Transcript
Speaker: AI can be the fastest thing on earth, faster than speed of light. and Eventually, in an ASI world, artificial super intelligence world, humans doesn't have work. Work becomes optional. I'm not as optimistic as you are. AI can now book your tickets, reply to customers and send emails for you.
Speaker: And sometimes there's no human checking what it does. Ederas and Natarajan built core stack around this idea. But for three years, investors struggled to see its potential. On this episode of Founder Thesis, Ereda Rasan sits down with Akshay to explain why AI needs guardrails and what happens when we let it make decisions on its own. It's not the software that is the limitation today, it's the hardware that is the limitation today. way the earth does not have even one GPU per human being. Do you have an IPO on the roadmap or do you think one of the hyperscalers will acquire you?
Speaker: EZ, welcome to the Founder Thesis podcast. You are the founder and CEO of CoreStack. Help me understand what exactly is the business of CoreStack. What does it do? It is not a name that a lot of my audience in India would have heard about.
Speaker: So I'd love to go deep into that first. Actually, first of all, thank you for inviting me to the podcast. you know Glad to be here. Happy to share our journey and what we are bringing to the table for our customers and partners.
Speaker: ah So interesting questions and of course, tag is in the B2B space. I don't expect a common technology consumers ah to know this. CoachLag is basically a B2B enterprise, cloud, AI, and SaaS governance platform.
Speaker: When companies consume new generation technologies, and especially in the newer age of AI consumption, they want to use the technology ah you know with the safe and secure feeling in their mind.
Speaker: They are getting the best return on investment for the investments they've made on the technologies. And they're continuously optimizing and increasing the velocity for which they're consuming the technology, whether is a healthcare company, fintech company, retail, media manufacturing. We have thousands of customers across every segment of businesses.
Speaker: And we enable them, empower them to use the technology. with a peace of mind from a financial angle, from a safe and security angle, from an operations angle.
Speaker: ah you know That's why most businesses will know us, but not the end consumers as such. Where does this product make sense? Because I feel like most startups may not need this product. It it must be a product for a certain size of business and a certain maturity of business. but what When does this product start making sense? ah Yes. now hours In turn, I will take it as I know who who's your ideal customer profile, right? When does it make sense? when When businesses or consumers are consuming at least few million dollars of
Speaker: ah technologies they need to use from hyperscalers like the Microsofts and the Amazons and Googles of the world or the larger ecosystem of technology players. you know Today it can be a you know AI computing capabilities that are available from NVIDIA to Google to Core Weeks of the world all the SaaS experience technologies from Oracle to SAP to Salesforce to Atlassian to, you know, if customers are consuming this breadth of technologies for a few million dollars,
Speaker: that's where they start seeing more complexities arise. How do I use the technology safe and secure? How do I continuously optimize it? So our entry point starts with, you know, from a company size standpoint, ah they must be in a position to ah spend couple of million dollars on technology, probably, you know,
Speaker: A few hundred employees in in employee size scale. in In terms of the Fortune 500 to 2000, would say probably Fortune 50,000. Those are the where they will find real meaning or real value outcome of from Coastan.
Speaker: So you kind of told me that it's a cloud governance model, which gives a feeling of safety to the enterprise buyer. But but um can you like make a little more concrete? What exactly is the product offering? Is it a cost optimization product that ensures that your cloud will is optimized? or but what What does coach that do?
Speaker: so Okay. So... We have multiple product or value categories. you know we Our products offer value to CFO and his functional units in typical companies, which means that it's a financial you know value outcome.
Speaker: It helps customers optimize on you know a total cost of ownership, investment, cost distribution, cost allocation, cost visibility, cost control, automated may of you know continually continuously managing the cost, and all of that. So it's a financial package.
Speaker: Then comes a security package, we call it as, ah you know, CostaGraphion. And that is primarily making sure the technology you are using is safe and secure for internal consumption and you are delivering safe and secure outcome for your customers.
Speaker: So technically, if you are in a regulated industry, I'm a healthcare customer or I'm a healthcare company consuming technology, I need to follow HIPAA, SOC, uh you know our regulatory uh compliances if i am in a financial industry i maybe have to follow know pci nist uh i am in government uh companies you know in the us uh you know i need to follow fedram fisma and all that kind of compliances those are all driven from a security foundation thought process but then costac does not limit to industry compliances
Speaker: ah Customers at ah at at a large scale, our Fortune 1000, Fortune 500 customers with their hundreds of thousands of employees and the technology consumption, they can define what is the security standard you know starting from you know authentication, authorization on a fully automated way. to make sure that no single person on a company can have access to the data, can have ability to you know misappropriate or inappropriately use the system of records, which may cause inadvertent damages eventually.
Speaker: You know, protecting from external hacker or external compromises that can happen to internal compromises that needs to be avoided and to meet the industry standards and compliances that are regulatory bodies recommending you to follow our security package delivers all of that.
Speaker: Okay. ah Did you start with both of these offerings or like what was the starting point? um Our starting point was, you know, the the one that we have not talked about is cloud ops. Basically, you know, 10 years ago when we started CoreStack, at that point of time, cloud was in the nascent stages. We were talking about, you know, 2016, 17 timeframes where the total cloud industry itself was...
Speaker: ah probably less than $10 billion. dollars Today, we are talking about a trillion dollar, multi-trillion dollar industry, but a decade ago, it was a total $10 billion dollar industry where adoption was limited because of automated abilities because of security compliances and because of cost overruns.
Speaker: And we started provisioning our... Companies would buy servers and do the hosting in-house instead of yes we were hosting it on and AWS because they felt it was more secure ah to host in-house with more control over things, e etc., ah How did you see that this is a problem, that companies are not adopting cloud ah because of certain blockers and there is an opportunity to build a business to remove those blockers?
Speaker: that that The landing of CoreStack, the idea of CoreStack was originally born to make the adoption better faster. And that stemmed from the foundation of all the cloud companies, ah cloud technology provider companies back then, Microsoft, AWS, IBM, VMware, and probably Google was not in the foray at the time.
Speaker: ah All these companies were cash negative. customers were trying to adopt. They saw you know more than 100% growth over growth you know every quarter.
Speaker: But the adoption was challenging because there are set of critical challenges that were there for CFO, set of critical challenges that existed for a CIO or a CTO, and the the Chief Information Security Officer. right So the three critical personas who came together, they all had set of challenges.
Speaker: because they had to coexist between the traditional investments. This is the times the data centers and servers and you know offline software, what we used to call as box products that were there. And then there is a cloud version of everything combined together and hardware and software bundled together being offered by the hyperscalers.
Speaker: your workflows, your automation, your methods of managing, methods of optimizing, methods of securing, everything was different in these two different worlds. And customers found it difficult. We wanted to ease the adoption of cloud. Only when more adoption happens, you move in the maturity curve to start solving the next problem. So it was the cloud adoption problem first we solved.
Speaker: Eventually that led to now more and more people are adopting, a you know what, I have no problem in moving to cloud. Now i have a problem of ah security from a data standpoint, from a usage standpoint, from a regulatory standpoint. And once they solve the once we solve the security problem, then comes, oh, now I'm starting to spend more.
Speaker: I want to have control on my finances. I i want to get you know my usage being totally optimized. So it's an evolution. And we drew the natural progressive path of enabling our customers to use the cloud.
Speaker: then make sure they feel more secure. And then, okay, now you are ready to, you know, use it on a scale mode. Then you need continuous financial, you know, optimizations that needs to happen.
Speaker: I'm wondering what was your vantage point? What were you doing that you saw that this is a problem about ah cloud adoption having blockers? Well, I as a founder, I used to be part of the larger Azure building Microsoft ecosystem, being one of their partners.
Speaker: Built Azure from its infancy days when Azure was not even named. but So this you're talking about system integrators. So system integrators are companies who help ah ah enterprises to buy and implement IT solutions, software. Not really system integrators. it if If you really step back and see what was happening in the industry, after AWS started making inroads, when there are few billions in business at that point of time,
Speaker: cash negative, but few billion dollars of annual business. All the technology providers were box providers at that point of time, you know including Oracle, VMware. What does that mean, box providers? Microsoft.
Speaker: Box software is something that you download in a CD or you you download from the web or you get it a CD. and that you install it in your on-premise servers to run the software. right and It can be a Windows Server, it can be a HP server, it can be you know from a server operating system to an Oracle database. to you know You have always been given software in the form of downloadable or consumable installable mode.
Speaker: that was transitioning into consumable mode, right? You don't need to download, you don't need to install, you don't need to own anything as a software license or a hardware, you know, in in your data center.
Speaker: Everything is bundled and given to you. And all the traditional software providers who were building the so-called box products, right, an installable, downloadable product, they were starting to build consumable products.
Speaker: And when the customers had during the transition period, a downloadable product and a consumable product, it it has a different set of grammars. It has different set of methods of surviving. You know, this is like speaking Greek and Latin.
Speaker: How do you bring a common mechanism by which you continue to live in your old, you know, tenets of living and you are able to adopt with the newer tenets knowing that at some point of time you will let go your traditional models and move completely to the new model.
Speaker: This was our vantage point to figure out. And you know thanks to the hyperscalers, we call them the today's large technology providers, the big four in America, the Microsoft, Amazon, Google, and Oracle.
Speaker: They all decided and announced to the world that we are no longer going to build installable products. all our products will become consumable product.
Speaker: So customers were forced to change, but they're looking for help. hey Who can help do this? And yes, there were systems integrators, there were traditional service providers for the technology world who are trying to offer the transition ah in a manual way, right ah throwing human labor as a differentiation to transition.
Speaker: But what was difficult was the software capabilities. You were working for one of these software ah development or like system system integrators or? oh At that point of time, you know, when but just before starting CoreStack, I was in the larger ecosystem of Microsoft and Azure, that the the hyperscale building itself through a systemstimeator through systems through a systems integrator who was actually helping Microsoft to build,
Speaker: the Azure cloud and make it available for customers. Okay. So as part of that, I have been being in the industry for the last 30 years. we have I've seen the IBM era, mainframe era.
Speaker: I've seen the AS400 middle frame era and then to the complete client server and client server moving into a cloud era. So we had the previous x experiences that clearly told us There are methods in which we can provide the transition ability to the customers through software and need not be done through manual labor.
Speaker: And that was the birth of Kostak. And your experience was largely on sales side or were you like product management, product building? or Oh, i I've played multiple roles. I've been a technologist. I've been an architect. I've been a software developer way back in my career. i eventually transitioned into, you know sales, running business, P&Ls. So it's it's a typical career progression that you could have imagined, you know, in in late 90s and early 2000s, from software development to sales to business.
Speaker: Okay, okay, okay. Got it, got it, okay. So you saw, ad so you were essentially a P&L head and you saw that there is a software solution to helping companies migrate to the cloud from on-premise servers to cloud.
