深入 Anthropic 的1000亿美元 Al 算力承诺｜CFO Krishna Rao

Every time we have a new model, there's a set of capabilities that are different. People tend to think about model intelligence as IQ. We think about it somewhat differently. Intelligence for us is multidimensional. It's not just a score; it's the real-world capability of the model.

Each model generation gives you the chance to do more with it, to do it better, and to do it more efficiently, because we think the returns to frontier intelligence are extremely high. And they're extremely high, especially in enterprise. That's a core thesis of our business.

Krishna, I have been so excited for this conversation because you get to see from the inside one of the most interesting businesses in world history at maybe the most interesting time in world history, at least if you're a technologist or care about technology.

One of the things that fascinates me most, just to dive right into something that I think we're both quite passionate about, is this question of compute that you have to deal with all day, every day. It's a key part of what you do. It's a key part of what these companies are doing, and there's just this whole revolution happening.

I'd love you to start by explaining what it's like to have to deal with that. I understand that at one point you were having a daily meeting about how to allocate compute, and to whom and why. Just bring us into that part of your life, because I think it's right at the cutting edge of what's going on.

Look, the compute that we procure is the lifeblood of our business. It is the most important thing in the company. It is the canvas on which everything else gets built. The decisions we make about how much compute to buy are some of the most consequential and hardest decisions to make in the entire company.

Think of it this way: If you buy too much compute, you go out of business. If you buy too little compute, you can't serve your customers, and you're not at the frontier. It's the same thing. We talk a lot about this cone of uncertainty, but the idea is that these purchases have real-world implications. You can't just go out and buy a gigawatt of compute and have it delivered next week. You have to really think ahead and plan for this.

We take a very disciplined approach to how we think about it. We look bottom-up. We model what we think demand will be. Obviously, we sometimes get that wrong. We think about the compute we need to stay at the frontier, and we really look ahead and try to estimate that.

1. The Compute Canvas

As we go out and actually do these deals to procure compute, flexibility is really important to us, so we build that flexibility into the deals themselves. We build that flexibility into how we use the compute as well, because the way in which we bridge from where we are today to where we want to go, when the business is growing exponentially, is to use that compute as efficiently as possible. I would say I spend 30 or 40 percent of my time on compute, even today.

What does flexibility mean in that example?

It means a couple of different things. Number 1, we use 3 different chip platforms. We are customers of Amazon's Trainium chips, Google's TPUs, and NVIDIA's GPUs.

We use these chips fungibly. If you think about the compute we buy, we're using it for model development. We're using it internally to speed up our own product and model development, and then we're also using it, obviously, to serve customers. Across those 3 chip platforms, we're using compute for all of those internal and external uses.

That flexibility actually took us a long time to achieve. We've invested in it over multiple years to be what I believe are the most efficient users of compute among any of the frontier labs. That's not something that just happened overnight.

When we started using TPUs, I think it may have been the third-generation TPUs that were the first ones we used at scale. People thought, "Oh, well, you're crazy. Everyone's using GPUs. Why aren't you using GPUs?" We've invested very heavily to be able to use that compute incredibly flexibly.

Then we look across the different generations of those chip platforms and use each generation for the best workload internally. We really built this orchestration layer that gives us the flexibility to use all different types of compute. In doing so, we're also able to get the most value out of it.

Am I thinking about this in the right way? Something like CUDA has been a part of NVIDIA's story for a long time now. It allows you to do a lot with the underlying hardware, and you want to work your way toward being as close to the bare metal as possible. That's part of this flexibility and being able to control as many of the variables as you can. Is that the journey that you've been on?

That's part of the journey, for sure, but it's also been pretty collaborative. We work really closely with the Annapurna Labs team at Amazon to help influence the roadmap of these chips because we believe what we're doing is really stressing the limits of what these chips are capable of.

That means that a dollar of compute inside our organization goes further than, I think, it does anywhere else. Importantly, we want to utilize each chip for its best purpose within the company.

That does mean that we're building our own compilers. We're really building things from the chip level up in order to have the customization and flexibility to use it internally in the way we think will generate the most ROI.

Can you explain this cone of uncertainty thing? I want to ask about all the component parts of this, but that feels like a really key starting point or overall frame for how to think about both sourcing and the uses of compute. Can you explain what that concept is?

Sure. When you're building and growing a business exponentially, really small movements in monthly or weekly growth rates result in compounding, very different outcomes. As we're thinking ahead, even with our revenue growth, it's really hard to predict this business.

Humans mostly think linearly and incrementally. I've been at the company for 2 years, and that's a paradigm I've had to break for myself: to stop just thinking linearly and think exponentially. When you're in this exponential environment, the range of outcomes starts to be really wide.

We look at a range of scenarios, and we look at different points in that cone of uncertainty over a 1- to 2-year period. Then we work backward from that. What we want to do is be in a place where we can obviously still be at the frontier—that's the most important thing—to be able to serve customers, and then to have enough internal compute to accelerate our employees.

It's interesting: If we were to say to our employees, "You can't use our models anymore," we could serve billions of dollars of revenue with the compute that we allocate to employees internally. But we want to take a long-term view and a long-term perspective on that cone of uncertainty because we want to range toward the top end of these outcomes. We have to plan for that.

2. The "Cone of Uncertainty" in AI Growth

As we go, that's how we think about buying compute in a disciplined way. The most important thing is what happens if you're at one point in the cone of uncertainty, but you've only bought compute for a different point. That's where this compute efficiency has really helped us out.

Can you bring us into the room for the conversations around the tradeoffs between those 3 buckets—training, research, internal use broadly speaking, and then serving customer demand? Naively, you might think it's a third, a third, a third allocation or something. How much does that range around? What are the tradeoffs? What is that discussion like on an ongoing basis?

In addition to meeting about compute procurement, we meet a lot about compute allocation. What's important is that it starts with a culture that's incredibly collaborative, and that informs how this conversation happens. There aren't fiefdoms. It's done in a very collaborative, non-zero-sum way.

There's a level of compute for development that we will not go below. Even if it means it's harder to serve customers or we have to do unnatural things when it comes to that, we want to continue to make that long-term investment in developing the best models because we think the returns to frontier intelligence are extremely high. And they're extremely high, especially in enterprise. That puts a floor on the compute that's allocated to model development.

As we think about the internal use of compute, it really helps us speed up that model development and speed up finding those compute-efficiency multipliers that get us more from each dollar of compute.

When we're talking about it, each team is representing what they would do with that compute. Then we have a very open and frank discussion about how we think about ROI. Because we can allocate that compute so dynamically, we can make changes and adjustments on a relatively short time horizon.

The efficiency thing is so interesting to me. I'm curious if you have a sense of how much more efficient you are versus your own internal benchmarks from a year ago, or versus others that you have some sense of how efficient they are.

How do you measure what efficiency means?

There are a couple of different ways I would think about it. From a model perspective, I think the analogy people have when these new models come out is that they're kind of like cars. You had a sedan before, and then you might have the higher-end version of that sedan, and you're moving up the chain. I think that is true in terms of model intelligence.

