Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Cloud Commitments Without the Lock-In with Archera's Aran Khanna
0 Plays3 seconds ago
Scott talks with Aran Khanna, co-founder and CEO of Archera, about a new category of cloud financial tooling: "Insured Commitments." Instead of locking into 1- or 3-year reserved instance contracts and hoping your usage matches, Archera offers commitments as short as 30 days. They get into the economics of cloud purchasing, how AI workloads are changing capacity planning, and what FinOps looks like in 2026.
http://archera.ai
Recommended
Transcript
00:00:00
Speaker
i was at I was at Costco and I bought, I want to say, 900 Advil for like 25 bucks. And then I was at the Hilton a couple of weeks ago and I bought two Advil for $5. It was unbelievable. So I'm starting to think that like on-demand pricing is the convenience store price of compute and not the real price.
00:00:26
Speaker
I think that's that's generally it, right? it's It's almost the promotional price. That's that's where you want to prototype at. That's where you want to like understand, hey, you know this works. But then if you're actually going to run the thing.
00:00:38
Speaker
Hi, I'm Scott Hanselman. This is another episode of Hansel Minutes. Today, I'm chatting with Aran Khanna. He is the co-founder and CEO of Archera. How are you, sir? I'm doing well. Thank you so much for having me.
00:00:50
Speaker
So I, as a individual developer who does like side projects and startup consulting, I kind of pay as you go. And I've been doing pay as you go in the cloud for gosh, 17, 18 years. But I'm realizing that if I were to get serious about the cloud, I could probably save a ton of money if I would just commit.
00:01:10
Speaker
But then I'm realizing that cloud commitments are like this financial instrument that's hiding inside of a developer infrastructure. And I just don't want to buy a year's worth of compute.
00:01:22
Speaker
And Archera seems to have a solution for that, for for you know for at least most businesses. Is that right? Yeah, absolutely. As you said, you know, commitments um are are almost on their face to some extent antithetical to what the cloud promise was to developers like you and me, which is, you know, pay for what you need, turn it on, turn it off and, you know, pay as you go. But, um you know, my background is I actually spent almost a decade working for the hyperscalers, working for Azure, working for AWS in launching a lot of these services. And what you realize is there's a ton of capital expense that goes into these projects. And so having committed revenue from customers is actually incredibly valuable. And in fact, as the cost of data center builds go go up, it's actually more and more necessary. And the main way that historically the cloud providers have incentivized customers who are really sold on this pay as you go model as you and I were when we first started in the cloud building and experimenting and doing hobbies. was to give deep discounts for actually committing to infrastructure versus running it on demand. You know, these things can go north of 80% if you take something from a, you know, second by second usage pattern to a three year committed usage pattern. And most developers, obviously, especially if you're running a small hobby, you don't know what's going to happen with it. um
00:02:41
Speaker
They don't feel really comfortable committing there. They're willing to pay the premium. But when you get to running a business on the cloud, especially a serious business running at scale, you know, we work with some of the biggest cloud customers in the world here at Archera. You have portfolios of hundreds of millions, if not even now billions of dollars committed to the cloud providers. And it's become you know one of the the largest problems that we see in the cloud ecosystem today, not just the the technology problem, but the financial problem that precipitates from it.
00:03:10
Speaker
Yeah, it almost seems like we as developers were so enamored with the idea of this infinitely scalable elastic cloud that we kind of forgot it's still hosting and there's still machines there. And if the machines aren't spinning, then who's paying for all of that and they're being wasted?
00:03:27
Speaker
So then that means that if I'm doing pay as you go, I'm effectively paying a huge premium for the elasticity, for that bursty kind of thing. And as a startup, you would think I would want to actually pay less. Like you said, 80%. mean, I've seen 50, 60, 70% like crazy upfronts.
00:03:46
Speaker
crazy upfronts Buying a VM is pretty reasonable, but buying it for a year is like you pay for four months and the rest is free. I'm leaving money on the table, but then I i worry that what if my company goes under or what if what if I need more? What if I need less? I i just don't want to spend. i want to spend the least amount possible.