Speaker: ah ah Instead of selling services to solve this problem, you can sell a software to solve this problem. um What did that software do exactly? Well, it was basically CoreStack. At that point of time, the the first, second version of CoreStack, the alpha, beta version of CoreStack, was making the cloud consumption easier for customers. you know Assuming say assuming in in your data center world, in your installable software world, you were used to certain...
Speaker: standard operating procedures of how do you configure an application, how do you build software and pop of it, and how do you make it available for your consumer consumers. consumers right So you had our clue.
Speaker: But when you take that and put it on cloud, it just does not work. It will not work. right Even though it was Microsoft operating system here and Microsoft Azure here,
Speaker: it won't work. You might have been running on Red Hat Linux and you might use the Amazon RHEL, Red Hat Linux on your cloud version, it it won't just work.
Speaker: So there were transition abilities that were acquired. There were workflow that needs to be you know updated to work on a cloud form factor. And those were our cloud operations. right So we used to call it as you know a provisioning, orchestrating, building the basic guardrails of the technology that can be used in a cloud form versus an installable form.
Speaker: right So that was our the first version of the software. How specifically did it work? like and This was sold as a SaaS product? Yes, it it was a SaaS product.
Speaker: So ah technically, what when consumers do consume, and we had ah some of the very large telco companies and VLSI companies, healthcare companies were using our platform.
Speaker: So when when customers use cloud, ah typically the cloud hyperscale cloud technology providers provided with certain level of authentication authorization. They call it as cloud subscriptions, you know your cloud accounts.
Speaker: And you know once you get into your subscription, you start buying or you start creating your instances of server your instance of database your instances of you know additional set of data sources uh your web applications you know those are called ah your workloads right they create the workloads so the workloads sit inside an account or inside a subscription and the entry point is an account or a subscription so the same account of subscription if customers
Speaker: give the credentials to costak costak this is my azure subscription this is my aws account right and that's that's pretty much we need you enter that account then costak has the ability to go discover everything that you have so far done inside of the particular cloud player right and we were a multi-cloud platform So we could discover if you were using you know IBM, if you are using VMware, if you are using Microsoft, if are using Amazon, we could discover everything you have done in the cloud and provide certain basic recommendations based on
Speaker: how well you are using the cloud from a maturity standpoint, because all this cloud providers, uh, you know, eventually started offering a framework of maturity. They call it as cloud, um you know, cloud adoption framework or, you know, well architected framework. So basically it's a recommendation by the cloud technology providers. Hey, when you consume my technology, you know, we hear all the recommendations.
Speaker: So based on that recommendations, CoreStack will go first to discover and then find what is your current maturity level. Are you at a 60%? Are you at a 20% maturity? Are you at 80% maturity? And again, those maturities measured against, okay, financially, how matured are you?
Speaker: Security compliance wise, how matured are you? Operations, uptime, availability wise, how matured are you? From an experience building standpoint, how matured are you? So, you know, it's a different vectors of giving the maturity and each customer will have varying maturity. Some will be way high, you know, there will be total cost of ownership. They may be significantly optimized. They're getting every dollar work back, but then They may, you know, that their security score may be, you know what, tomorrow when the next hack happens, when the next compromise happens, you will your application will be the most vulnerable application.
Speaker: Financially, you are protected, but security-wise, you are exposed. Or in some cases, it may be the other ways. Oh, you know what? Just to get this additional 0.01 percentage of security compliance, you are spending half a million dollars more. You don't need to spend that more. You can reduce your spending. You still get that kind of and you know security compliance. So that's the kind of maturity. So it all happens when customers enter just their cloud subscription and account details and Kostak goes, discovers, finds, then starts recommending one after the other. Again, I'm talking about the basic elemental capabilities. The advanced adoption, you know, if you're a cloud architect, if you're a technologist, then comes the next level of, you know, what do I want to build on top of my workload? How do I want to transform my workloads? And back then, we were talking about you know infrastructure, basically servers, database, and you know network set of capabilities. But today, in cloud, if you look at it,
Speaker: anything that you can think on earth from ai to cognitive services us to you know even the conversation that we're having everything is dynamically built and you know life transformed on the cloud so cloud is the form of technology consumption and not just as an infrastructure not just as a database not just as a network source but for every experience from analog to digital and all the services now have to be financially optimized taken care from a security compliance standpoint.
Speaker: And ai now needs to be managed in a way that the the resources doesn't have to be managed. Because now we are talking about AI, which is an intelligent piece of software. Now that piece of software can do things without your knowledge.
Speaker: you know it It can self-learn. It can you know have ah dimensions that are unacceptable. So all that needs to be managed. So CoreStrike evolved from a fundamental approach basic technology enabler for customers to consume to dynamically manage even the AI today and govern the AI. Today, AI governance is a big topic. and We have become a natural AI governance technology provider over a period of Okay. Okay. I'll come to the AI bit. i just want to make sure i properly understand the legacy products first. um So I guess a simple analogy to understand is like, say, ah ah company uses electricity and if it could install a software which will constantly be monitoring the meter, seeing that are you optimizing your load and what are your peak hours and non-peak hours and The equipment that's plugged in, like, for example, if you're using an air conditioner, then is it a power efficient air conditioner or is it an inefficient air conditioner and it's costing you a lot of money?
Speaker: ah So things like that is what CoreStack would do when it comes to cloud infrastructure consumption. Yes, your your an analogy is well suited and that is the foundational. You can take it to an additional easily understandable example of like, hey, putting additional sensors like light sensors, motion sensors, like, you know, hey, you put all those sensors, it calculates, I mean, it starts optimizing on multiple devices, not just a large energy consuming source.
Speaker: Even your light bulb may consume a smaller amount of energy, But over a longer period of time, if it is unattended, even that adds to the cost. So it is having some intelligence piece on a large consumption capabilities and kind of ah motion sensors and you know sound sensors and you know things of that kind is what Kostak does typically for the cloud cloud world.
Speaker: Got it. Okay. Okay. Understood. So so ah your source of data for... uh calculating maturity level etc is just how the cloud is configured like uh you can log on to azure's dashboard and then you can see how various services that you're consuming of azure are configured ah and that remains the core source of data today also for like say finance finance ops ah or security Yes, that that's that's a largely primary source of data, right? I know, what are you consuming on the cloud providers landscape?
Speaker: Comes from the account and the account specific, it can be a financial data, it can be a performance data, it can be security data. That's a foundational data.
Speaker: But then what do you really do with it? Right. ah Every minute, every next minute, what is going to happen? You know, that's a dynamic set of data that needs to be gathered.
Speaker: And we have a lot of sensors to gather the data. And even the cloud providers provide the bells and whistles and, you know, tools to gather the data. We continuously gather that information.
Speaker: and eventually to provide the outcome, you can all can have all the bells and whistles, but then the the outcome is primarily how well you connect these data points between finance, security, operations, availability, all that, and take dynamic actions.
Speaker: right i mean that are Some of our customers, our largest customer uses probably $1 billion dollars of cloud infrastructure spend. They pay $1 billion dollars to Microsoft or Amazon or Google or ah probably distribute distributed across. And it is no human cost, no additional cost of anything.
Speaker: It is just technology consumption cost. At that level, imagine you know this is like a million equipment inside an airplane flying and which part of that is is functioning, not functioning, which is supposed to function the way it is supposed to function.
Speaker: gathering you know live statistics and making dynamic decisions to fly the aircraft. you know That is the kind of challenge that goes on inside the you know cloud when it comes to a large scale. As I said, if you're spending you know a couple of million dollars and less, this complexity is very limited. i know ah Probably one engineer can sit and manage all of it without any complexity. It's like driving a you know a simple car with no fancy equipment, right? i Anybody can drive. Whereas flying an Airbus or Boeing commercial flight,
Speaker: That's the complexity. Got it. Okay. Understood. um So you are not only able to read that data from the dashboard, but but you're also able to write there. Essentially, you can make changes on behalf of the customer. Yes. you know that That's the uniqueness of Kostak. Kostak is not a visibility engine. Kostak is an outcome delivery engine.
Speaker: It's a SaaS platform. First, it gives you visibility. Then customers have the ability to let how much Kostak can do on their behalf. behalf and It's like planning the model today. you know But we have been having this for the last 10 years.
Speaker: ah They can decide the production workloads, Kostak don't take automated actions. right It needs a human intervention. But whereas it may be a development environment, test environment, and all the workloads when they're in the creation process, they would let Kostak take ah automated actions.
Speaker: So yes, Kostak has the ability to recommend automation. take dynamic actions and deliver the outcome on behalf of customers.
Speaker: And that was one of our unique differentiators because taking actions across multitude of platforms which are totally indifferent and the way Microsoft functions versus Google functionssoral functions versus Oracle functions, AWS functions is different. But you give a seamless experience when you take actions.
Speaker: So that was one of our unique selling points. Okay. Okay. Very interesting. um I'm going to bring in another example. ah So, you know, a lot of people are now, for example, letting Claude run their meta ads platform, you know, so so running running the meta ads, Claude is able to run it much better than humans. And meta has started banning those ad accounts, which are run by agents. um So I'm just wondering that do the cloud providers, the hyperscalers, the
Speaker: play nice with you in terms of letting you read and write on behalf of your clients? ah Is it with explicit? it Oh, absolutely. you know that we We enjoy the ultimate level of relationship, partner relationship with the cloud players or cloud providers.
Speaker: Because what we do fundamentally is give the ability to a customer to consume more technology Because they think that whatever they are consuming, they are getting the best returns.
Speaker: And it is safe and secure. And their velocity is better. right What they have done yesterday versus is what they are doing today is better. Their businesses are growing. Their customers are happy.
Speaker: So technically, we are enabling the end customer to consume more technology. For the cloud technology providers, the fundamental DNA of driver is they want their customers to consume more.
Speaker: And that is what Corset enables. right So we make customers consume more cloud. Yes, we optimize it. Yes, we save. But if you don't save, customers are going to find out at some point of time, or I'm wasting my money. i don't want to, you know, the CFO will come and say, okay, I'm putting a budget cut now.
Speaker: Or I'm having a budget freeze. You cannot spend more. Which is going to directly hurt the cloud technology providers, you know, if this decision happens. Kostak helps customers or companies to not have the freeze, to not have the budget cuts to spend on technology.
Speaker: And, you know, we offer one of the other unique selling points of Kostak is if a cloud provider offers the new capability today, Kostak has the ability to you know, secure it, operate it, optimize it on the same day.
Speaker: Typically, you know, when a new service is offered, you may have to wait for another 30, 60, 90 days to to get everything operated. So, you know, because Kostak can allow the customer, but's it's a new shiny thing. It's like your new Apple phone that is out in the market. Everybody goes to the doorbuster and get the phone, right? So inside the technology world, a similar thing happens, right? Today, cloud clock cloud launches co-work. Everybody wants to go get their hands out. But hey, is co-work secured? Is co-work managed? Is co-work optimal?