The place that analogy breaks down a little bit is that people think, okay, I'm going from the sedan to the sports car, and I get much less fuel efficiency, right? I'm not going to buy the sports car for the gas mileage. In our case, we actually see both improvements—huge improvements in capability, but also in model efficiency. If you look at going from Claude Opus 4 to 4.5, 4.6, and now 4.7, each one of those leaps—they're not equal—has a multiplier in terms of how much more efficient it is at processing tokens effectively.

That doesn't just serve customers. It also helps us internally. If we're doing reinforcement learning on the model, it's basically inference within a sandbox with a reward function, right? If the model is better at efficient inference, that reinforcement learning is more efficient as well. We're able to do this kind of win-win where the customer is getting more capability when we release a new model, and then we're able to serve that model sometimes at a multiple more efficiently than the prior generation.

In between generations, we're dynamically deploying efficiency improvements between these more step-function model changes. It is always getting more efficient over time. What fuels that is the research team. All these things are very connected: the various tasks and workloads that we have internally all fit together in this way of doing R&D for model capabilities, compute efficiency, and serving customers, and then having internal workloads that can be sped up by using the best models—sometimes models that we haven't released.

You said something really important before, which is that the returns to being at the frontier are really high. Can you explain that in as much detail as you can? It sounds obvious when you say it, but there have certainly been some camps that are like, “Oh, I'll just use the 6-month-old model. It's a fraction of the cost, and I'll just use that, and that'll be catching up all the time.” That just hasn't been the case.

The second Opus 4.7 comes out, even I, as a consumer, switch it on right away. Or GPT-5.5 comes out, and you switch on the new one right away. I want the best. So, talk about the returns to being on the frontier and why they're so high.

I think it's a couple of things. Every time we have a new model, there's a set of capabilities that are different. People tend to think about model intelligence as IQ. It's a single number: okay, this model was at 110, and then it goes to 125. We think of it differently. Intelligence for us is multidimensional. It's not just a score.

In fact, we find that, yes, everyone publishes their model benchmark cards, but a lot of those benchmarks are saturated. We publish them, too, but our measurement is what the customers tell us: what is the real-world capability of this model? As we've released better and better models, what we've seen is that it's not just the outright intelligence. It's also the ability to do long-horizon tasks, the ability to use tools or computer use, and the ability to do things for agentic tasks that have specific value even faster, right?

If you have 2 employees and they're maybe both equally capable, but someone takes a week to do an assignment and someone does it in a day, that second person, if they continue to do that, can be 7 times better, right? They might be equally capable at something, but maybe they just take longer to do it. All of those factors then determine how customers experience it.

What we found very consistently is that by releasing new models, the TAM is unlocked in a unique way. More TAM gets unlocked, and more use cases are possible. A good illustration of that is the last 4 months that we've had at the company. We started the year with about $9 billion of run-rate revenue, and we ended the quarter with north of $30 billion of run-rate revenue. That kind of change is really enabled by these model-intelligence leaps and then the products that we build around them.

That's what I mean by the returns to frontier intelligence being really high. I think that's unique to enterprise, because in consumer, sometimes you don't see that as readily. You don't see as readily that consumers are really pushing the limits of what the models can do, whereas in enterprise, our customers are always now—it started with coding, but it's really expanded beyond that pretty meaningfully. Each model generation gives you the chance to do more with it, to do it better, and to do it more efficiently. Customers see that, and then they invest really heavily in more tokens with the newer models. We've seen that cycle play out again and again. That's a core thesis of our business: especially in enterprise, the returns to frontier intelligence are not slowing down.

The things that push that frontier are like a sci-fi story or something from the books I was reading when I was growing up. It seems as though, in the major labs, we've reached this point—someone on your team said it recently—of recursive self-improvement, where the models themselves are building and doing a lot of the research to do the next generation of improvement.

If I think about the frontier that you're pushing and OpenAI is pushing, and compare that to the open-source models, maybe the gap will widen as a result of you getting there first to this recursive thing. How do you think about that? Tell us how we should think about this idea of recursive self-improvement in the models themselves, because it seems like getting there first is incredibly important. Then you can just continue to separate yourself from those that haven't reached it yet.

I would say we do see progress accelerating. We see—I can't speak for other companies—but for us, the scaling laws are alive and well, and we're seeing that even with more recent releases like Mythos. Right now, within the company, 90-plus percent of our code is actually written by Claude Code. A lot of Claude Code's code is written by Claude Code.

3. Recursive Self-Improvement

Why do we allocate compute internally? Why would we forego revenue for it? It's because the models themselves are helping us to build that next generation of models. In addition to this capability leap that you would have just from the scaling laws, talent is really important, and that talent with the best models can really accelerate the development of capabilities. We're really seeing that.

We don't really think about models as closed or open. We think of them as frontier or not. The ones that are at the frontier are clearly capturing this economic value and driving meaningful ROI for customers. We're investing behind that thesis, and that means both compute and talent to use that compute and use our own models to really accelerate the development.

The other piece of it, too, is that it's not just the models; it's the products that get built on top of them. We had 30 different product and feature releases in January. The pace of that has accelerated as well, and that's enabled in part by utilizing the models with the talent that we have to really accelerate ways to access this underlying intelligence.

That's kind of our theory of the case on the product side.

How do you think about this weird world where you mentioned the talent and the leverage, and they're not writing code themselves, and Claude Code's writing its own code? It seems like the last step of that would be that you don't even need the talent to tell the thing what to do. It just figures out what to do on its own, and that's the ultimate version—the thing then just runs and is only constrained by compute or something. Am I being too crazy about that, or is that future possible, do you think?

I think that the core of our company is still a research lab. I think it's maybe not as well understood—maybe it's getting more understood from the outside—but we're doing experiments. We are doing things that push the limits of what our models can do. That research and that engine are upstream of everything else that we've talked about.

4. Raising Capital & Overcoming Investor Skepticism

That is enabled by the models today. It's not entirely done by the models. Over time, we think the models will get better and be more helpful in that process, but having the best talent to set the direction—not just the priorities, but some of the new areas of discovery—actually makes that research talent even better, right? I think of it as accentuating and accelerating the talent that we already have.

We talk a lot about how talent density beats talent mass, and I think that's true here. We want the densest collection of AI research talent and inference engineering talent. That, enabled with the best models, is a really winning combination.

How are scaling laws talked about internally? The consensus has been that you've got different components of them: pre-training, post-training, and reasoning, and that all of these are moving at different paces. To hit a true wall, they would all need to fall down. That's sort of how the world is conceptualizing scaling laws now. How are they talked about internally? How do you think about them?

We look at models at various points in their development. During a pre-training run, we can see how this model compares to a prior model that we did on these loss curves, and that gives us a sense of model capability. You can do the same thing as you think about RL.

Then, probably as importantly, when customers get their hands on it, what are they seeing? Where are they identifying pain points? Those pain points then become training targets for us, right? We don't train on customer data on the enterprise side. On the prosumer side, it's only if you opt in.

5. Scaling Laws

But customers tell us things like, "Hey, I wish the model were better at this," or, "I had this particular place where it got stuck. I could build this other product, but the capability needs to be further than that." What we usually tell them is, "Okay, build your product for that, because we're going to, on the R&D side, improve that over time." There is this connected loop.

Internally, we're always looking at different models that are being trained and different snapshots that we have, and comparing them internally and, to a lesser extent, externally against our own measures and, ultimately, how our customers view them as well.