00:04:07
Speaker
and That's exactly it. And what's really interesting, you know you talked about that illusion of infinite capacity before ah that existed in the cloud. Now with with tokens and GPUs and capacity shortages in AI, like that illusion itself is almost broken. right We see that often in 2026, you're required to get a commitment just to have access to something like a latest generation or even an older generation GPU nowadays, given the the capacity crunch. and And that's really what our Chara is kind of holistically helping customers solve for. If they need to commit to get access to capacity or if they're interested in getting the discount, but they have uncertainty. They don't know if they're going to be using this GPU for the whole year. They don't know if they need all this compute you know three years on you know their business, if their startup might not even be around or it might be doing something completely different. And what we do is we help them hedge that downside risk by a unique product or the only ones in the space offering it called insured cloud commitments, which is essentially as simple as saying, hey, if you commit for three years, but you don't use say the last 24 months, we'll pay you back the difference for that whole 24 month period if you're basically having a commitment sitting there in your inventory that's not being used. And this gives customers the confidence to commit with that downside risk protected and capture more of that discount, you know, get resources that actually require commitments they otherwise wouldn't be willing to commit to, like GPUs, and really provide a new option they can put into that overall commitment portfolio. What we see is that customers generally you know the most sophisticated biggest ones that we work with. um you know Like Hex has become a really big AI native customer that's been running with us for seven years. you know Traditionally, it would be a mix of on-demand, one-year, and three-year commitments. Now when they're mature, they're actually looking at a mix of on-demand, insured one-year commitments, regular one-year commitments, insured three-year commitments, regular three-year commitments, and it becomes a lot more of a dynamic portfolio that helps them really match the risk in their business with the kind of best pricing that they can get out of these cloud provider constructs.
00:06:14
Speaker
So the cloud is an abstraction layer. And as a developer, I get to pretend that there isn't capital expenditure. i don't get to pretend that the giant building and the giant data center doesn't exist. I just, you know, I i burst when I want to burst.
00:06:30
Speaker
But let's pretend for a second that I've never heard the words FinOps or reserved instance. Maybe talk to me like I'm five. There's a giant Costco with a bunch of computers in it and they're not all working very hard.
00:06:43
Speaker
Why is it okay for them to give me 80% off and why do they want reserved instances? Yeah, so, you know, ah a core part of this just goes back to how these services, like you said, get built, you are actually going and buying machines and putting them into a data center, which takes a lot of money up front. In fact, the best analogy I found in an explain it like five I'm five sort of way is looking at this almost like building an apartment building. You know, we all live somewhere and there's a big apartment building down the block. It takes a lot of money to go and build these apartments, but then you got to sell them and you could put them up on Airbnb. You'll probably get the best price, but you might not fill it, you know, the whole year, right? You're going to have a lot of vacant apartments if you're only Airbnb-ing and you might not recoup that cost. And in fact, you generally have a loan against that apartment and they don't want to, so you know, underwrite Airbnb revenue. They want to underwrite, you know, one year, three year apartment leases where there's a guarantee of spend every single month coming in, not some variable expense.
00:07:43
Speaker
The same way these cloud providers go and take loans, you know, CoreWeave is a very extreme example, but all of them do this, to go and put up GPU capacity to build these data centers. They don't want to just rent it out on demand with no guarantee. In fact, they probably can't secure their loans if they do that. They want to have those one year and three year leases.
00:08:00
Speaker
And the same way as when you rent an apartment and you sign a one year lease versus a three year lease, you're going to get a better rate. If you sign a one year lease on a database, you're going to get a better rate than running it on demand. But if you sign a three year grade lease, you're going get best rate. And then really the way to think about our chair and all of this is thinking about essentially having the right to break that lease. So let's say, you know, get a new job in Memphis, you know, you want to get an apartment there, but you need to sign a lease. You don't if you're keep the job after a year, but you sign a three year lease, you get a great rate. We'll essentially sell you the right to break that lease after that first year of your job circumstances change. means you can get the right capacity, you that right apartment. You can get it at a much better price than say Airbnb-ing the whole time that you're in Memphis. But you have that downside protection baked in.