Speaker: That course I can do on day one, imagine now the power. You are allowing customers to consume a new technology with peace of mind. Because we were enabling that, the hyperscale cloud technology providers were I mean, I would say we were like deep in bed. Okay, you know, you are one of our best partners. You are a build partner, you are a part you are a first party partner, third party partner. There are multiple titles. With all of them, we enjoy all the titles.
Speaker: Got it Okay, so they see you as a system integrator almost, like a productized system integrator instead of the usual system integrators who throw people at the problem of cloud reduction.
Speaker: Yeah, rudimentary you can call that. So they see us as an enabler for their businesses to grow. So we we critically influence their top line. Got it. And do they ah bring you in pre-release of these features? Is that how you're able to optimize features from day one because you've been there pre-release?
Speaker: Yes. that i have the The simple answer is yes. for Not for all capabilities. You know, it's that Every cloud provider has hundreds of products and services that are being offered.
Speaker: ah you know There are certain set of products, yes, the partners have will be given, have been given early access. So you are able to build the guardrails, the additional mechanics that are required for customers to consume.
Speaker: Some of them are like vanilla. It doesn't matter whether the customer gets you first or the ah they enabler gets first. right So, you know we get it at the same time. um you know So, in in in short, the answer is this. If a partner needs to have access to the technology early enough so that they can build the additional ecosystem that is needed for the consumer to consume, we typically have access, yes.
Speaker: and Understood. um I'm going to ask something very, very naive about I understand cloud as ah virtual server that you have a hard disk and a memory and a processor ah which is available to you virtually instead of running it in your basement. ah You are connected to AWS and Amazon is running it for you.
Speaker: ah So now you said that they have hundreds of products. um Just help me understand, how does how does cloud work? and I know it's a super naive question. I'm not from a technical background, but I'm just curious about that. like Is my understanding of cloud accurate and from where do these hundreds of products come? what is ah What is happening there?
Speaker: ah Broadly, what you're saying is right. But that's like you know a the the language as or english as a language as 26 alphabets. yeah okay Why does the why does the you know grammar book or a dictionary for English is having 10,000 pages? Right. At the end of the day, the only time is Eclatius needs to be reused. right It's it's ah that kind of an analogy. so that Everything that is going to run is primarily going to run on hardware and that requires a basic operating system and that is what you call it as a server.
Speaker: And you know there is database and different kinds of data that needs to sit in from an ah you know structured data, unstructured data, you know large set of data, you know whether it's a video file, audio file, everything is a data.
Speaker: So the data has to be designed in some some places, in structured and unstructured formats. And then to access all of it... These are all products. Like, you know, when I say database, there are 20 different data and database related products that are available.
Speaker: Right? There are structured database products, unstructured database a way, these hyperscalers are cannibalizing the... At one point, I'm like, say, Oracle was a database company. So... The same thing is available, you know, Oracle, MySQL, SQL Server, all of these are databases. Those are available in the cloud form. Okay. In addition to that, cloud has its own additional data structures, data forwards that eventually have come in, right? So that's why in the world of data, you will find probably 20, 30 different kind of products.
Speaker: Again, each of it are fundamentally they deal with the data. but what kind of data they deal with. Are they high performance data, high throughput data, you know like a live, if you want to telecast a Super Bowl, you know how many people are watching and you need to have a system that can manage you know hundreds of millions of live streaming capability. right that requires set of systems which is different from a ott system where hey it's the same content new movie that is coming in but you know how much people are going to watch at peak level at a normal level you know there's a different system compared to you take an you know a reservation system of your flight booking versus your next train booking or your next movie theater booking it's all data but
Speaker: the experiences, the outcomes are differently handled. Accordingly, there are categories of you know data management technologies available on site cloud. Now in the new generation, lot of cognitive services that are available.
Speaker: like As you and me are talking, probably whatever we are talking is live transcribed by the software that you're using. So, you know, the the system does not know what I'm going to say next, but it has the ability to understand my dialect, my language, translate that to a text format which you can consume. So, these are the more new generation cognitive services, whether it can be AI service, whether it can be a deep learning service, visual service, audio service, touch and feel service. know, if you go into a healthcare company, you're using a smartwatch.
Speaker: It recognizes your motion, right? It recognizes your heartbeat. It recognizes your pulse. All these are getting translated and there are cloud services, again, that are available. So now think of each experience that I talked about is available as a software product by cloud providers.
Speaker: Okay. What are broad categories? One category i understand is data storage and access and like databases. What else are the product categories? so but At a broad level, you can probably say five or six categories. right know and You are talking about compute, storage,
Speaker: data, network, cognitive services, and sensory services. you know So you can can broadly classify all of them. In the cognitive services, I include you know your visual, audio, deep learning, you know all all that kind of services.
Speaker: ah AI service is on top of everything. Now, artificial intelligence as a service, if you're consuming your TensorFlow models, you're consuming. Those are AI services, right? So broadly, we could look at it in compute, storage, network, database, cognitive services, AI services, sensory services.
Speaker: Okay. ah What is the difference between cognitive and AI services? ah Like, wouldn't the AI, like, if you are, like, say, you have an API key of cloud, doesn't that do everything? Yeah, so that's a very interesting question. um You know,
Speaker: Prior to 2018, if you asked that question, the answer would have been different. Because the AI, what humanity has been thinking ah since the birth of ai as a concept, AI as a knowledge, and the neural extension of artificially building intelligence outside the human brain, you know, this is all part of neural networks, right? So during our college days in engineering, computer science, we used to study neural networks. And then, you know, how artificial intelligence was built.
Speaker: ah We were thinking and building this based on traditionally then existing computing models and data management models.
Speaker: In 2018, what has happened was, you know, a set of people in Google sat and wrote what's called Transformers today. And it's a new concept of achieving...
Speaker: paper Yeah, correct. You know, it is a transformer paper written by sort of Googlers, which eventually gave birth to the AI that we're experiencing today. So today's AI is not based on the foundational neural network models. Rather, it is based on the speed of computing that can happen.
Speaker: Today, every AI, the the reason why we use NVIDIA, our GPU as the most used ai computing fundamental hardware is that it can process today,
Speaker: the the latest NVIDIA chips can process 200 trillion transactions per second. If I have to loosely translate that into a data size, right, and if you're not able to visualize, you can probably transact 13 terabytes of data, you know, by by that one second, you would have probably transacted, processed 13 terabytes of data.
Speaker: right which is huge. you know as As humanity, we wouldn't have created 13 terabytes of data until 2000, forever. And that data can be processed in one one second. Is this speed of light computing and a model that is built around that, how fast you can process any and every data given to you and then translate it into an intelligent model.
Speaker: So the traditional computing, the neural network model was building. How do you build as if you are a child, you are learning, you are, you know, you are gross motor, you are fine motor, then your are association skills, your commutative skills. That's how this intelligence was built.
Speaker: Right. Now you have the ability to grab every information in a less than a second, and now you need to only sort to which data is where. Hey, this is ah you know cognitive data, this is analog data, this is ah intelligence data, right and then pick whatever information you need to answer the query that is in front of you. So the model-based AI is different from a neural network-based AI.
Speaker: So that's why you are, when I say cognitive services, is it data coming in? Whereas today's ai that we're talking about is all model based. You need to build a model. You need to train a model. If you're talking about a healthcare model, that needs to understand every nuance about healthcare, your you know human systems and you know data associated to that and find a needle in the haystack.
Speaker: And this services are different from the cognitive services. Cognitive services are, I'm able to see, gather everything. Earlier, while you're gathering, you would have figured out, you know, what could have been happening after you gathered, you have to go process one by one. But now the model while i'm seeing i can figure out oh this is a tree this is a structure this is a cloud this is a mountain ten this is a human human is walking this is a you know a zebra crossing on the road and that is how the current model works so the ai services and cognitive services are different
Speaker: today what is the relationship between data center and cloud so data center companies are different from cloud companies i believe and in the current world because data center is being a word that is used extensively and probably data center companies are on steroids right now because uh you know the world needs more data centers uh it's a very interesting irony right so
Speaker: 70s, if you went back, you know it was mainframes, large, big computers. Then you have so-called dumb terminals, you know a screen that remotely connects through wires and you're able to you know use it.
Speaker: Processing happens in the big computer, right mainframe days. And then from there, we slowly move to a client-server technology where your computing, your experience, everything is locally residing. You needed a large computing. Then you had something called as a server sitting inside a data center, a large storage, a large server, a large database that was available to you to exclusively connect, to process whatever you wanted to process.
Speaker: And when cloud transformed it, what happened is we went back to a mainframe kind of experience where The faster processing computing resources were made available again inside a data center, but you did not own it.
Speaker: You did not control it. Rather, you controlled the but processing that happens and you paid it as a utility. It's like I built my own power center, power generator at my backyard to, oh, I no longer need to build my power generator.
Speaker: I'm using a utilities-provided power service. The utility provider seems to be very reliable, giving me continuous services. And during bad weather conditions, I still have my backup generator or a backup you know ah power generating system.
Speaker: That is how customers move to. So basically, from a large... utility, we moved to a local self-generated model, which is a data center. From there, we came to cloud, which is a utility provisioning model. From there, now we are going back to data center world because during this period between mainframe and to plan server to the, you know, cloud to AI, the computing processing powers have grown significantly higher to an extent.
Speaker: the The computing power that you have on the phone today, in your you just smartphone, is probably thousand times more capable of a mainframe that powered the entire city.
Speaker: and i So now you translate it to a centrally available capacity. If you have to process 13 terabytes or 200 trillion transactions per second on a computing device, it requires that amount of energy and I mean, that that information has to be stored somewhere. 13 terabytes processing can be easier, but it needs to be stored somewhere.
Speaker: This requires again a large scale setup, a physical infrastructure where ah you need to have a significant amount of power and a place which cannot be made available in the cloud form factor alone.
Speaker: It is available as the cloud form factor. Today you can consume AI through the cloud providers as a cloud service itself. It is coming at certain cost, like a utility provider selling a service at a cost.
Speaker: When you think that I can build this power unit at my backyard, my unit cost is lower than the cloud provider providing or the utility provider providing, then you start to think, oh, what is better now?
Speaker: Should I continue to use the utility service or should I go build my own service? There are trade-offs from a cost to vantage point, from a security compliance vantage point, from a, you know, improper usage vantage point because AI, you don't control it, you know, AI builds things on its own a lot of times.
Speaker: So, the companies are weighing the risks between cost, security compliance and the unknowns that can happen in the system. It can be consumed through a cloud or you can go build your own data center.
Speaker: So there are thought process that says that if I have the AI cloud as a data center service, my cost won't shoot up because I will not pay for tokens. I you know i can definitely control my cost because I'm owning it.