It feels like there's just no slowdown in the scaling laws themselves. Is that a fair characterization?

For us, that's a fair characterization. Yeah. Obviously, a bunch of the authors of the scaling laws papers are among our founders. Notwithstanding that, we can be a skeptical bunch. We hold ourselves to a really high standard. Again, it's this idea of a research lab that's very much about the scientific method, and people are constantly challenging previously held assumptions.

But from what we see, the scaling laws are not slowing down.

So if that's true, you said before that it's hard for humans to be exponential in their thinking and not linear. If that continues to be true for however many more turns of the crank here, how do you do that thing of not thinking linearly and thinking exponentially yourself in your job and for the business? The implications are really hard to reason through. Exponential growth rate is one thing, but exponential growth of capability—I don't even know how to get my head around it. How do you get your head around it?

We think about the world as scenarios. It's very hard to have a point estimate in this business, so we have a very low bar for updating our current priors or our current perspective, because it could be the case that something that was true a month ago is just not true today, and that breaks your model.

This old idea of, "Well, we'll forecast once a quarter and revisit this in 3 months at the next board meeting"—that doesn't work for a business. It's so dynamic that we always have to think, "The models couldn't do this before, and they can do this now. What does that mean for the TAM?"

We've seen this in coding first, right? Starting around Sonnet 3.5 and 3.6, we started to see this really remarkable jump in capability, which was then followed by adoption, usage, and revenue. It was a little hard to predict that, but now we can use coding as an analog for a lot of what's happening elsewhere in the economy and elsewhere in our business. We look at pattern recognition in our own business to try to predict what's going to happen in the future.

Literally 15 minutes before you got here, the news came out about your partnership with xAI and the Tennessee facility. It makes me curious about how you're canvassing the world for opportunities like that. That's an opportunity you decided to pursue, and I'm sure there's a universe of things that you've explored. What is the strategy for trying to get more compute in creative ways? Bring us a little bit more into that.

We announced a partnership with xAI for their Colossus facility in Memphis. We're really excited about that. It's going to allow us to continue to expand, especially on the consumer and prosumer side. But that's just one example of us, as you said, looking for near-term compute wherever we can get it.

6. Sourcing $100 Billion in Compute

As the compute base grows, that near-term compute becomes a smaller and smaller fraction of what's available and what's out there. But we look at it as, can we deploy that compute that's available productively? Sometimes the answer is yes and sometimes it's no. If we can, then we look at the economic return on it based on what it's priced at, what duration we have it for, where it's located, what type of compute it is, and how efficiently we can run it.

We have a process to assess that. That same process, by the way, we use to assess longer-term deals as well. Last month, we signed a 5-gigawatt deal with Google and Broadcom for TPUs, starting in 2027. We also signed a deal with Amazon for Trainium for up to 5 gigawatts as well.

It was an over-$100-billion commitment, and a lot of that compute is already landing and will land in the rest of this year and into next year. If you think about it, it's a bit of this layer cake of compute that's starting at different times with different capabilities, and we're dynamically comparing that compute—its price-performance over time, when it lands, and what we think we can do with it internally in the business.

There are so many different variables you have to optimize for around what compute it is, at what cost, and over what time horizon. But we have a pretty dynamic way of looking at near-term compute and then medium- to long-term compute. The things we're assessing are largely the same. What is different is just the time horizon.

What about the tradeoffs of price-performance, the tradeoff between cost per token or something, throughput, and speed? From the customer perspective, they care about both. Speed unlocks some capabilities and use cases that are really interesting, that we don't know about yet as these things get faster. Can you talk a little bit about that tradeoff in compute as you're assessing it?

As we look across 3 different chip platforms, we also have multiple generations of chips within them, right? It could be TPU V5e, V6, and V7, and Trainium 2 and Trainium 3. All of them are at different places on the price-performance curve.

We also importantly look at how we will utilize them. Price-performance is important because of efficiency. Speed is also important for certain use cases. We look at the compute down to a very granular level in terms of what it can deliver for us and when.

Our compute team leads that, but we closely collaborate across the business to say, "Where do we need this compute and for what?" We might need CPUs for RL. We might need more leading-edge compute, and we're going to deploy it for our best and fastest models or for training them.

From our perspective, it's customer demand, but it's also really quite granular in terms of what each chip is best for and then what we will have when.

I'm always so curious about the metabolism of Anthropic, in this case, for new compute.

How fast could you take it if I air-dropped on you twice the compute that you have tomorrow? Would you consume that, and how fast would you consume it? If I air-dropped 10 times the compute on top of you, how fast would you consume it? Can you calibrate us on that sort of thing?

Is demand actually—it feels like demand's unlimited. Between these 3 uses—training, internal, and customer demand—everyone is saying the same thing: shortages everywhere, memory stocks mooning. Is it that extreme that if you 2x, 5x, or 10x the amount available to you tomorrow, you would just more or less instantly consume it?

This goes back to how we use it and the fungibility of it. The answer is, we're constrained across those use cases internally today. I would say that a year or 2 ago, it would have been harder to consume—especially a heterogeneous kind of compute drop in your example—really quickly because these chip platforms are different. They are different: some are harder to operate, and some have idiosyncrasies in terms of how we use them.

I would say today that getting a bunch more compute, I think it would be deployed very rapidly across those different use cases. We probably have the same kind of allocation or calibration that we do with compute today. But it's become a lot easier for us to spin up very quickly and deploy almost any type of compute. That's something we think is a real advantage.

Back to the ways customers are using Anthropic. One of the interesting tensions and trade-offs that I'm fascinated to hear how you think through is between the platform approach, where I build my business on top of Claude and it powers my thing, versus you doing the thing that I wanted to build. This is the classic Claude Code versus Figma, or something like this.

7. Platform vs. Application Strategy

How do you think about the right balance of how deep into the application layer you should go versus just being a pure enabling layer of, "We're going to provide the reasoning engine and the intelligence, and the world, go forth and build whatever you want—pass through the API or whatever"? That seems like a fascinating internal discussion and tension, to some degree.

The way I would think about it is that most of what we're building is a platform. We think there are so many examples of where a platform can accrue a lot of value, but the customers who are building on that platform actually create even more value. We think that's what we're setting up for today. It's maybe akin to the early days of AWS, right?

If you think about the Claude platform and all the tools and services that are now built into it, because it's not just the raw model access—it is prompt caching, the ability to use virtual machines, Claude Code being called within there, Dispatch, or the Claude Agent SDK for managed agents—all of these are, I think, effectively vectors to access that model intelligence for other companies to build into their own products.

That's most of what we're focused on, and really most of where we think the business is going and will end up from where we are today. That said, we will also build our own applications on that same platform where a couple of things are true.

Number 1, if we feel like we have a vision into where the models are going and we can demonstrate that and create customer value in it, that might be something like Claude Code, right? We were able to say, "Actually, a lot of what's out there in the market was developer-led. Claude Code is a platform that's Claude-led, and we thought the models couldn't quite do that when it was launched a little over a year ago, but we thought they'd get there—and they have." So, one is building ahead to model capabilities.