00:08:47
Speaker
So is this a trick or is this, the do the cloud people know you exist? Because if I break my lease, the landlord will be upset with me. But if the landlord knew that I had Archerra insurance, they might be okay with it.
00:09:01
Speaker
Yeah, and and that's exactly how we've gone to market actually, is we've partnered incredibly deeply with all of the hyperscalers, AWS, Azure, Google. In fact, you know most of our revenue actually comes through selling on the cloud providers marketplaces. ah So the Google, Azure, AWS marketplace is where we do actually most of our revenue and most of our volume. So we're really tightly integrated with them and they love it. In fact, their sales teams love it because they are incentivized to sign bigger deals, to sign these commitments. And often it's very difficult to get customers to commit larger amounts and to commit for longer because of that downside risk aversion that they have. So having something like this they could bring into a deal really helps in terms of closing deals, getting more velocity in terms of their contracts and what they're able to deliver to customers.
00:09:51
Speaker
um And then obviously the customers love it because instead of sitting on something that's unused and getting mad at Amazon or mad at Azure, they actually have an out, they can take those dollars and reinvest them in new initiatives. So it's really value added for the whole ecosystem the same way that you know, Ford probably wants Geico to exist, right? They they want to have someone who insures these car purchases. So people will actually purchase the big expense of F 150 in the first place.
00:10:16
Speaker
Now, I had a fender bender recently, or my son did, and there was an interesting discussion as I was teaching my son how insurance works and explaining the different flavors of insurance that one could do. And he was wondering if the other person could have bought insurance and then pretended to backdate it and you know basically game the system. So we had conversations about insurance. How do you prevent abuse from someone who might know that their startup's going out of business and try to go and buy insurance and then and you know make you insolvent?
00:10:47
Speaker
Yeah, absolutely. Well, that comes down to the core of insurance, which is underwriting, right? And part of underwriting is to avoid this thing that's called moral hazard, which is essentially where someone goes and buys insurance and then takes a lot of risks because they have insurance. So there's a few kind of ways that we do this. I think the simplest way is essentially ah we ensure that the customer has some skin in the game. They get you know usage data when we work with a customer going back at least a year, if not more. And we can see what they've actually been spending money on you know, in their cloud provider, we can see some of that history. And those are real dollars spent. So they, you know, if they're going to try and trick us, they would have to spend a lot of money to do it. So I think there's kind of a a default guardrail that we have around, you know, when we even make an offer. Beyond that, we actually have, ah you know, now a petabyte plus of historical usage data across thousands of customers. So we're able to actually go in and very granularly understand, hey, other customers that look like this customer,
00:11:46
Speaker
how long do they keep their you know rds database up and we have very strong priors is what we call it in the science world on like what the longevity of these instances are and then finally there's kind of the last layer which is almost in my co-founder came from the finance world he was at de e shaw doing you know quantitative pricing and and uber after that it's really this portfolio construction problem because you actually want to build a very diversified book where You're not overexposed to any one instance type, any one GPU type, any one customer type, or any one customer in general. um And so that kind of last layer, you know, those are kind of the categories of the guardrails that we have in place to ensure that we're not getting selected against. We're finding, you know, deals that make sense for both parties, and we're able to surface that up in the right way to end customers.
00:12:33
Speaker
Is this that term I've heard, quant, where there's a quant who is the person who is really doing the deep statistical analysis, and the more diverse your portfolio as an insurance company, the better? That's exactly it, right? we're We're taking a very data-driven quantitative or quant approach. And in fact, we do have folks who act as quants within our business ah in terms of maintaining our underwriting algorithm, understanding where we should offer insurance versus pull it back, and and even thinking about like pricing very granularly. Where should we price a little bit more or price a little bit less depending on the actual implied risk that we're taking? So it's very much part of the moat of this business, the fact we've been doing this for seven years and have such a rich data set, such a broad array of customers, and have really built expertise in this very niche thing that we're kind of the pioneer in.