Speaker: But what I will not control is you know unexpected usage of the cloud and i'm not unexpected usage of the AI and what disastrous outcomes it can create. I need to control it. right So there are companies that are taking responsibility on it going the data center way because they think that unit cost of AI computing, otherwise they have to consume as a utility from a cloud provider is more expensive. This is less expensive.
Speaker: for you know the the typical Fortune 5000 and above companies. I don't have unlimited budget of spending hundreds of millions of dollars as a capital investment to go buy my own AI data center or build my own data center. rather i want to use ai okay i will continue to pay utility service because that's the only mode that i can consume this newer capability so the two worlds are emerging the you know if if you have to call it as a rich world and a poor world The rich world is moving towards, you know, I want to build my own AI and I want to control control the destiny of it. Whereas the poor world is continuing to use it as a consumable service. Again, there are a multitude of factors that come in. ah Today, the rate of invention of AI's capability, you know, this is why NVIDIA is the world's most admired company today. Every year, every year, they are quadrupling the capabilities.
Speaker: of the chips or the you know processors they're inventing, quadrupling. um you know The traditional CPU world, if you are looking at it it, always one one year to the next year, you know, Intel has been our, you know, no home in earth you can find without an Intel processor inside today, right? But Intel has not been innovating on an annual basis like quadrupling kind of a capability.
Speaker: like Like Intel used to follow Moore's law, right? Moore's law was that every 18 months. Yeah, correct. You know, and and they have doubled it, right? You know, whereas today it is doubling every quarter.
Speaker: right So that has a big impact. right i know Every 18 months you double something versus every three months you are doubling something. In 18 months, imagine the amount of you know exponential growth you would have seen.
Speaker: So this plays a role in the consumable as as a utility AI versus the data center owned AI. The data center owned AI will have latest and greatest and most powerful because you want it.
Speaker: You can replace your one year old ah device and place a new device. Whereas if you are buying it from a utility provider, the utility provider, you know, if they keep changing the backend capability, then they're going to price you more and you don't want to pay more, you want to pay less, then you are using a little older technology.
Speaker: hardware at the back end, but it is still meeting your demand. So the the reason why data center world is in steroids today is, and it the most in talk ah conversations today is how well latest and greatest can be controlled and used by, you know, set of companies that are, that can afford that.
Speaker: If I understand correctly, A data center company is different. Like AWS is not a data center company, but a data center company would be selling to AWS. Like like AWS would be a customer for a data center company or AWS would have its own captive in-house data centers.
Speaker: AWS will have its own captive in-house data centers. Today, Like how the these mobile towers, right? So at some point of time, know, probably during 3G, 4G periods, if you have multiple, you know, cell phone service providers, they all used to own their own towers, right? You know, an eighty will eighty and will have their own or in India probably Airtel will have its own, Verizon will have its own.
Speaker: But eventually, more and more, the technology consumption subscriber base was increasing. ah you know what happened was in one tower, you can have an RTL versus Verizon versus in the US probably T-Mobile and AT&T can be powered ah from one tower.
Speaker: So similarly, today, data center provider can create a data center capability of ah space, cooling, network connectivity, hardware, all of that, can give it to ah Microsoft, a Google, Amazon,
Speaker: And instead of Microsoft, Google, Amazon going and figuring out all this logistic first level of you know land, physical building, then air conditioning, network gears, that is getting outsourced as a commodity service to someone else, which is a large set of data center service.
Speaker: And now the tenant can come in and put their equipments and connected to their service providers. So today, the the large cloud providers, they have their own data centers.
Speaker: The data center providers offer services which actually the cloud providers can use. It need not be only cloud providers as an end customer. Oh, i'm I'm Toyota, right? As an end customer, I'm Bank of America, right? End customer.
Speaker: I have the ability to go buy it from a cloud service provider. Or if I think that I'm better off ah from a multitude of vantage point to host my own hardware inside a co-located data center, I go to a data center provider and buy their foundation service and then bring my equipment and connect it to my network. know if If I do this, this is data center.
Speaker: If I'm not owning the equipment, if I'm not owning the software persistently, rather than I'm consuming it as a service, then it is a cloud service. So in in large cases, a customer can consume a data center service.
Speaker: A cloud provider also can consume a data center service. But a cloud provider converts a data center into cloud and offers it as a utility, whereas a customer consumes the capacity they are building inside as if it is their own service.
Speaker: Got it. Okay. And ah you're saying that the reason why today data center as an industry is growing is because companies want their own models instead of being a cloud. They want to use an open source model or a small language model that is exclusive to them.
Speaker: The reason for this ah one is, of course, cost. But is it also data privacy? um Yes, you can say the primary driver of data center or data centers on steroids today is the cost factor.
Speaker: right Today, ai consumption is primarily based on tokens.
Speaker: you know, you may get an outcome for a low number of tokens or you may not get the outcome until you complete processing tremendous amount of data and get the needle in the head stack, which may consume, you know, significantly higher cost.
Speaker: And the technology providers, the businesses still have, while they are reducing the number of employees that are required, software is being built on machines or built by machines, you still have, you know,
Speaker: So you are a having your bot employees or human employees, which are all working. And if you are giving all these tokens to so many employees, the cost can significantly go up.
Speaker: So you are now starting to measure what is the total cost of production, whatever they're producing. I'm a hospital scattering to my patients or I'm an airline operating my flyer fleets or I'm a bank servicing my customers. It can be anything. So end of the day, your measurements is now starting to translate into, you know, what is the number of tokens I need to service my customer, to service my patient.
Speaker: And that cost, if fundamentally you think that I have all the technology available in a data sample form, but my final translated unit cost of customer service is lower than consuming it as a token, then this set of adoption is increasing.
Speaker: right For the mid-market businesses, for the you know lower SMB segment kind kind of customers, the tokens are still cheaper. right So even if tokens are expensive, they don't have the ability to go and buy super expensive capital incentive equipment. So they will continue to be in this world.
Speaker: That's a faster driver. It's like taking a Uber versus buying a car. Correct. Buying a car. Exactly. You got the right example. Right? For people who can buy the car now, hey, do do you want to buy a Toyota or do you want to buy Rolex or you know whatever is the most expensive car out there? right um That is primarily because of the additional dimensions you are thinking, hey, what is my sophistication required on security? What is the sophistication required on data privacy? What is the sophistication required on even sovereignty for that matter?
Speaker: Right after the wars that are happening now, hey, you are thinking about ah My data should not leave countries premises. It needs to be hosted inside the sovereign borders of the country.
Speaker: So additional drivers are driving. So many large companies today the the nations take pride of they are recommending the the national governance is recommending saying that you need to run this within the soil do not have it outside the soil that also mandates them to start thinking about okay now i i need to have this computing available in my premises so the drivers once customers decide that i need to buy a car whether you buy a Ferrari or you buy a Toyota, depends on the additional drivers that I was talking about. it It is not just cost now, it is privacy, it is security, it is compliance, it is sovereignty that drives you to decide.
Speaker: So ah now, if let's say I'm going the data center route, so then typically ah open source models is the way that most people do or do they build their own models? Like would they take a DeepSeek or a Lama to create ah their model like and based off DeepSeek or Lama or would they build from scratch?
Speaker: No, they would say today today the foundational models are all available. mean Starting from OpenAI, OpenAI is an open source model. right i You don't have to really take a deep seek, but you can even take OpenAI foundational model. you it's It's open source. It it was open source. Only the latest models of OpenAI is not open source. The older models, are it's it's open.
Speaker: right So you can take that and you can start building your specific model. And and today there are a lot of additional models that are being built from large language models. So that's one way of processing huge amount of data.
Speaker: But you know starting from small language models to tiny language, in out in order tiny models are on the because you are looking at small devices, like even the in the and the smart ring, can have some level of intelligence that is processing. Because, again, imagine the computing capability that sits on this ring is way higher than a mainframe that operated the moon landing that happened in the 1960s and 70s. So, I mean, if if a TLM can work on this, this this can probably moon you know take a rocket to moon.
Speaker: if it can take a rocket to moon, it better do something for you on your phone ah to to help you keep your health up, to help you eat better, to help you wear better clothes, to go to office on time, to take the, you know, whatever you want to do.
Speaker: So. Okay, okay. Fascinating. Now, when you are consuming AI, ah what is the role of cloud companies there? Because wouldn't you directly go to a cloud or an open AI to get the tokens? Oh, the biggest backend that is needed is not the software, it is the hardware.
Speaker: Today, the world is limited by two...
Speaker: factors on AI. One is the amount of hardware that can be produced to do all this computing. right Every time you ask the query, every time I ask a query, there is some GPU somewhere on Earth is processing trillions of data for you and me. It can be ah as simple as dumb question as a how many alphabets English has.
Speaker: right To build my entire business plan to land a rocket and moon, You know, all these are happening today. There is some GPU somewhere on Earth is running.
Speaker: And the the the models by which the GPU functions, I'm not talking about the AI models. It it is, ah you know, the method in which the GPU processes data and the amount of power it consumes is significantly higher.
Speaker: So even though we have more GPUs, the second limitation the world has is power. The earth does not have enough power to, you have enough power supply to power all the GPUs that are created.
Speaker: Right? You know, if we create enough power, then the bottleneck will move to number of GPUs that needs to be created. And all of this needs to be hosted a data center, whether you call it as a cloud or whether you call it as a data center. If a cloud provider provides it, you're calling it as a cloud because you don't own the software, hardware, anything. You consume it as a service and just pay for tokens. You call it as a data center when you own the hardware. You have the ability to pay for the higher capital,
Speaker: Totally, you own everything as if it is your equipment. And that again still requires power, still requires processing capability. So the challenge that you asked is foundationally to the amount of power Earth can generate today and the amount of computing that is needed at the hardware level.
Speaker: Once these two are, you know, crossed, then you will see the real potential of the software usage of tokens.
Speaker: Today, the earth does not have even one GPU per human being on earth. We don't have 8 billion GPUs on earth today. But you know, how many GPUs every human being needs today for their processing?
Speaker: I don't know. If you have to just take a guess, how many GPUs you would need ah you need for all the processing that you are using for ji from a GPU standpoint? at Half a GPU or something like that?
Speaker: you you You at least need 10 GPUs. wow people People like me probably need 100. People like Sam Altman you know probably need 1,000 GPUs. But the world does not have GPUs, that many GPUs. We all get probably...
Speaker: quarter GPU or half a GPU on an average basis. ah So we are not utilizing the true power of AI because the resource is not available at scale.
Speaker: The hardware there is not available at scale. The hardware is not available at scale because the power supply is not available at scale. So if we resolve this too, then we will see the unleashing power of AI from a software token standpoint, which will become commoditized again.
Speaker: Then every damn thing, and this is where we are talking about, and and you know GAI, generative AI, becoming an AGI, which means artificial general intelligence, and eventually becoming artificial super intelligence, meaning the AI is...