The second is thinking about ways to demonstrate value for the ecosystem that others might emulate, right? If you think about Claude for Financial Services, Claude for Life Sciences, or even something like Claude for Security, these are ways in which we've composed the platform. Again, we're building on the same platform as our customers, and we think that creates a level playing field.

We also think that there's so much value that's going to accrue in some of these areas that our customers can win and we can win as well. That's why you've seen, as we've launched some of these products, we've done them in a collaborative, partnership-oriented way. Whether that be on the security side, design, or financial services, we've partnered with the ecosystem.

I think of our strategy as mostly horizontal. We'll build vertical where we think we have some value to add, a perspective that's useful, or a way to demonstrate to the market how we think about our platform adding value. A lot of the value is going to accrue to the customers that are building on top of it. Our goal is to build the best models, and then build the products, tools, and services that allow that intelligence to proliferate within customers.

How much do you care that it's just a reality that people are scared of you? There's a sense that because you control the most essential piece of these new applications—the underlying intelligence, the reasoning engine—that it may totally be true, and maybe already is true, that more of the value is accruing on top of the Anthropic platform than is being captured by it.

But nonetheless, it's still scary to imagine. I guess maybe you could say something similar about cloud and AWS or something like that. How much do you think and care about the fact that some of your would-be customers or existing customers are, in fact, scared of you as a competitor?

Part of what is hard in this business is that it's changing so quickly. The model capabilities sometimes even surprise us. When we release models or products on top of that, there's an element of what's happened in prior waves over the course of 5 years, 10 years, or 20 years happening in months now.

When we release things, people are also surprised by them, in some ways in the same way that we were surprised by them. But fundamentally, what we are trying to do is be very partner-oriented toward the ecosystem.

Yeah.

That means that we have early-access programs, we work very closely with customers, and we listen to them about what capabilities they want. That doesn't mean that the things we release aren't sometimes moments where you're like, "Wow, that's way more powerful than I thought it would be," or, "I didn't realize the models would be able to do that this quickly."

I think that part is a reality of where we are in this cycle and in this development of intelligence. But part of it is also that we want to make those capabilities really accessible, and that should accrue a lot of value to customers as well.

Customers that are front-footed on that and adopt—and, frankly, also those that are building and using the tools that we offer on our platform—we think we can actually accelerate them. I think some of it is a reality of frontier-model development, but our approach to it is probably a little different and more partner-oriented.

8. Pricing Dynamics

You said before, going from 9 to 30 in the first quarter—the pace is so insane—which makes me wonder about pricing. The dynamic of how to price tokens, or use of the system, is so fascinating to me because I think a lot of people a year ago would have said price is going to constantly fall, but actually what's happening is it's going up in many cases.

This is true at different levels, whether it be the Mythos pricing, which is quite high because it's so powerful, or the cost of an H100—the rental price of an H100 looks like a smile curve. I'm very curious: if everyone is compute-constrained, why doesn't everyone just raise prices a lot to try to find the right equilibrium?

I don't know if you could just riff on pricing—how you think about it, what the trade-offs are, and why not raise prices a lot.

Anthropic is only a little over 5 years old. This past March was the 3rd anniversary of the first dollar of revenue into the business. We only had a frontier model, for real, for the first time in March of 2024. So, the time scale of these things is an important backdrop.

Our pricing has been relatively stable across Haiku, Sonnet, and Opus. Now Mythos is obviously newer. But we've made very few pricing changes.

The biggest pricing change we made was to bring down the price of the Opus family when we launched Opus 4.5. If we think about why we did that, it's really because we found that Opus-class models were underutilized relative to their capability, right? People were often trying to fit an Opus problem into a Sonnet workload.

Because of the efficiency improvements that we were able to make, we were able to serve that very efficiently from our perspective, but actually bring down the price, which made it more accessible to customers. It goes back a little bit to the fact that we want our customers to generate a lot of value from it.

They're generating a ton of ROI from our models today. We want that to just continue because our goal is to proliferate this throughout the ecosystem. We think we're in the very, very early innings on all of these use cases.

The best way to do that is to get this intelligence in the hands of as many businesses as possible, from startups and digitally native businesses to the largest companies in the world.

Some of that means that you have to make it at a price point that's accessible and allows them to get a lot of value from it. Changing the pricing for Opus actually demonstrated the Jevons paradox. We lowered the price of it, but consumption went up way, way more than what you would have expected.

Because we hit that sweet spot for customers, they were able to use it a lot more. We had the efficiency to be able to serve it to customers at scale. Then they were able to build that into their workloads, such that when we released Opus 4.6, it was a model improvement they could slot in. We didn't change the price, and so we think pricing stability is important. We also think that pricing to get that value and to see that kind of Jevons paradox happen is really important.

The other component of this is margins and how you think about margins as a business. Again, because this is so unbelievably capital-intensive to build these frontier labs, you've got the levers we talked about: efficiency and price. Both those things relate to margin.

Apologies if it's a naive perspective, but given how much capital you need, why not just say, “We want to have a healthy margin,” and set the price accordingly? Maybe that price can come down if efficiency is better or whatever. I'm curious how you think about margins as they relate to pricing in the business.

I would say we think about what the return on our compute spend is, writ large. That is all of the different workloads that we've talked about, whether it's serving customers or model development. If you think of all of those as being in support of revenue over different time scales, if I serve inference, it's in support of revenue today. If I do model development, it might help with a capability that unlocks TAM and drives revenue 6 months from now. Everything in between—if I do internal acceleration to launch a new product, all of these things are in support of that.

I will say our returns on that compute expense today are robust. They're robust, and we think of it as the return on that full envelope of compute. We feel really good about where we are from that perspective, and we're balancing delivering value to customers with also seeing a really, really strong return on that compute ourselves.

If you think about when revenue grows, as we mentioned in Q1, it's not like we onboarded a bunch of new compute in that time period. We talked about how compute comes based on a ramp that might have been determined 12 months ago. This idea of a variable cost that's incremental to serving a customer doesn't really fit our business. It tries to fit our business into a software paradigm, but that's not the case.

In actuality, compute is supporting all of these activities, and we're really generating a robust return on that compute. That's our measuring stick. I think it's something where you think of the compute envelope that we have as the thing that's able to govern how much we're able to drive revenue, both over the short term and the long term.

If you're this great customer of the compute providers, what does that group need to do to be a great provider to you and help you drive that return?

We're fortunate that we have really great partners in Amazon, Google, and Microsoft, but also with Broadcom and NVIDIA. Our ecosystem is really strong. We are the only model that's on all 3 clouds today. We're the only language model lab that's using all 3 of these chip platforms.

9. How Anthropic’s Finance Team Uses Claude

These collaborations are much deeper than just procurement. I think that's something that's often lost. If you think about our relationship with Amazon, our teams are deeply embedded with the Anthropic Labs team. We're really good users of training. We've spent a lot of time and energy working closely with the team, and internally we plan capacity together.

If you think about the 3 clouds, they're great distribution engines for us, too. We have a really robust first-party business as well. These are multifaceted partnerships, whether it be developing the chips themselves, landing that capacity, serving it, and then ultimately distributing it to customers.

I'm thinking about your function, like the finance team, and the ways that you might use this ROI-on-compute idea on different horizons with all these complex variables. That makes me wonder: How do you use these powerful tools yourself internally to run your group and the business? What is the deployment of Claude Code and Claude in general on the finance team at Anthropic?