00:13:20
Speaker
So this might be a little spicy, but I'm sure you wouldn't appreciate a spicy question. Like when did cloud billing first feel broken to you? And, you know, was the pitch was only pay for what you use. And I'm curious if you think at enterprise scale, that's still true, or this is the the correct solution to that broken problem.
00:13:37
Speaker
Yeah, the cloud billing started to feel broken, frankly, when those two initial promises that we opened the show with, you know, the illusion of unlimited capacity, and then, you know, the pay as you go nature started to break down. Now the original idea of like a reserved instance, um it was interesting, but it didn't feel unusual when it came out in the early two thousand and ten s from Amazon. you know, hey, this box is one that we're going to keep up for a while. You know, we should be consuming this in a slightly different way.
00:14:06
Speaker
But, you know, fast forward to 2017 when we were launching SageMaker, you know, I was part of that original launch team over at AWS. And we were saying, look, for specific SKUs, we're not going to buy that much NVIDIA gear. this was This was back then. I guess they're buying as much as they can.
00:14:21
Speaker
And so we're going to have a limited and capacity. So people are going to have to make a commitment just to show they're bought in so we can give them some of that capacity. And so this idea of a commitment turned from something that could provide financial leverage and match sort of the customer's consumption pattern, you know, a a reserved instance, to something that almost felt like ah a hammer that everything started to look like a nail with. Oh, we have a limited capacity. Okay, throw throw a commitment in front of the problem and that'll fix it.
00:14:45
Speaker
oh, we we want to actually stop customers from going multi-cloud. Let's make a big commitment cult program like a private pricing agreement program and and th throw a Microsoft Azure commitment program and throw everything into that. Oh, we we you know we have issues around customers wanting to modernize their infrastructure. Let's just make a new commitment type called a savings plan and throw everything in that. So you you know this an initial idea that started to make sense around a cloud commitment, you know more and more stuff got piled onto it to solve more and more problems. And I think it started to become quite a spaghetti mess for customers to both understand and then ultimately to take advantage of in a way that made the most sense for their business.
00:15:24
Speaker
i was at I was at Costco and I bought, I want to say, $900. nine hundred Advil for like 25 bucks. And then I was at the Hilton a couple of weeks ago and i bought two Advil for $5. It was unbelievable. So I'm starting to think that like on-demand pricing is the convenience store price of compute and not the real price.
00:15:50
Speaker
I think that's that's generally it, right? it's It's almost the promotional price. That's that's where you want to prototype at. That's where you want to like understand, hey, you know this works. But then if you're actually going to run the thing, you know it's the imperative to start looking at commitments. and And maybe a one-year commitment doesn't make sense. Maybe you want to start with committing for – a shorter amount of time. And that's where something like Archerra's insurance comes in, where you're like, hey, this project works. I'm going to keep it up for a quarter.
00:16:16
Speaker
i don't want to be paying the crazy premium. I might even keep running it after the quarter, but I don't want to pay for the whole year. i don't feel comfortable there. And that's a great use case for something like an insurance provider like us to come in and say, hey, we'll share some of that risk with you. But by default, you're going to get the best rate. And one of the really cool things about Archerra is that if you get confidence that, hey, i'm going to use this to the rest of the term,
00:16:36
Speaker
you can just cancel the insurance and stop paying. ah So you're you're never really locked in. You're always able to go back to what you're able to do kind of by default with the cloud provider. But you have now a partner that can offlay that risk and allow you to kind of crawl, walk, run into those bigger commitments as the project matures versus being stuck at that promotional pricing where where you're not really going to be able to get a lot of leverage on your dollar.
00:17:00
Speaker
Yeah. Yeah. It feels like developers love elasticity. I definitely do. But like a CFO, they love predictability. They don't care what the developers think. And it feels like the modern cloud is arguably trying to sell to both sides.
00:17:14
Speaker
And it's something that can't fully coexist. And you are trying to make it coexist. You're trying to make both sides happy. Yeah, and and that's FinOps broadly, but I think in our space specifically of FinOps where we deal with you know rate optimization and commitments and upfront spend, we're very much at the forefront of that interface between you know the DevOps team, the infrastructure team, the office of the CIO or CTO, and the VP of finance, the FinOps team, the you know VP or the CFO, right? Yeah.