Speaker: ah going to do things way better than and human being could ever do. And these all require units of consumption that are in the magnitude of thousands per human being on earth.
Speaker: So, you know, typically it's not the software that is a limitation today. It is the hardware that is a limitation today. So, I mean, what what I'm wondering is this, ah you If you want to, let's say, decide that you don't want to build your own model, but you want to buy tokens, ah wouldn't you directly be buying tokens from a cloud or like an Anthropic or an OpenAI and not through AWS or not through Azure? Yeah.
Speaker: I mean why did why did as your why did Microsoft and invest in OpenAI? what What is the significance of that partnership? You know, how are cloud hyperscalers and ah AI companies in bed together? That's what I'm trying to understand. like Okay, so but and know fundamentally,
Speaker: see the token-based model is becoming a fast-standard new norm of paying for any utility that we're consuming, you know, you're consuming gas, ah gasoline, you consume, you know, in some parts of the world, per gallon, per liter, you know, things of that kind. If you're consuming electricity, you are consuming, you know, per units of power that you're consuming, right?
Speaker: So similarly, Tokens is becoming a new norm of consuming AI, AI-based service. And it need not be just Cloud, it need not be OpenAI, it need not be Gemini. It can be any software, right? You know, you are consuming SAP software, and if the backend they're using ai you know Today, they may give a number of seats, a number of licenses, number of employees. right or There are companies that are pro they're giving the the units based on throughput, you know how much of data that you are transferring, how much of time you are using it for. So measurement has been there in different form factors because everything is now back-end done by AI, and AI is...
Speaker: fundamentally measured because of scarcity of resource, hardware and power, it is measured on tokens. Every player who is sitting on top of that is starting to translate their service offering into a token based model.
Speaker: That's why you hear the word token repeatedly. Oh, you're paying plot tokens. Oh, if you, you know, eventually in the next few months, you buy SAP software, you'll buy but but by it on tokens. And eventually even Codestack will move into token based licensing model.
Speaker: yeah you are paying for tokens because back end i'm giving you experience back and how much tokens i need to use for a hardware level how much tokens i you need to use for my processing how much tokens i need to use to build a dynamic software experience i'm giving it to you an outcome i deliver that's a sum of all tokens that i use and then i keep my margin of how much margin i need and i will give you a token cost to you so it is like Everything eventually in the software world um could be paid by tokens, right? In the foreseeable future. ah the The difference between an AWS, Microsoft, Google giving the traditional service versus AI service, because AI service is measured on tokens and they are giving foundation capability, that's it, right? so I'm not giving anything more. I'm giving you open AI service. I'm giving you a cloud as a service.
Speaker: You can go to Anthropic directly. When you go to Anthropic, Anthropic does not have their own hardware, right? They are running on Azure. And this is where the big companies today, oh, NVIDIA is investing on Microsoft. Microsoft is investing on OpenAI. OpenAI is investing on Gemini. Or Gemini is, or Google is investing on Anthropic. All these things that keep on happening because fundamentally, there are three components.
Speaker: There is GPU. There is data center, there is power, there is software. right let's Let's remove the power supply out. right So that's being provided by the utility providing. So then there's only three components.
Speaker: There is GPU, there is data center, there is software. And if you look at these companies, they are offering one of it or two of it or all three of it.
Speaker: right Nvidia today offers a GPU that you and me can buy or companies can buy, build my own data center and host it or Nvidia is now already getting into data center business where they are saying, oh, you have the ability to buy my GPU as a GPU or I will host it in a data center and they'll give it to you. You just consume it as a service.
Speaker: But when you consume the GPU, you need a software, AI software that needs to run. Oh, what software you want to run? Now comes OpenAI and say that, oh, on NVIDIA software, or you can use OpenAI. On NVIDIA software, you can use ah Anthropic or ah you can use a Gemini or you can use a Lama. So these three players with three colors, they are working amongst themselves to give the power to an you know the the consumer, whether it's an individual or a customer,
Speaker: And but if you're in if you're thinking you're an individual consuming it, you are not consuming AI directly. You have in-between business layers that are making the AI consumable for you.
Speaker: You are using a healthcare AI, you are using video editing AI, you are using podcasting AI, or you are using banking AI. So there another business provider who uses the foundational AI giving you a specific domain experience for you to consume. So there are already layers.
Speaker: There is fundamentally GPU, then data center, then foundational AI layer, then specialized ai outcome, and then you as a consumer consuming it. So all of them are standardizing on tokens.
Speaker: So you have the ability based on your capability, you have the ability to consume the last mile, or you have the ability to consume in the previous mile or you have the ability to go consume from the bottom most layer. If you are consuming in the bottom most layer, I assume you are the richest person on the earth or you are one of the rich companies on the earth.
Speaker: ah So from what I understand, when you are ah an AWS customer and you are, let's say, ah you want tokens from Anthropic or Cloud, is AWS a reseller of Cloud tokens or is AWS actually running the Cloud foundational model on their data centers and providing you the tokens?
Speaker: In the traditional way, think of it, I would give an analogy as if it is like a reseller, but it's not a truly reseller. So basically you pay the token cost to AWS. So AWS collects the money from you.
Speaker: But internally what they do is they have agreement to pay with Anthropic. They take the Anthropic software, If Anthropic has funded the data center, then they like their distribution, percentage distribution model will be different. If AWS, say, in Ireland, i built my own data center, I have taken Anthropic software alone and I'm running it on top of my hardware, I'm collecting the token cost for my customer.
Speaker: In Ireland, they may say, oh, I'm paying Anthropic, I'll pay only, you know, two cents out of ten cents I collect. Whereas in California, you know what? you are building the data center, you have provided the software, I am giving you access to customers because I already have customers who can easily consume and they want to give a simplified experience to coexist with their existing usage they already have on AWS. So in that case, AWS may still collect the 10 chance money, but they will pay 8 cents to Anthropic because in California, Anthropic owns the hardware, software combination,
Speaker: and AWS becomes a distributor of their service. So the combination can be anywhere. and know It can be everything owned by AWS to everything owned by the fundamental provider.
Speaker: And it varies from every data center, every location, every geography. And that's why I said loosely, I said reseller because the the margins, the distribution, the last mile collector decides okay, you know, in this region, I pay you fifty fifty In this region, I'll pay 80-20. In this region, I pay 10-90. It keeps varying.
Speaker: Just like ah Coke works with bottlers in some countries or they would also have distributors. and So the bottler actually takes the Coke formula and creates the drink and sells it and gives Coke a royalty, whereas the distributor is receiving the bottles that they sell. in Yeah, yeah.
Speaker: It can go from complete owned model to a complete, you know, the franchisee model. Yeah. ah Wouldn't these AI companies be extremely protective of their models? Why would they let ah Amazon run the cloud model on AWS hardware?
Speaker: It's going back to supply, short supply of hardware. What ah cloud providers are doing is, because they they are sitting on hundreds of billions of dollars of savings or earnings they're sitting on,
Speaker: they go and buy the bulk order. This is why you see, oh Microsoft is investing $10 billion. dollars You know, Amazon is writing $37 billion. dollars Google is investing $60 billion. dollars You see all of it because these big companies, they have the money.
Speaker: They go to, and today, the the the largest you know GPU producer is NVIDIA. You know, there are other providers who are emerging, including AMDs and, you know, others are ah emerging, but their capacity, capability is nowhere comp comparable to NVIDIA. So NVIDIA being a monopoly.
Speaker: This won't be the case. Eventually, in the next 5-10 years, there is going to be other set of players who will start providing. So today, in the demand-supply gap, supply is really at the lowest level.
Speaker: Demand is the highest level. And the richest companies are going and blocking the supply available and pre-ordering it for the next so many years.
Speaker: So now, if you think I'm an anthropic, I'm coming to the market just as a software provider and I'm a startup. I don't have enough money like a Google and Amazon and Microsoft. Now, how will I go and buy 20 billion worth of Nvidia for me to run independently?
Speaker: So what I do is I go sign a pact with Google or I go sign a pact with Microsoft and Amazon saying that, you know what, you already ordered 60 billion. You're going to run some software, right? Why don't you use my software instead of OpenAI, instead of Gemini, instead of Lama, you buy my software.
Speaker: I will help you do this better, right? um you know That's the commercial models that are working in place. But everybody today in the industry, if you look at it all the big cloud providers, all the big model providers, all the big data center and hardware providers, they're hedging their bet on everything.
Speaker: right If you take Google, Google pays to OpenAI, Google pays to Anthropic, Google builds Gemini. So between OpenAI and Anthropic, they compete with each other. They go and pay to NVIDIA while they are building their own TPUs.
Speaker: right the TPUs and GPUs are fighting in the market. It's the same thing. you know If you go to Amazon, same story repeats. If you go to Microsoft, same story repeats. If you go to Oracle, same story repeats.
Speaker: Everybody is hedging a percentage of their large-sized spends. Those are on tens and hundreds of billions of dollars. Expecting whoever wins at the end,
Speaker: everybody will be a winner. but Right. Okay. Fascinating. I want to dig in a little more into your prediction that everything eventually will be sold as tokens. um So what what you're saying is that today, if I'm a Salesforce consumer, I'm paying on a per seat basis, for example, and then there are different tiers, you know, there'd be some basic tier and there'd be an advanced tier and so on. So eventually, this whole UX of using Salesforce, the whole software experience of Salesforce is going to be irrelevant when compared to the intelligence that you are using, like like in terms of the cost. the The cost of serving the intelligence will be so much higher that the cost of building the UX and maintaining the UX and updating the UX, all of those things. costs which used to be the primary costs are going to be negligible and so therefore Salesforce would just put a markup on the tokens that they are using to serve you and that markup will also cover their cost of creating the UX, maintaining the UX etc. That is what you see the world going towards.
Speaker: That's probably that's probably ah probably one of the one of the easily thinkable adaptable models that will level evolve. But where I foresee it will mature to is your outcome is going to be measured not on consumption but on experiences.
Speaker: and So today, you know, you don't have a method or in the traditional world, you didn't have a method. You were having methods of, okay, how many employees I had, I'll pay it on number of seats.
Speaker: As I said, in know a if if I am using heavy processing, and I could pressure it on throughput. In some cases, it was measured on consumption, like including cloud providers. i know I know my total cost of building a data center.
Speaker: I may give it to you as a compute. I may give it to you as a database. I may give it to you as a storage. Based on some consumption, I'm i'm giving you a consumption-based pricing model. But what is going to evolve is an experience outcome.
Speaker: So you don't need a lot of the software layer. right i know Fundamentally, you have a hardware which has the ability to process trillions and trillions and trillions of data in seconds.
Speaker: And all those data needs to sit on a data layer. Now the model is the one going to work. And traditionally, like your booking engine, like today, you can go to OpenAI and say, book my ticket, right? Point A to point B, look at the cheapest ticket, use, i mean, you can just converse with the the browser and you don't need to go and click through one by one, enter the data, you check everything manually, nothing and but nothing needs to be done.