This is really interesting because we were using Claude Code about a year ago. I started asking people, “Is everyone just live coding?” We started to use Claude Code almost like an assistant, a digital coworker, not just for coding tasks. That actually was early in what eventually became Co-work. That was an extension of Claude Code to say that what it's done for agentic software development, it should do for all knowledge work.

We started to productionize that, and I'm actually really proud of it. We spend a lot of time with our product team, too. They see how we use it and get input and feedback from us. Today, all of our legal entities can produce their statutory financial statements using Claude. The human checks them, but all of those financial statements are produced with Claude.

We also have a more real-time platform called Ant Stats. It used to take a lot of time to sift through all the data, get to the conclusions, write a memo about it, or publish a regular report on what's happening over the course of the day and what's driving it. We now have a library of skills for Claude that are specific to finance. The last time I checked, there were 70 of them that everyone could access through this common repository.

On top of that, we built an MFR, a monthly financial review skill. It can produce our monthly financial review, and it's 90% to 95% ready. Then all of the discussion becomes about what we do and what the implications are, not exactly what happened, because Claude isn't just reporting the weather. It's also helping to think about the drivers and why the number changed in the way it did.

That gives you tremendous insight into the business, both in terms of this MFR that we do and also on a daily basis. What used to take hours to produce—a weekly report on what's driving revenue or what's driving our compute utilization—now takes 30 minutes. Then we can spend our time on the actual strategic implications of the business. We can also get it in the hands of business leaders much more quickly, so it's just meant that the insight engine is a lot faster within the company.

We also have a dashboard. I look at token usage across—

A leaderboard?

Yeah, we don't compensate people based on it. No one's trying to token-max for that, but it's really interesting because some of the most senior people within the finance team are actually the biggest users of tokens.

It's not just the 22-year-old who joined, has a coding background, was doing that on the weekends, and brought it to work. It's also people using the tools to change how they're working. I think our number-one user is our head of tax, and he's really focused on tax policy engines and automating large parts of the workloads that are happening within the team.

I love seeing that, and I tell people, “If we're not superusers of this, if we're not pushing the limits of it, how can you expect customers to do that?”

Just as a human, does it freak you out at all that all of these—I’ve heard so many examples like this—it starts to feel like we just start doing the stuff that AI tells us to do? In the sales example or the calendar or whatever, maybe that's great. Maybe it's just such a better coordinator and wide thinker and optimizer than we ever could be that we should do what it tells us to do.

But it feels ever so slightly dystopian to me that this reality is coming quickly. I’ve had examples of it, too. It feels kind of cool—like it’s helpful—but at the same time, if I really close my eyes, I’m like, “Oh, I’m just doing what it tells me versus me telling it what to do.” It’s a really interesting human dynamic, and I’m curious for your take on it.

I maybe have a slightly different view on it, in that I think we’ve been able to hire great people at the company, but it has made even those incredibly talented people so much more productive. There’s a little bit of this—I think of it again like Jevons paradox, but for labor—which is that we have people who become incredibly more productive.

We’ve actually hired a lot more people because of that, because there’s no shortage of work to do. And now, with the assistance of Claude, people are spending less time in that MFR trying to reconcile some number, but they’re actually thinking, “How do we reinvest this in the business? How do we think about dynamically allocating resources?” Whereas before, I’m working to tie out a number or, in that accounting example, taking a long time to close the books.

So I actually think of it maybe even more optimistically: it is an accelerant to our productivity. That actually means that we can get a lot more done, and even as we grow the team, those people are more productive as they come up the curve on how to use Claude within our company. I think that’s starting to be true across many companies as well.

I’d love to talk about investors and capital formation. Of course, you’ve had to raise tons and tons of capital. At the same time, it seems as though, if I just squint my eyes and think about the multiple on current revenue, it’s not that crazy in terms of where you’re raising money.

I’m so curious for you to teach us about what it’s been like to interact with investors and how you’ve seen their understanding of the company evolve and mature. Where do you think investors, as a group, understand it now? Where are their misunderstandings about Anthropic? Tell us that side of your life.

I joined the company about 2 years ago. We were closing our Series D at the time. That was not a straightforward fundraising. The company really only had a frontier model in the middle of that fundraising, and toward the tail end of it, the FTX transaction was happening, which was liquidating a bunch of Anthropic shares.

10. Why the Returns to Frontier Intelligence Are So High

That was the starting point. At that point, the questions were around, “Why do you need to have a frontier model? What are the returns to this?” They were also around our mission and how we approach things. People said, “Hey, aren’t AI safety and building a really big business at odds?” There were also a lot of other misconceptions: “Your sales force is really small. Don’t you need to scale it like all these enterprise software companies?”

There was just a paradigm around trying to fit us into a particular mold that had existed before. Over time, it’s evolved. At the end of 2024, we raised the Series E. The business had scaled to close to $1 billion of run-rate revenue, but the day of our first close was the day the DeepSeek news came out.

Obviously, we got the close done, but there was certainly a ton of volatility as people then said, “Wait a minute, should I just totally re-underwrite how I think about AI in total?” That was a Series E. We brought on great investors across all of these, but people still had some of those questions.

They looked at our forecast and thought, “Okay, I get it. You’ve grown—wow, you’ve grown to $1 billion of run-rate revenue so quickly—but there’s no way you’re going to be able to keep it up. That’s just not possible, right? There are laws of physics. You’re an enterprise, which is great, but the adoption’s going to take so much longer. Look at how long it took with cloud and how many people are still on-prem.”

The business continued to prove out the thesis that the returns to frontier intelligence are really high. What’s really happened is model-led growth, enabled by products, our go-to-market team, and our distribution.

I think investors also saw that this thesis of, “Hey, it’s really important to build this transformative technology, but to do it in the right way and do it responsibly,” had this really interesting interlink with our business that most people didn’t really understand or believe.

We invest in research not just in model development, but also in AI safety research. We pioneered interpretability, which is—think of it as an MRI for the model—to see inside the neural network and how it works. We pioneered alignment science, which is: you want the model to do what you tell it to, so how often does it do that, and how often does it stray from that?

Those things are important for our mission, and that’s why we did them. But they had these downstream effects. It turns out that if you can look inside the model, you’re better at building them.

The last linkage is that, if you’re selling to enterprises—we now sell to 9 of the Fortune 10—all of those enterprises are entrusting us with customer information and their data. They’re interacting with their employees, sometimes even interacting with their customers as well. Those are the most sensitive workloads.

The more and more these businesses are running on Claude and our Claude platform, the more this investment that we made and will continue to make in safety, interpretability, and alignment accrues to the benefit of enterprise customers as well as all of our customers.

If they’re going to entrust us with all of that access and all that data, and the ability to work in the most sensitive workflows within their company, they want a company that they can trust. That’s not why we invested in it, but it did have this downstream effect that we’ve really seen prove out again and again: we’re a company that is both at the frontier and investing in safety, and that you can trust.

We’ve raised $7.5 billion since I joined the company. We have another $5 billion that will come in the future from the Amazon and Google deals that we closed last month. That’s a tremendous amount of capital, but it’s a capital-intensive business, and we need this capital to support that growth.