00:17:46
Speaker
it is nuanced because they want to see things and understand things in very different ways. But ultimately, they have the same goal, which is to get the best leverage on the dollar and the most predictability in a fundamentally unpredictable space. And so that's what we're trying to do is trying to give them guardrails and bounds around what do they are you know going to use and what they're going to spend, um but also provide, again, we have a completely free tool that comes along with you know, the underwriting and all that we provide and the options that we we give customers they can opt into to give them some of that additional visibility. And we have a lot of partners that help us with that as well, because I think without that visibility, which we partner heavily on and give a lot of tools away for free on, you're not going to find alignment.
00:18:29
Speaker
And that alignment then speaks to the core of the problem on like, hey, there's a principle agent problem here the devops person doesn't want to take risk the finance person wants predictability and wants to actually put some risk and some commitment in place that they can actually track against how do we reconcile that without that visibility you never get that conversation and then a solution like this never becomes you know clearly the go forward Yeah. You know, you would think from talking to cloud folks who are just out there buying GPUs and building new data centers that everything is maxed out at a hundred percent.
00:19:02
Speaker
And the only problem in the cloud is like build more spaces in a field in Utah somewhere. But it feels to me like the the the thing that no one's talking about underneath this is that there there is unused capacity and they need to find a way to to to sell that. And it seems like selling commitments is the way to solve that. I wonder what's true.
00:19:24
Speaker
I think it's it's a textured problem because for certain skews like take you know the grace blackwell 300 right. Those are very new NVIDIA machines, you know, tough to deploy because they require liquid cooling and a lot of power.
00:19:40
Speaker
There's a clear supply demand imbalance there. um You know, more recently, you know, you look at things like arm CPUs in the data center. There's still plenty of those out there. In fact, for the last two years over 50% of the new CPUs in the data center. on the AWS fleet were Graviton and ARM-based.
00:19:57
Speaker
But you know you look at something like an x86, Intel CPU in the Azure data center, at least in a popular region, those might be harder to get. And so again, it's a very textured problem in terms of like different SKUs have different demand supply profiles. But generally, i think in the areas of the market where there is a high degree of demand relative to supply, the commitment is the way that that problem is being solved. um It's not a clean solve, obviously, because, you know, people want to make sure that if there's high demand, they can get the capacity um and they might not use it. They might just sit on it, but at least they're renting it out versus it's sitting idle and not being rented in the data center.
00:20:35
Speaker
And again, it's, it's obviously a very dynamic market. But I would say that that's been the primary tool. And when we talk to customers, we work with a ton of startups um that are running on the hyperscalers and even the Neo clouds. you know, this is kind of the biggest hand wringing for them, how much do they over commit, so that they're ensuring their capacity and having again, a partner like us in the mix where we can say, Look, you know, you want to over commit 5x, we're happy to take one x or 2x of that, and risk share that with you for a price. um it It gives them some of that flexibility.
00:21:07
Speaker
It's so funny how the things that I naively would think are intuitive are are not when you look at things at scale. I'm thinking about, you said the word overcommit, and I'm thinking about times that I've been on a plane and they've told me they overbooked, or the times that you've rented a car and you do that Jerry Seinfeld thing where it's like, I have a reservation, and they're like, well, no.
00:21:28
Speaker
Like, no, you should have the car here. I reserved it. I should have my finger on the car. Like if they nail it, they're going to reserve everything. They're going to be at 100% or 90% CPU all the time and everybody wins.
00:21:40
Speaker
but if they do it wrong, then it's a problem. So it feels like right sizing, is it a technical problem or is it a organizational problem? Like is cloud waste technical waste or is it organizational waste?
00:21:52
Speaker
It's really a combined problem. And I think the way that we think about it is there's really two sides to the cloud cost ah kind of world. There's the usage optimization, which is saying, hey, it's how I'm running the most efficient way, let's say I'm like serving you know Netflix videos, is my cost per video served as efficient as possible? if Even if I just take the on-demand pay-as-you-go rate, am I architected in an efficient way? um you know That's very much an engineering question, an engineering problem. On the other side, there's rate optimization, um which is distinct from usage optimization. And that's kind of where we play, which is really, what have I spent in the past? What can I comfortably commit to going forward? And what does that stack of commitments end up looking like over time? And there's more fundamental questions there than just the infrastructure questions. It's like, how fast do I think my business is going to grow?