Speaker: You just tell your experiences in 100 words or even you speak to your OpenAI and OpenAI makes the exact same booking probably in the least cost possible that you are not able to do it for yourself in a sophisticated way.
Speaker: So you don't need the UI experiences, right? UI experiences will more and more will shrink. And what is going to become a measure of outcome is real experience outcome you are getting.
Speaker: So you will be able to say, oh if you are able to book me a ticket of point A to point B, yeah you know, in in this particular date timeframe for $200 or less, in less than two minutes, I am willing to pay even $5 more.
Speaker: Right? If you are going to book it in the next 10 hours, you take your own time, you wait and figure out, but you give me the booking in the next 10 hours,
Speaker: I will pay X money. right So if you're in an emergency service, it's it's like you're you in India, probably it's called tatkal servers. I'm under prepared. I don't have much time, but I need to get things done in a faster way. Then you pay the premium fee.
Speaker: I have time to wait and you pay a standard fee. And okay, I don't need all of this. Just give it to me as a commodity. i can wait forever, but I need it. right it can come to you as ah as a free service which the governments are giving it to you hey education is free your basic you know local commutation is free some of the food services are free your are basic 100 units of electricity is free you know things of that kind will also evolve so in the a world there will be a free tier There will be a standard pay tier for which time is the measure. Then there is an ultra premium tier where again time is the measure and experience is the measure. But you longer the wait, you may end up paying more.
Speaker: So you end up giving to AI more to make sure that your cost of supply chain is not going to go up. Cost of supply demand is going to hurt you.
Speaker: So it's it's an evolution that's going to happen. in my mind, it's purely based on outcome and experience and any consumption that is going to happen. There will be a supply chain limitation. There will be a supply demand or demand supply limitation.
Speaker: And the the new model of tokenized cost will incorporate all these units and give you a unit outcome. Okay. So what you're saying is that ah the way people in future will use Salesforce is not through their UI, which is rather clunky and difficult to understand for someone who's never used Salesforce, ah but it'll just be a chat interface. You'll say that, okay, I...
Speaker: Whom am I meeting today? What's the follow-up that I should do today? What's the status of my pipeline, which is a client that is likely to close soon? How much revenue am I predicting I will ah get this quarter? You'll just chat with an interface and Salesforce will just be ah doing a markup on tokens and that's how they will monetize. And based on ah how much intelligence and the speed at which you get your answers, there will be different tiers.
Speaker: Yes. you know it It will do all the workflows. You can probably just interact with it as if it's a natural language. And you can use it as an assistant and you can say, okay, send a follow-up to so-and-so and it will draft a mail and send it also. Absolutely.
Speaker: okay Okay. And this is where when Elon Musk says, and i'm I'm a subscriber of that philosophy as well, where oh work will become optional eventually.
Speaker: And before work will become optional, the number of hours that humans are working will start going down. ah Today, worldwide, we are working for 40 hours. I mean, we know that during industrial era and colonization era, the human beings have worked for 80 hours or more per week, but right? And that has gradually changed to 40 hours per week now, largely.
Speaker: And we can expect that in the next 10 years, it will become 20 hours per week. ah No human being will be required to work more than 20 hours up per week and all your you know compensation structures, everything will change to that while your experiences get better. And eventually, in an ASI world, ah artificial super intelligence world, humans doesn't to work. Work becomes optional.
Speaker: I'm not as optimistic as you are. I feel like there will always be inequality. Even ASI world will have winners and losers. Like there will be some human beings who will get disproportionate benefit of ASI.
Speaker: ah so So I have a keynote speech, you know, maybe it's for another session. I'm happy to walk through the session. You know, how AI will change the world? you know, it's my philosophy. i built I don't know if I can share on the screen in this mode, but i'm happy to walk you through. So largely my construct is AI will make foundational elements of civilization democratized.
Speaker: Over 2,000 years, right I'm taking you know existence of, I'll even take 5,000 years, ah humans, average life expectancy of humans until 1900, if you take, ah less than 50 years.
Speaker: The amount of mobility people had probably was few tens of miles, other than the warriors who would have walked for months together to go to the next territory and win the war. And probably, you know, warriors would have fought one or two or three are battles. That's it in their lifetime.
Speaker: Otherwise, typical average mobility of people was restricted to few tens of miles. um You know, if you look at the number of hours of work, it was 80 hours or more during that period.
Speaker: um you know So we talked about life expectancy to mobility to your you know survival period. right To do all of this, even to true two travel to 20, 30 miles, you had to earn your bread.
Speaker: Food was not given to you. You were doing all of it to put food on the table. Your commutation was something that you had to pay for. Your food is something that you have to pay for. Your health care, to live even for 50 years, you have to pay for.
Speaker: So you have to work to make all of this happen. Fast forward 1900, you know, fundamentally invention of automobiles, invention of electricity, invention of railroad systems.
Speaker: If you think, it has changed all of this to a new standard of what we are experiencing today, where life expectancy on an average across the earth is 75 to 85 years.
Speaker: Your ability to move around is pretty much across the world, but in about 24 to 36 hours. right Point A to point B, anywhere on earth you can move around in 24 to 36 hours.
Speaker: Food is still not something that is available for everybody, but largely people are not you know not starving. While there is percentage of population that gets only one meal a day, people are you know having food.
Speaker: Shelter, probably 70-80% of the people have shelters. Still, 20-30% of the world is still not having shelter. homeless Homelessness continues. This is what last 25 years have changed.
Speaker: If you have gone to a human being before Thomas Alva Edison invented electricity, if you said, you know, at that point of time, people would have comprehended what a light means because they have seen a kerosene light or a light lid on a, you know, wood or a charcoal, right?
Speaker: and you say that I'm going to make it available in a smaller form factor bulb, you put it, that would have people would have been comprehended. But if you have said, I'm going to use this electricity to invent a refrigerator, I'm going to use this electricity to invent a microwave, I'm going to use this electricity to invent, you know, air conditioning, meaning, you know, I will live in a controlled weather environment at any point of time, right?
Speaker: I don't think so people would have been able to comprehend how this is possible. How can you make and weather controlled the situation because appear in Antarctica it is cold or in Chicago it is cold. and In you know Mumbai it is hot. How can you control the weather? Nobody would have been able to comprehend.
Speaker: Today people can. i mean With the AI we are standing in that kind of a juncture of what AI will do to us in the next 25, 30, 50 years we are not able to comprehend. but What will happen?
Speaker: The comprehensible change I will tell you like how 1900 years to the last 125 years, you know from 35, 50 years of average age has become 80 years, right? This 80 will become 150. In the next 25 years, average life expectancy of human beings will become 125 to 150 years, meaning you can live on a healthy basis up to 125 years.
Speaker: You can move around on your own. You can live on your own. You are you you know you don't have shrunk skin. All this will be possible because of healthcare improvements, DNA edits, prevention of many diseases that can happen, eradication of many diseases that can happen. With all of that,
Speaker: you will be able to live for 125 to 150 years. On the other side, as I said, today you are working long hours to put food on the table. Food will become available for every human being on Earth.
Speaker: It will become a birthright eventually. Food is free. Shelter is free. Today, in some parts of the world, commuting for 50 miles is free. right i in Even in Seattle, you come to hey pretty much the state subsidizes up to 50 miles local local transport is free. Only my long distance transport I need to pay for, which will become truly a global travel will become payable in the next 10 years.
Speaker: Thousand miles travel will become free for everybody. So you can get on, hop in, hop out anywhere, local flights, everything will become free. which is not the case. right ah Ten miles became thousand miles travel possible. Now, thousand miles become become free. A shelter, everybody will have a shelter on Earth. Nobody will have a situation of homelessness because the AI and robotics and the automation will be able to produce enough food for everybody without human labor.
Speaker: Without human labor, we can produce shelter for everybody and that eliminates homelessness. Now you will ask a question, oh, if everybody gets a home, how do you differentiate between the rich and the poor?
Speaker: Basically, what you're saying is the the unaffordable class becomes the you know the the lower middle class. And what will happen to the rich class? Today, Earth composes of 71% of water and of of the of only three percent age of her is what is ever human habitated
Speaker: 26% of Earth today is unhabitated. what And this has happened over the last 5000 years of civilization. right What I anticipate in the next 50 years, additional 1% of Earth will be habitated by the humans, not to create the basic homelessness go away, to create super experiences that humanity has not seen.
Speaker: What is super experience? Until 1900, the super experience is a palace. It's a large Colosseum construction. Broader, wider, but shallow constructions was super experience until 1900. Between and the super experience is a skyscraper You go to a Dubai Burj Khalifa, you go to a Taipei 101 and say, oh you know what? I am standing at, you know, 3000 feet above earth on a constructed building.
Speaker: But there you see what is the circumference area that you have is only few square meters, and right? A few tens of square meters. Now combine this two, can you build acreage of experience at 10 kilometers above space.
Speaker: This is where the blue origins and the SpaceX of the world already starting to take humans to. On Earth, the combination of horizontal and vertical experiences will be combined and built on the new one percentage of Earth because this Earth may be desert, this Earth may be a ma Amazon, this Earth may be, you know, mountains like Everest.
Speaker: you know, nobody has built a, know, Rich Scalton experience center on Everest. so AI will make a Rich Scalton experience center possible on Everest.
Speaker: So the top 10 percentage of the earth, the so-called rich, will move to start living on experience centers. The today's middle class will start to live a life of today's rich class.
Speaker: Okay, I'm living on a Taipei 101 80th floor, which I could not afford earlier, right? I can live in a large suburban, you know, single family home, I can buy few cars, you know, all of that are possible.
Speaker: When the poor class the unaffordable moves to say oh I have a shelter I can commute I'm eating every day. I'm wearing good clothes, right? I can watch content but that is available because internet is going to be free for everybody so AI robotics automation is going to change the world in a way human beings are not able to comprehend today. That's why you are thinking, why I'm not optimistic this much because you you are an individual 1900 at nineteen hundred who has not experienced electricity, but Thomas Edison comes and says, you know what, I'm going to have a, you know, refrigerator in your house, microwave in your house and air conditioning in your house where you can cook, you can heat your milk or water in less than 30 seconds and your temperature 24 hours will be controlled. You don't want to wear thermal clothing or a summer clothing in your house.
Speaker: where you are like, what? That's not matter possible. yeah You are just sitting in that experience level. So now, hope after my explanation, you're able to see through how life will change forever and your your ability of comprehension has increased now, I'm assuming. so Yeah, I mean, I see your point.
Speaker: but Again, think of it. You don't have buy the point. you know this is This is inevitable. it It will happen. It's only a question of whether it will happen in 5 years, 10 years, or it it's going to happen in next 50 years, right? You can always argue that.