It all goes to the fact that the business is running very efficiently. The reason we raised this capital is more because of that kind of uncertainty than it is to fund actual losses in the business today.

What was your own perception of this 10x growth of the business? The first time that happened, did you personally believe that it was possible? Did that seem absurd? Has it continued to feel that way?

Now that it’s becoming consistent, maybe it’s becoming more commonplace to you, but what was your own view, staring at this curve, about the odds of hitting a 10x type of growth so many years in a row?

When I joined the business, it had about $250 million of run-rate revenue, and the plan was to get to $1 billion. I said, “Great. In what year?” That was linear thinking, right?

Consistently, Dario has been a much better predictor of the revenue than I have. I think we’re going to close the gap over time as we get better at forecasting and understanding the business, but definitely, the first time I saw it, you have all these arguments about the laws of physics and the law of large numbers: Where is the revenue coming from? How can it be added this quickly? How can customers move this quickly? Is this even possible in enterprise?

All of those things start to get broken down over time as you see how the business works internally, and you see the adoption curves and the exponentials that are happening. Again, we have the exponential that’s happening on revenue, but underlying it are these many other exponentials that support it. You start to see and believe in that more.

That doesn’t mean we’re not disciplined and thoughtful about the forecast and how we think about the range of scenarios. But it does mean that my thinking has at least shifted a lot more from linear and incremental toward leaning into this exponential and really believing in its potential, and how this is just different from how other businesses have evolved.

As you’ve talked to investors at every stage, every round that you’ve raised, I’m sure there’s something that’s the most common or hardest thing to explain to investors, or that they’re struggling the most to understand and get their heads around. What is that today?

11. Public Perception, Risks, and Government Regulation

I think it is this paradigm of how compute is used. Thinking of it as not just something that is a variable cost over some time period, but really this resource that’s so fungibly utilized. We run workloads on a chip in the morning for inference, and in the afternoon and evening, we use it for model development.

That paradigm doesn’t exist in a software company or a factory, right? You can’t repurpose it if you have a bunch of people doing R&D and that’s your R&D expense.

They can't go and become cogs, right? And vice versa in most traditional companies. Here, you really have that fungibility that's possible, and I think that's where the return on compute is so important. I think people are beginning to understand that, but there's still a tendency toward treating it like, “Oh, I have to separate these 2 costs,” when in actuality, they're very self-reinforcing, and that flexibility is actually what helps to drive revenue short-term and long-term.

If I was to force you out of your role and into an investor seat at a great big investing firm, and then I said your job is to go grill these companies and invest in the best ones, what questions would you be asking of the labs or companies that are building models to really get at the heart of the points of uncertainty, of skepticism, of things that might not make these the best businesses of all time? I'm curious, maybe from that angle, how you would approach it.

So I would say a couple of things. First, what is the ROI on compute, kind of all up? How are you utilizing it, and what return are you seeing today? How is that coming over time? These are the massive, unprecedented investments that companies like us are making. What is the return that you're generating on that, when does it come, and what is the shape of it? So I think that's one.

Second, how do your customers see ROI in what you do? Are people just using this for testing? Are they actually deploying this at meaningful scale? I could say for our business, we're seeing that in spades. Our net dollar retention rate is over 500% on an annualized basis, and with 9 out of the Fortune 10, these are real, real customers making significant buying decisions.

Not pilots anymore.

Exactly. On the way here, I was in an Uber, and I signed 2 double-digit-million-dollar commitments during the car ride, which was 20 minutes. From that perspective, we're seeing it, and we're now being judged by some of the biggest companies in the world—the most sophisticated buyers—and startups that also have choice in the market, and they're choosing us. But I think one question I get a lot, or I would ask from the investor seat—the skeptical investor seat—is, “How are your customers getting a return from this?”

Maybe a third one is, how do you think about compute in the future, and where does it come from? Because obviously, some of the places that we buy compute from sell the compute to others; they may use the compute internally. What is the balance of that over time?

And so your philosophy there is just to be involved with great players and have flexibility.

That's right. That's right.

There's this crazy stat about AI, just the generic concept, being less popular than Congress among the general populace. It's funny when you first hear it, but when you really think about it, you're like, we need to solve this problem. It doesn't seem like the general world—the people who aren't in technology and don't live in the Bay Area or New York—yet feel or understand why this is good for them, as measured by their opinion of it. What do you think we need to do as an industry about that problem?

Look, I think that if we think about the transformation that's happening, there have been other transformative waves before, right? All the way back to the Industrial Revolution, the internet, cloud, et cetera. I think one of the things that's different about AI is that it's all happening so quickly. You can have years or decades of progress compressed into months. Going back to humans thinking in terms of exponentials versus linear, that can be jarring.

12. Mythos

I think we are very optimistic generally about the potential for this technology. I think that we as an industry can continue to do a better job of articulating. Dario wrote this essay, “Machines of Loving Grace.” It's all about the potential for this technology to transform the way that we live, whether that be in drug development and curing diseases that are more mainstream, but also rare diseases, and, number 2, in healthcare and how healthcare is delivered, to raise our standard of living in the developing world and in places where resources are not as plentiful.

I think all of those things are part of the promise and potential of AI. We could probably do a better job of painting that picture, and we want to show more tangible results for that over time. I think that is coming, and that's one of the things I'm most optimistic about.

I think on the other side, though, we do—and this is again cultural to us—we do want to articulate the risks. I don't think we should just tell everyone everything's going to be great, because there are likely to be bumps on the road. People generally gravitate toward more honest and balanced assessments, right? If I feel like somebody's just telling me all the good news and none of the bad news, then I'm like, “Okay, do I really trust this perspective?”

That's where there's a need for balance: to say, “Look, these are some of the things that happen when change is compressed over a short amount of time. How do we work across commercial and government to actually come up with some of the solutions to that?” I think it's about a clear articulation of the opportunities. It's about really thinking about what those solutions may be, and that's not any 1 company that can come up with it.

We don't have the blueprint that's going to solve everything, but we should at least have that dialogue about some of the risks and downsides and what we can do to address them. Then I think it's being transparent about both of those things when we talk about it. Over the long term, the opportunity is going to be significantly higher and greater than some of the risks and downsides that will happen, but that doesn't mean it's going to be perfectly smooth on the curve.

The release of Mythos was such an interesting moment. It was the first time many people—friends of mine who are careful watchers of this stuff—said something like, “This one kind of makes me scared.” So it relates back to the safety question. It's also the first example of you coming out and saying, “We want to make sure this isn't used for bad,” and it's maybe the first one that we are worried could be used for bad.

I'm curious what that discussion was like internally before the world heard about it—the decision-making process around it. And just using that as an example to talk about the things that do scare you as we continue to advance and the scaling laws continue to hold.

Yeah, I think one of the things about Mythos is that people may misconstrue it as just a cyber model. It is an incredibly capable model across many, many different dimensions. What we found was that cyber, in particular, was a place where it spiked. And so this was the first model that we decided to release in a different way.

The way in which we did that, again, is consistent with our mission and our principles. We wanted to do it in that way. We have this phased approach to it because we think that when a model is this capable—and again, cyber is the thing that people focused on, but there are other things as well—we think it can be used in a positive way, right, to patch code bases.