00:22:43
Speaker
Do I think I'm going to see customer churn? What does that mean in terms of what I'm comfortable committing to? In fact, even looking at that first world of usage optimization, you might say, hey, I have the ability to save 30% on my cluster. It's gonna take a two year migration, but I have that ability. Do I wanna factor that in ah to what I commit? So, you know, visibility, forecasting, and really planning that commitment portfolio is in that second bucket. and And again, that's more of a finance problem, but they're very related, right? These things, again, have to speak to each other and use some sort of common language as I was saying earlier.
00:23:15
Speaker
Right. Now, you said that you have petabytes of data, like you're getting aggregated data across, you know, kind of anonymized aggregated data across all of it. And that's how you are able to do your pricing. But you would think that companies would have their own usage data. So why would a company overbuy or underbuy when they themselves have years of usage data? Is that a blind spot on their part?
00:23:37
Speaker
Well, the thing is that, you know, there's kind of two aspects to this. One is just understanding that usage data in and of itself is difficult. That's why there's this whole vendor ecosystem of folks to help, which is getting visibility on cloud costs. Many of our partners actually kind of play in that space as an example. um And so just the expertise to understand distill and then forecast based on that data is non-trivial. um But even beyond that, there are known unknowns. So even if you have perfect forecasting in your data, known unknowns like how fast will I grow my business? If I am just towing that line and and committing you know to every marginal GPU as a customer comes on,
00:24:15
Speaker
What happens if we have a capacity crunch and I'm not able to get more GPUs to serve the marginal customers? Should I overcommit now? That's something that actually a lot of our startups are thinking about. Like, do I just buy a lot more now, a bigger chunk and grow into it and not take the risk that I won't have capacity down the line? You know, these are kind of the known unknowns. I mean, will my architecture change? Even if you are the person doing the usage optimization, you don't know if you're Your database architecture is going to look the same in 18 months. No one does. So it's really those known unknowns that having a partner like us can come in and help because we've seen now thousands of customers that look like you. So you might only see your own data. We've seen tons of customers that look like you's data. So we actually have a prior. We're able to price that in a sense more effectively than you might be.
00:25:01
Speaker
Okay, that's funny that you say that because that's kind of where my next question was kind of going, which is like all of this sounds simple, but pricing risk by its definition is hard. So I was going to ask what are the signals that would allow our charity to say this is a workload that is safe enough to ins ensure that's data.
00:25:19
Speaker
Yes. um Well, there's there's a lot, right? So it's what is the specific? and And just to give you a sense, we um are are pretty lightweight. We'll get to your to your point, kind of anonymized metadata on what is the spend, what is up, what is down, you know if there are tags on these machines like production versus R&D, basically just the the metadata around the billing and usage in the environment.
00:25:42
Speaker
um But even through that, you know just what are the machines being used? What is the type of business that this is? right you know Just a simple crunch-based lookup or looking at our CRM, understanding the employees and funding, understanding the tags on the machines, understanding um you know things like very specific you know services and ah their lifetime. So if someone is migrating and on something like you know, Azure VMware service, they're very likely to start migrating some of that capacity to AKS within a year. Okay, so I probably won't offer that much on the, um you know, Azure VMware service commitments relative to something like AKS. So there's a lot of really granular decisioning. And, you know, the beautiful thing about modern machine learning is that you can basically put this all into a feature vector and have, you know, a set of algorithms basically rank these things for you from random forest through to deep neural networks um and actually come up with some of these statistical correlations and outcomes of the historical data that we're able to apply in a future looking way.
00:26:47
Speaker
Now, but all the data you have, is it all people with insurance or because wouldn't insurance by a definition change customers behavior yeah and someone what would take a different risk if the downside is bounded?