Speaker: And that is the limitation on power supply and hardware creation, because the more power supply earth has, more ah hardware that we can generate, the the intelligent hardware that we can build. Now, all of labor work that is happening by human beings can be outsourced to robots machines intelligent missions and the machines will constructing all the things and it is no longer going to be human beings human beings are going to be controllers of it so what will human beings do i get more time in hand i'm going to live 125 years working only 20 hours a day i mean 20 hours a week that means that i can consume more entertainment
Speaker: I can go, you know, sports will become the most, you know, energized ah thing in the in in the newer generation of experiences because everybody will get into fitness, entertainment, sports, food and culture, travel, unexplored territory of world.
Speaker: Today, if you think that, you know, only, you know, single digit percentage of population has even gone out of their country, right? You know, that will become like 80% of the people would have gone out of their country because they have more time and commutation becomes more easier.
Speaker: Again, one of the fundamental things that will change during this transition is point A to point B. Today, we are saying 24 to 36 hours, right? That will become less than four hours. In the next 10, 15 years with the advent of AI, point A to point B, we can go to any place on earth in less than four hours. Imagine the logistical unblocking that is going to happen when point A to point B things can move in four hours.
Speaker: I can sit in Seattle and order a delicacy from Delhi or Guga Woon or a Mumbai or Bangalore delivered in four hours and eat the fresh food.
Speaker: right again, if you comp compare, try to compare and comprehend it, it's a limitation of a technology, not from breeding, but not being able to power up today.
Speaker: So, inevitably, what I'm saying is going to happen. And the 10% of the people will make sure the technology is not used in a bad way. It's not, I mean, any technology from internet to, you know, AI, today you want to use it in a good way, you can use it. You want to use it in a bad way, you can use it. There are bad elements always happening. It will happen with AI, but there will be governments companies like CoreStack. There will be Godrail providers, including the AI technology providers. Like, for example, Anthropic, when it said,
Speaker: that to the US government, they want an, the US government wanted an, you know, uncensored version of Anthropic, what they can use for everything. But Anthropic as a company said, no, it is way too powerful. I cannot give it, even if it is a government, I cannot give it. So that layer of governance will get built.
Speaker: So AI... will not have a catastrophic outcome in such a way that we all decide like the you know the the the last edition of Tom Cruise movie, right? The world will end if this AI is not controlled, right?
Speaker: I don't think so we will have that kind of a situation because people who are innovative, does not have anything else to do, have to work only for 20 hours, are going to build Godrails that will keep AI in check.
Speaker: So a lot of prosperity is what we can expect overall. There will be some damages that AI will inadvertently create that exists with every other technology.
Speaker: But I mean, that's why 10 billion people are going to be there to make sure that it is used for humanity and not for destruction. What is the the future of CoreStack in such a world?
Speaker: ah We are the god rail providers, right? For AI, we make sure the AI models, AI tokens, AI... you know, the the so-called agents are supposed to do what they're supposed to do.
Speaker: It's not doing things that are not allowed, that are unethical. da Is this a present capability or this is a capability in the pipeline? ah I mean, we are we are building faster with AI, right? So earlier, our releases used to be, you know, once in a quarter, then once in a month. And now every week we are producing.
Speaker: So, you know, as we are talking, you know, we have, Many of it has our current capability. Today we can manage the tokens. We can give visibility control operations. We can control your models. We can control your agents. But can it take automated autonomous actions? Those are in the pipeline. Hey, what are all the things that can autonomously act? Like CodeStack traditionally, as I said, save money for my usage, right? So that was autonomous action. But today, save money based on my AI usage
Speaker: disallows certain AI usage, right? If somebody is coming and asking how to build a nuclear bomb, right? Today, AI itself disallows it. I will not give you this, right? If you get a go and ask AI, how do I commit suicide, right? AI will not tell you.
Speaker: But now think of it in a business use case, there can be a legitimate way of asking a destructive thing. Hey, how do I delete an employee who left? All their content has to be deleted.
Speaker: You know, give me a method to delete delete all their footprints. right An employee can take that and apply it on an existing employee and and destroy them. right So, these are all the use cases that COSTAT now governs, monitors and make sure that it is legitimately used with the legitimate circumstances in thenron the right en environment and don't allow a legitimate action in an illegitimate environment.
Speaker: a legitimate action in an unallowed situation. that We are building that governance guardrails. So technically, we are one of the guardrail providers of AI being used for constructive purposes.
Speaker: So ah I understood how the legacy business works. You have read and write access to the dashboards. You have algorithms which compare how ah the dashboard data is showing up versus what is the ideal way. And then your solution is able to autonomously make changes if it is so allowed ah or it is able to flag it ah to the user that this should be changed. And those changes can be on finance or on security side.
Speaker: ah Help me understand how you are doing AI governance, because the cloud governance is very clear. You have that dashboard of Azure, etc., which on which you are able to read and write. But how does it happen for ai consumption?
Speaker: So the current level of ai governance and and the current usage model, first of all, because the demand supply gap is there, Everybody and and the cost of AI is expensive because of the demand supply gap.
Speaker: And everybody is bothered about cost of using and get a higher return on investment on every dollar they spend. So you need, first of all, visibility of what you are using it for, what AI you are using, how much you are paying for, and is is the outcome that you have gotten paying a unit cost, is it worth it?
Speaker: Or could you have gotten the same outcome a traditional technology? How do you get access to all of this data to evaluate? Today, AI is consumed primarily through
Speaker: utility model, right? Either from a cloud provider or from an NVIDIA kind of a data center provider, you are consuming it, right? Even if you have it in a data center, inside the data center, you bought the entire rack, right? You paid $5 million, dollars you bought the rack, cost is no longer an issue.
Speaker: Whether you use it, you don't use it, you've already paid $5 million. dollars That's a max cost. Right. In that scenario, the problems are different. I'll come to that. In a consumable model or in a utility model, the problem is I'm continuously consuming, I'm continuously paying for it.
Speaker: is my usage versus consumption is giving me returns or not. To identify that, you know, am I who's using, what is it being used for, why is it this being used for, how much it is being paid, all that in the utility mode.
Speaker: Translate, I'm calling utility mode as a cloud consumption mode. So in the cloud consumption mode, we already know how to get the fingerprints of cloud consumption, you know, base prints. So we get that and then we analyze it dynamically and tell you, you know what, you you basically asked how many letters language has.
Speaker: You know you don't need AA for this. You paid $2 for this. It is it is worthless. Whereas how to land a you know rocket and moon, that's a very valid question. That's you paid only $10 for it. It's way worth it. So we can make a dynamic decision based on our algorithms.
Speaker: So first do we bring visibility. Then we apply what is acceptable cost, what is not acceptable cost at the first level. Next, come to the model. right This is at the token level.
Speaker: At the token level, what is worthy enough to spend spend on token, what is not worthy enough to spend on tokens. First, you need to observe all the tokens and then make a determination of this is good, this is not good from a cost standpoint.
Speaker: Come to models. On the models, like, you know, is is my employee is allowed to use, that you know, Anthropic and Gemini and CharGPT and Copilot and in all of it and different levels of access and data, you know, grant privacy levels are granted and HR can access every employee's personal data.
Speaker: Now, an HR employee can also misuse it. right It is in the model. The model can do things based on certain conditions. But if I am impersonating as an HR director of a company to go access things that I am not supposed to access, then the model will allow it.
Speaker: right So now you need to guard the model from outside to say that, hey, it's an impersonating employee. Don't allow that. It is not a normal employee. right Somebody has to watch outside and say that model, I'm not controlling you, but I'm only telling you, you can do this, you cannot do this. So you start controlling the model.
Speaker: and right So that is the model governance that comes in. Then comes the agent governance. Now, agents are, as you're interacting, you're going to ask, like SAP, as you said, you talk to an agent and said, I met with so-and-so customer. There's a prospect. There's a detail. There's information. Go send an email. I've sent all of it. right The agent is going to do all the actions.
Speaker: Assuming you said, send it to tomha Tom Hanks. right Instead of sending it to Tom Hanks and the customer, it sent it to Tom Hanks, the actor. So you need to govern the agent is not inadvertently going to do something and reaching a person which is not. I'm i'm giving you very lame use cases. So everything needs to be governed at different level.
Speaker: It's not a model. It's an action that inadvertently happens. And this is where agentic governance has to happen. Again, you and me as two agents exchanging some information, we are not allowed to. Again, that comes in the agent governance. So, Corsac has the ability to do the token governance, model governance, and the agent governance.
Speaker: What will it evolve to as the next stage of AI consumption? We don't know. But the consumption of customers is going to drive the need.
Speaker: Customers are going to say, oh, this is what I need next. I need autonomous capability. I need ability to predict and prevent. right I mean, we have seen it with healthcare, care we have seen it with the financial world, we have seen it in the traditional world. What is the next evolution that is required? So I predict you know prevention and prediction is what is going to be needed. Autonomous actions is what is going to be needed.
Speaker: And eventually you know bring down my total cost of ownership to a fractional level of getting all of this done. So meaning I don't care you run it on Claude or you run it on Gemini.
Speaker: dynamically get me the outcome whichever should have given the local lowest cost so now all the software experiences you built on one platform may have to be plumbed dynamically on traditional platforms so the the evolution will come on its own as we have seen in in the past and every software provider will adapt to it The way you get visibility and therefore ah the ability to analyze and say, yes, allow, no, don't allow, is by virtue of the fact that all the AI consumption happens through code which resides on the cloud. Like, like let's say, Zomato has a customer service chatbot that that code of that customer service chatbot resides on the cloud and you have access to cloud. And therefore, you are able to see how that chatbot which Zomato is using for customer service is behaving, ah how much token consumption is going on. And if it is inefficient consumption, if maybe you need a small language model and not a large language model for the Zomato customer service bot, then you are able to give that recommendation.
Speaker: um yeah so So all of it comes from the fact that you... have full visibility of what's happening on the cloud consumption of the customer, including the software which is being run, the code which is being run, and things like that.
Speaker: Am I correct in that? Data and the amount of capacity required. See, the software layer, we are in a transition mode. It's it's like, you know, pre-electricity to post-electricity. It took 25 years for humans to adapt, or 50 years probably to to adapt. For in the era, it will probably take only a few years.
Speaker: We in the transition phase. In this phase, the old software... And in the new mode, you don't need a software. yeah is That is the model, right? It is the model.
Speaker: The model can directly work on the data and give you an experience. In the traditional world, that same layer existed, but because you didn't have the ability to process, you know, humongous amount of data, you had multiple layers and then experience as a web UI or an app UI or something of that kind.
Speaker: You may still need a web UI and app UI, but the software itself will shrink and shrink and shrink to an extent. What will happen is to process it, a software needs to be written, but it will be dynamically written for your query on the fly.