You've seen these examples where we had an open-source code base that a prior model found 22 security vulnerabilities in, and Mythos then found 250. That is kind of scary, right? But that informed the way in which we released it. So we didn't say we're never going to release it. We said, “Let's do it in a phased way. Let's do it with a group that we'll expand over time, where we can focus on this one cyber capability and how it can actually be used positively, in a defensive way as opposed to in an offensive way.” We think that's a template that could be used for the future.

Because of this one particular area, we wanted to be cognizant of that in how we released it.

You're so big now that you run into everything and everyone. One example of this is the government, just a couple of days ago, saying maybe there'd be this new system where you have to preapprove the release of a new model with the government before it was released to the public. Obviously, you had the crazy experience with the Department of War, and I'm really curious what that was like as you went through it.

Now everyone cares about this company and this technology and the couple other companies that are building it. How do you navigate that stuff? Some of it is just beyond your control, but I'm sure you're trying to work with people as best you can. Maybe talk about those 2 examples of how the government is now a very relevant partner, player, overseer, et cetera.

Yeah, so I think first and foremost, we prioritize having a strong relationship on this because we do think that regulation has a role to play in how these models are developed over time. We are very America-first in our approach. We want the technology to support the US as well as democratic countries around the world, and that's one of the reasons why we've been working closely with the administration on something like Mythos.

I do think that there’s a balance, right? You want to be able to have innovation happen really quickly and not have that be slowed down, but you also want to have this kind of responsibility framework for how these things are deployed. We’ve long said that this technology has implications and that we should have an honest conversation about them, and that includes with the government. I think the Mythos process is a good example of that.

Can you teach us a bit more about the culture—how you would describe the cultural tenets to your parents or something like this?

What feels like it really drives most of the culture? I’m especially curious about the writing. You hear often that Dario publishes these long essays externally every so often. My understanding is he does that much more frequently, and there’s a lot of writing culture internally. I’m trying to get a sense of what the culture is like to be in, what makes it most distinctive from other companies that you’ve worked at, or from other companies that are trying to do the same thing, and your sense of the differences and the distinctiveness.

Yeah. The culture is a really unique aspect of Anthropic. It is something that we do talk about externally, but it’s different when you’re in there living it. Maybe I can tell you a little bit about some of my observations. First of all, we have 7 co-founders, right? That shouldn’t work on paper, but it really does in practice. I think they’ve really set the example for the culture and the things that really matter to the company.

We do a culture interview, and it’s not some pro forma thing we do just to check a box. It is a real part of the evaluation process. Somebody could be passing with flying colors on everything else and be the smartest person you’ve met for this role, but we won’t hire them if they don’t pass the culture bar. The way I would describe it—I like that frame of how you would describe it to your parents—is that it’s incredibly collaborative.

This means that we don’t really tolerate fiefdoms, sharp elbows, or the attitude of, “I need to take credit for this.” It’s incredibly humble. Our competitors are incredibly capable, and success is far from guaranteed. I think that’s really part of how the company operates. If we reach a milestone and something good happens, there’s not confetti on the floor. It’s, “What’s next?” I think it’s just that focus on the mission and the alignment that’s imbued throughout the culture of the company.

The other thing I would say is that there’s rigorous debate, right? There’s an intellectual openness and intellectual honesty where people question things. People will really express a point of view, but then there’s productive dialogue around it. After that, we’ll decide on a path forward, and then there’s real alignment. In something like compute allocation, which we were talking about before, people might have different perspectives on how to allocate that compute, but they’ll engage in a thoughtful discussion about where the returns are highest or best. When we come to a decision, there’s alignment on it. There’s no second-guessing, politics, or fiefdoms.

The other piece of it is that the culture is remarkably transparent. Dario gets up in front of the company every 2 weeks, usually writes a short document, and talks about 3 or 4 topics. Then he takes open questions from the company. These are not softballs or planted questions; they’re just real questions that are on people’s minds. He answers them the best that he can. It’s not a decision-making forum, but it is a way for the company to get a window into how leadership is thinking, how he’s thinking, and to have debate and dialogue around that.

I think that’s something people really value. It is a transparent culture, and all 7 of the co-founders are still at the company. The vast majority of the first 20 to 30 employees are still at the company, and I think the culture underpins why we’ve been able to attract and retain some of the best talent in the industry. We don’t always pay people the most. We have very competitive compensation packages, but when Meta and others were out with these huge packages for some of the technical talent across the large language labs, I think we lost 2 people. Other labs lost dozens.

What parts of the business and the culture, specifically for researchers, do you think make that true?

That is true. I think it really is underpinned by the culture. That’s not just something we feel; empirically, when you talk to people, it’s, “I want to have the most impact possible. I want to work in a place where, again, there’s this idea of talent density mattering more than talent mass. I want to work in a place that’s actually collaborative, rather than having to fight for this one thing and feeling like it wasn’t discussed and debated in the right way, or that there wasn’t transparency around how a decision was made.”

I think that actually really matters because most of our team just wants to do really good work, and they’re attracted to the company for the mission. The idea of having an impact at a company like ours that is trying to develop this transformative technology, but to do it in a responsible way, really matters to people—not just on the research team, but across the company. We think that’s a real advantage for us, and it’s not something that we take lightly.

We have this concept of a race to the top. We don’t always have all the right answers, and we don’t always do everything perfectly, but we want others to look at some of the things we do, maybe emulate some pieces of that, and actually have the technology developed in a better way across the industry. I think people are also really attracted to that—not that we have all the answers, but that we can be a part of contributing to and leading how this can go well for humanity.

As you’re having conversations with people internally, what does the frontier feel like to you? I don’t just mean the model frontier. I mean, what are the next couple of rolls of the dice here in building AI in general? Everyone is aware that these things are powerful. Everyone’s using them. It’s diffusing, and people are becoming accepting of them. What feels to you like the frontier from the inside?

I think it’s this idea of a virtual collaborator. Again, because we’re focused on enterprise and really trying to change the productivity of knowledge work that’s done in the economy, I think we’re moving toward this vision or goal of a virtual collaborator.

Think of this as something that has context within your organization and can use all of the tools that are specific to you, whether they’re homegrown tools or tools that you purchase. It has the memory and the ability to effectively learn from mistakes you’ve made, but also mistakes that it’s made over time. Then it has the ability to work over a very long time horizon on not just a task, but an actual idea. What that means for us is that model capability has to continue to grow to support that, and then the products we build on top of it can unlock this virtual collaborator that we think can really accelerate knowledge work.

But you have to get it in the right form factor, right? This is where intelligence is not just a single dimension. It’s multiple things. The virtual collaborator combines many of those things. It’s not just generically smart; it’s smart for your use cases.

I think what we’re seeing in coding is something that we expect to see elsewhere. For us, Claude Code has led the way on that, as has much of the business we have with great customers that are pushing the coding frontier as well. But then you also see something like Co-work come along and start to unlock that. Claude Cowork is growing faster than Claude Code was growing if you index them to the same point in time. That’s remarkable, because developers are really fast adopters of this technology.