00:26:58
Speaker
Yeah, actually, and it's a great question. It speaks to that moral hazard point. So so two things. One is we actually don't see um too much moral hazard because of some of the things like you have to run the workload for a certain amount of time beforehand before we actually will offer insurance, things like that. But but even that aside, only about 40% of the customers on our platform at any time are procuring insurance from us.
00:27:20
Speaker
They love to have, ah you know, one, we offer a lot of free services and tools to understand your bill, to do some of the basic stuff around forecasting and commitment planning. We'll actually automate the full native commitment lifecycle management for customers entirely for free. um You know, and a lot of customers will use us for that. ah Really, our goal is to put these quotes for, you know, if and when insurance makes sense for both parties, for us and them, in front of the customers in in a timely manner, like in the right way. And so as a function of that, you know, we see a good amount of uninsured committed spend, even from customers buying insurance from us. But we see a lot of customers that that don't have that spend that are, you know, maybe waiting just for the right quote from us to go and buy it at some point. All right. That's the hope.
00:28:06
Speaker
this ah This might be putting you on the spot, but you know cloud bills are notoriously weird. is What is the gnarliest billing edge case that you can think of you know keeping anonymous? but like What's the weirdest thing you've bumped into in your experience?
00:28:20
Speaker
Yeah, without talking about any specific customers, ah once you start getting to the scale of like 80, $90 million dollars a year with a single provider, there are crazy things that come up. like First of all, you have to upgrade your database just to ingest the bill and parse it. But but two, you'll actually see that on you know every bill or every other bill, depending on the scale, the provider themselves will often have billing errors in putting it together, especially because you have custom pricing and other things you know at that scale.
00:28:49
Speaker
I think the other thing, you know, especially when you look at things like the Azure ecosystem and how they've changed their contracting structure from like enterprise agreements now to max is there's so much being lost in translation there, especially for the big scaled accounts that um the bills themselves are often not treatable as a source of truth. And I think once that starts to break, a lot of things downstream of that, you know, ah start to break. so So those have been kind of the class of problems that I think have been the most gnarly and interesting when you can't trust the bill itself coming from the provider. Wow.
00:29:20
Speaker
Okay. So for folks that are listening, what is the right size when they think, you know, I think we're big enough. We need cloud insurance. We're a startup. we' we're We're selling this much or we're spending that much. Maybe I, Scott Hanselman, don't need it because I'm little, or maybe I do. What's the right size for someone to reach out?
00:29:36
Speaker
Yeah, well, frankly, we work with customers that you know range from a two person startup on credits where they're making their first commitment ever to like a database that they want to save like 50 bucks on. And again, it's a very lightweight, there's no infrastructure changes, five minutes to share your bill and get a quote. um so So again, we work with like small YC companies like that. But generally what we say is when you're around, you know, 100 to $200,000 a year of spend in the cloud, that's when the numbers start to become very meaningful. Like we're talking about, you know, funding an initiative, like a whole new AI initiative, or potentially we've been talking about funding a headcount once we get to the 500k plus a year ah level. And so, you know, I would say that really people use us as an individual developer tool, small startups use us, especially now with with just being required to get a commitment to sign a GPU. If there's two guys in a garage and want a GPU, they'll often use us for that. um But but the meaningful numbers start to come at that to 500K scale in terms of to a business, to an enterprise. That's that's that's more accessible than I realized. I mean, a startup and a couple of folks in a garage really could potentially save some money on that.
00:30:45
Speaker
Yeah, and it's often meaningful to them. Like we have a guy who runs a side project that, you know, has scaled up to about 20K per month, but it was running at $200 a month. And they bought some cloud insurance, ended up saving 75 bucks a month. It was meaningful for him, you know, running off of his credit card.
00:31:01
Speaker
Fantastic. All right. Well, folks can learn more at archera.ai. That's A-R-C-H-E-R-A dot A-I. And I have been chatting with Aran Khanna, co-founder and CEO of Archera. Thanks for hanging out with me today.
00:31:15
Speaker
Thanks so much for having me, Scott. It was great to chat. This has been another episode of Hansel Minutes, and we'll see you again next week.