Speaker: It won't be a persisting software continuously that exists to be used by multiple people. Rather, it will be a dynamic software the model has built on the fly in in less than five seconds for you, processed it, and then the software went away. right and That is the model, I mean that is the evolution stage we will go to.
Speaker: Meaning, software, after after three, four years, ah software stack importance will become like very, very lean. The model will replace the software. Yeah, because the the model is better at UX design. Everything from UX design to writing code, the model will be better at all of that. So on the fly, like my experience of a Zomato app to order food would be very different from your experience based on whether you're vegetarian or not or what kind of cuisine you like or things like that.
Speaker: it and And it will have, I mean, it's a lot of predictive, preventive and based on historic factors, right? It will know all of it. hey you you won't even go say Zomato, right? You have the best, fastest to delivery with high quality, low cost.
Speaker: You know my preferences. And you know, today is you know what day is today Get me the food that I would prefer today. That's all you would say. and And it knows your, you know, whether you have had your breakfast or not, how much of calories what you consumed, how much of work you did. you know, it it knows all the biological things that have happened.
Speaker: And if that thinks that you need to eat more carbs, it will deliver a carb food which you like. If you think that it needs to deliver deliver a protein food, it will deliver deliver a protein food, which you will like again, because it's it's going to become hyper personified experience.
Speaker: i mean, our our mobile phones, in my mind, have the maximum number of applications or apps as of 2025-26. Going forward, your number of apps that you require will keep reducing.
Speaker: You'll have more of a chat interface which does things instead of having to go through. One chat interface. And this is why the open AIs and perplexities and the all AI providers are going towards a browser browser way. right They are giving one browser that will do everything.
Speaker: right You go to the browser, you just talk to it and it takes care of it. So that will become a super app. Okay, fascinating. I want to end by understanding a little bit more about the the business of CoreStack. What kind of ARR are you at currently? How much funds have you raised? Tell me a little bit of that as well.
Speaker: um I mean, we are a privately held company, so we don't disclose exact ARR, but I can give you a range. of but We are walking towards our 100 million in annual recurring revenues. Wow. By when will you hit? This year? No, this is the current financial year. Our financial year is March to April.
Speaker: mean, April to March. So we're walking towards our 100 million journey. We are existing in the business for 10 years. Now, this is our 10th year. We have raised an equity investments of about 85 million dollars.
Speaker: to build a $100 million dollar recurring revenue company. We do have some you know loans or ah debt investments that have come in, which is typically non-dilutive, non-EQD-based investments, which our revenues are paying for.
Speaker: So I don't consider that as an investment, but it is a capital that was available for us to build. ah From a total investment standpoint, less than $100 million to build $100 million dollar and anything going forward,
Speaker: We may likely be able to do it without any additional investments, but it will take more time. If we take additional investments, then we can be at a AIA speed to deliver more because today, if an AIA company are not spending on tokens, then you are not considered AIA company.
Speaker: right So today our token spend will be probably less than a million dollars. right So you know that's that doesn't qualify CoreStack to be a true AIA company. at a hundred million dollar revenue journey because if you're a hundred million dollar company, you are expected to spend at least few tens of millions of dollars on tokens.
Speaker: And, you know, and that would put us in a $500 million dollars revenue growth from 100, right? So we definitely have an opportunity ah but to hypergrow using more AI with additional investments.
Speaker: But given the current market conditions, we are, we we We are tending to be capital-efficient company, cash-positive with moderate growth. I believe you also took funds, like major funds, pretty late after inception, a couple of years after, like like you bootstrapped in the early years. Yes, in 2015 is when we had the idea of CoreStag to build an intelligent platform that helps customers maximize their ROI, increase velocity in their digital journey.
Speaker: That was our foundational principle, our vision with which we started. ah We wanted to build it for multi-cloud. a you know The world was not ready to accept as a multi-cloud.
Speaker: The world did not see the problems that we foresaw needs to be solved. It took probably three, four years for the world to accept these are all problems in cloud and this needs to be solved.
Speaker: When you are little ahead of you know time with a problem you are trying to solve and the world doesn't comprehend but that these are all the problems to be solved, you don't get funding because people think it's a problem worth not solving.
Speaker: So, but we were determined to that the problem to be solved and we went ahead bootstrapped and solved and showed that this problem should be solved. By the time we we started demonstrating customers who are appreciating the problems we're solving, the world acknowledged already, yes, this problem needs to be solved.
Speaker: And we started getting money. watch Right. the When you're ahead of time, people are not able to calculate your TAM, your total addressable market, because of which… Yeah, I mean, they they't even they didn't even think that the investment world, the analyst world, the technology world largely… at that part of time did not see what we foresaw as set of problems to be solved when this explodes.
Speaker: you know and in In a funny way, you can think that you know as much as you're saying, I see what why you are saying how AA will change next utopia world, but you are not in a position to acknowledge that this will all happen.
Speaker: You see it, but you don't want acknowledge it. right We were in so similar situation. They saw, yeah, I mean, some problems will happen, but I don't think so. It'll be, you know, tough enough problem to be not solved.
Speaker: And we said, you know, no, we believe that it it will be a challenging problem for companies. So we need to create this, you know, God rails. We need to create a foundation. And we went on to create. the next three, four years, the world has, you know, 100% acknowledged what we were saying was right.
Speaker: Essentially, the way to look at it is like if there's a trillion dollars of spend on cloud, ah then at least some percentage of it, some single digit percentage of it will need to be spent on cloud safety and guardrails and optimization. And that is the market which you are going after. And and that was not immediately apparent to the VC world that there will be trillions of dollars of spent on the cloud. That scale was not apparent.
Speaker: Yes. I think now the new order of you know velocity, typically in the investment market, in the you know acceptance market, in the valuation market, everything is going through a significant reset.
Speaker: So, today you no longer talk anything in years. If you are talking the world anything in years, you are like stone age or a iron age human being. So,
Speaker: The longest time that you can talk about is quarters. The shortest time you are talking about in hours and days today. right you You think of some idea, and know I have got it done that day evening or in the next couple of days or more complex, i'm I'm getting it done in the next few weeks. right So in that way, the world also, the investment world also is is trying to come to terms with it and understand where the world is heading towards, how the valuation is changing, how the value impact is changing, what is the real enormity of the TAM that is getting created, where the bottleneck is.
Speaker: i There is enough acceptance, but because of the macro market and macro economy conditions, People are still not you know comfortable in making sure that i know AI needs to be backed up or AI will succeed because they see that big companies investing among themselves. It is a vicious cycle getting created. you know this This is a pyramid that will fall. you know There are a lot of theories of that kind.
Speaker: But what we need to understand is today's combination of cloud plus AI is the moment of electricity innovation or invention that has happened.
Speaker: Electricity has changed the world how we have lived. So this is going to transform the way we have we are going to live. And people can debate on timelines, ah but not on the outcome.
Speaker: And it just acquires probably a few more quarters for that arrived acknowledgement of this is how it is going to happen. Because it is just, we are in the very topmost of tip of an iceberg, really tip of an iceberg of AI showing. Because until Anthropic Cloud, Co-Work and you know all that came in, ah you consider AI as a, okay, it's it's fast moving, but not the fastest animal on earth.
Speaker: But and just in the last 90 days, that realization of, oh, AI can be the fastest thing on Earth, faster than speed of light.
Speaker: You know, we thought it was faster than speed of sound first. Then we questioned whether it is faster than speed of light. Now we are coming to terms with, yes, AI is ai can be faster than speed of light.
Speaker: So it requires... probably few more quarters. Once we all accept that this is the new norm, I think we'll all start doing things in in an order that we have not imagined even a year ago.
Speaker: Amazing. What are your exit plans? Do you have an IPO on the roadmap or do you think one of the hyperscalers will acquire you? Because ah you do seem to be a good fit to get acquired as well.
Speaker: I mean, it it can happen. We have opportunities always. Our competition, large players have consistently been reaching out.
Speaker: Who is your competition? There are multiple companies you know you know who are providing point solutions. Nobody provides an integrated set of outcome like a single pane of glass from finance to security to operations to you know AI control to visibility. no Nobody provides it. It's all point solutions today. But there are companies like Cisco,
Speaker: like Cisco has multiple tools, like you know ibm has multiple tool sets, there are companies, privately held, PE held companies, they have multitude of sets of tools, ah not like COSTAC.
Speaker: So these companies would potentially be interested in acquiring COSTAC? hyperscalers will be interested even the AI companies, right? You know, a open AI may want to, you know, build all of this with an open AI or, you know, Anthropic may want.
Speaker: But my belief is if a hyperscaler or an AI company buys, because we are like the the three or four pillars of governance, right? In every country, you have a political system, a judicial system, a media system that makes sure that know each system does not overpower the other and there is sanity that is maintained, right? So similarly, for the fast-moving technology, governance needs to be an independent layer You know you can comp compare core stack kind of technologies and any governance kind of technologies to a judicial system or a media system, something that sits outside to control the most powerful system rather than being inside of the powerful system.
Speaker: And as humanity, we have always seen that yeah we are not going to trust, hey, I am a self-governed AI. right Would the humanity ever believe Maybe there is a future possibility, you know, the the next two generations may may trust self-governed AI, but that is when the Tom Cruise end of the world might happen.
Speaker: So you don't want to get to a end of world situation. So then you need to keep the governance outside, not as a self-governing AI. So in that model, I foresee that if we get acquired by either large cloud players or large AI players,
Speaker: then someone else will have to be born to deliver what Kostak offers today from an outside perspective. So it is better for Kostak to stay outside and go and continue to thrive in this model.
Speaker: That way we are thinking of an IPO probably in the next and four years, i know get to become a billion dollar company. in in revenues and enjoy the benefits of continuing to make AI and and quantum and that in the next few years, quantum will also become more real.
Speaker: All these technologies governed well, useful for humanity ah is what I foresee at this point of time. But you don't know, you know anything can happen.
Speaker: How much equity is still with the founders? Between founders and employees, we control majority of the company. Like more than 50%, which is they pretty unusual for a company which has raised $80 million. dollars By the time you raise such big money, you are diluted down to teens. right so yeah As I said, you know we had a philosophy you know of creating wealth because when first three, four years, there were no investors. right no Nobody believed in us. At that point of time, we believed in our employees and our vision and our customers.
Speaker: So we have created, we have given more than 20% of the company to our employees itself. right Between founders and leadership team and key employees, we probably hold 30 to 40 percentage and other employees about 20 percentage. So that's where we continue to you know have little or above 50 percentage of stake in in controlling the company.
Speaker: Amazing. Amazing. Thank you so much for your time, Izzy. This was a very wide-ranging conversation and I really enjoyed our time together. Thank you, Akshay. I really appreciate you know hosting me on your podcast, Founder Thesis, and we've talked long enough. Hopefully, some of the perspective I've shared has been useful will be useful to the podcast listeners and the larger tech community out there.
Speaker: you know Thank you very much. appreciate