I think it’s because the model capabilities and the products are pushing toward this notion of a virtual collaborator. Even our product development today is not done by 1 product manager with 2 engineers shipping something over 3 months. It’s shipping daily, and there’s a fleet of agents working across the company on a specific task, so everyone becomes a manager. I think the implications of that, and the productivity gain that can come from it when it’s in the right form factor, mean that we’re very, very early in this. The potential for it is incredible.

Crazy to imagine. I’m curious how you’ve had to personally evolve to be able to stay doing this. You hear a lot of stories about how executives have to scale with the company or else get new executives. The business that you were at prior to this is a great business, but Cedar was a tiny, tiny fraction of the scale. So everyone is in this new, unprecedented thing.

You talked about the example of getting out of linear into more exponential-type thinking. That's one example of what I mean. But how have you managed it personally? What have you had to do? What's been the most painful? How do you manage your own ability to scale with this thing that's scaling faster than what we've seen before?

Yeah, it's really hard. But I think the important thing is to think in first principles, right? Everyone has priors when they come to something new. Thinking in first principles and having intellectual openness are really important.

I spent a lot of time with Tom Brown, our chief compute officer. He was actually one of the first people to interview me at the company. I remember we went on a walk a bit before I started. We walked around the Mission in San Francisco for 2½ hours, and he started to tell me about his vision for the future of the company.

13. What Could Derail the AI Revolution?

This was in early 2024, and I'll be honest: it sounded crazy. He walked me all the way home, and I remember coming in and telling my wife, "This is going to be wild. If even 10% of that is true, this is going to bend all paradigms—not just of things I've seen, but of what most people have seen." It turns out that a lot of what Tom said during that walk has come to fruition.

I remember that as an early formative experience, coming home and thinking, "Holy shit, this is going to be totally different and new, an incredible experience, but also a really challenging one." That's what it's been.

The other piece of this is just hiring great people. I try to hire people, and I tell them during the interview process, "I'm not really hiring you as a direct report of mine. I'm hiring you as a partner, and I want you to treat it as a partnership." That means there might be things that you and I disagree on. I want to hear that. I want to whiteboard it. I want to understand it.

14. Biotech and Healthcare

We've hired people from some of the best companies in the world. They come to this from a different perspective, right? They might come from a hyperscaler, a large software company, or financial services. In another lifetime, I worked at Blackstone in the private equity group. That training is really valuable for thinking about things at a granular level and not losing that. I'm not somebody who is comfortable at 50,000 feet. That's just not me, but you can't be at 500 feet for everything in this business. There's too much surface area. Having people who can be partners in that is really critical.

I think the last piece is to think about how the business evolves over time and where there might be moments or analogs to things that have happened in the past. I helped lead the financing that Airbnb did in the middle of the pandemic. It was a very different situation, right? The business lost 70% of its revenue in 7 weeks.

I know Brian just did a show with you. That was a harrowing time, but it was also a time without precedent, where you had to think about things with a clear perspective while everything was rapidly changing and there was not a good template or pattern to match.

On a personal level, it is hard to balance everything—family and friends, and certainly this job. It takes a big bite out of all that. But what I do try to do, maybe once a week, is, in a quiet moment, just think, "Wow, this is really cool." It's an incredible opportunity to work with this group of people on this problem, at this company, at this moment in time.

I try to do that again, maybe in a car ride, maybe late at night, or something like that. Just having that recognition and appreciation is really important.

What did he tell you on the walk that sounded most crazy?

We talked a lot about the scale of the computer infrastructure and what models could do in a short amount of time. I think he described a world that I would have said was kind of science fiction. A lot of what we're experiencing here and now has roots in that conversation.

15. The Kindest Thing

There are even more things he talked about that are probably beyond where we are today, but I think the commonality was that everything is going to happen much quicker than we think, and that both the implications and the capabilities of that can change.

He also had incredible optimism about the future. We talk about internally holding light and shade—that's one of the things we say. I came from that conversation with a bunch of questions, but also a sense of positivity about what could happen in the future.

It seems like we've spent most of our time talking about this because it's been the reality that we exist at the high end of that cone. What can you imagine would cause that to change to the low end of that cone? If we were to do some sort of pre-mortem a year from now and say, "Wow, actually, we didn't need nearly as much compute as we thought," or something like that, what can you imagine would shift us meaningfully in that cone?

I think the first thing would be the diffusion rate within our customers. The use cases are playing catch-up to model capability. These are humans in large organizations with a set of tools and practices and things that they've been doing for a really long time. Change is hard, right?

To the extent that diffusion hits a wall or slows down, that could affect the rate of change in terms of revenue growth. Certainly, scaling laws slowing down or not holding would be another thing. We don't see that. We can't say that with 100% certainty—I think that would be silly. We certainly believe in the trajectory, but model capabilities leveling off would be another thing.

Then, maybe third, is just how we think about being at the frontier. Today, we're at the frontier. I think we're defining the frontier of agentic AI. We need to stay there, right? It's a competitive market, and we're going to continue to invest in the technology, the compute, and the go-to-market that's required to be there. But that's not guaranteed, either.

What are you most excited about? You have a privileged seat. You get to literally see the future because it's happening inside the business before those outside the business see it. With that perspective and in that seat, what are you most excited about in the future?

I really think that the biotechnology and healthcare outcomes that can come from this technology are the things I'm most optimistic about. We may live in a world where you're diagnosed with a disease that is not curable, but in your lifetime, that cure can be found much more rapidly, and you actually might not die of that disease.

I think of this as: a lot of what we're doing today is helping to speed up the drug development process. A lot of the paperwork, clinical study reports, and things like that that need to be done—AI, and our solutions in particular, are helping to rapidly accelerate that.

I'm most optimistic and excited about when it goes further back into drug development and drug discovery because humans are incredibly capable at research. But if you think about these molecules and proteins, they're so complex, and such small changes have such big implications for the outcomes. AI is perfect for that.

If you think about what can happen when lab throughput goes up 10x or 100x and we can run that many more experiments, you could probably get better results faster, and that can help people around the world. It doesn't have to be limited to a small set of diseases or disorders. It can really go much further down the chain.

I think that has the potential to greatly alter the way that we live and the way that we interact, and that's really exciting to me.

I sure hope you're right. It sure seems like we're on that trajectory, and it's quite a future to imagine. This is so much fun. I feel like we covered so many interesting aspects of the business that I don't think you've done this before, so you don't get this amazing perspective.

When I do these, I ask the same traditional closing question: What is the kindest thing that anyone's ever done for you?

I have a brother who's 5½ years older than me, and we lived in California when he went to college. He got into everywhere he applied to, and he was going to go to medical school after that. I didn't know any of this at the time.

He ended up going to college in-state, and he did exceptionally well. It was years later that I had to pull this out of him. In deciding where to go to college, we were solidly middle class as a family, and this was 25 or 30 years ago. The financial aid packages weren't as robust as they are today.

A big factor in his decision, I found out many years later, was wanting to give me the opportunity to go wherever I wanted, even though that was 6 years out and who knows how it would turn out.

I didn't know that, and it was something that 12-year-old me or 13-year-old me would have never really understood. But now, many years later, I think that's something that was incredibly kind and is something that I still hold with me today.

Wow, I've done this about 600 times or something. I've never heard an answer like that. That's awesome. Amazing. Krishna, thanks so much for doing this with me.

Yeah, thanks for having me, Patrick. Really enjoyed it.