{
 "episode_id": "invest-like-the-best-2026-05-13-krishna-rao-anthropic-s-cfo-on-managing-compute-scaling-to-3",
 "title": "Krishna Rao - Anthropic's CFO on Managing Compute, Scaling to $30B ARR, and the Returns to Frontier Intelligence",
 "podcast": "invest-like-the-best",
 "published_at": "2026-05-13",
 "speakers": [
  "Patrick O'Shaughnessy",
  "Krishna Rao"
 ],
 "summary": "Anthropic CFO Krishna Rao joins Patrick O'Shaughnessy to explain how the company manages compute, which he calls the canvas on which everything else gets built and the lifeblood of the business. He details the 'cone of uncertainty' framework, the fungible use of three chip platforms (Amazon Trainium, Google TPUs, NVIDIA GPUs) across model development, internal use, and serving customers, and why Anthropic forgoes near-term revenue to let its own models accelerate the next generation (90%+ of its code is now written by Claude Code). Rao argues the returns to frontier intelligence keep rising—especially in enterprise—citing growth from ~$9B to $30B run-rate revenue in a quarter, sales to nine of the Fortune 10, and net dollar retention over 500%. He walks through pricing philosophy (stability plus a Jevons-paradox price cut on Opus), why compute is fungible and not a software-style variable cost, capital formation ($75B raised since he joined, $50B more committed), and how Anthropic uses Claude internally on the finance team. He closes on the phased release of the Mythos model, Anthropic's collaborative-but-rigorous culture, the goal of a 'virtual collaborator' for knowledge work, and his optimism about AI accelerating drug discovery and healthcare.",
 "topics": [
  {
   "name": "Compute as lifeblood and the cone of uncertainty",
   "timestamp": "3:53",
   "summary": "Rao frames compute procurement as the most consequential decision in the company—buy too much and you go out of business, too little and you fall off the frontier. He introduces the 'cone of uncertainty,' modeling a range of one-to-two-year scenarios and working backwards while building flexibility into deals."
  },
  {
   "name": "Fungible compute, three chip platforms, and efficiency",
   "timestamp": "5:14",
   "summary": "Anthropic uses Trainium, TPUs, and GPUs fungibly across model development, internal use, and serving, via an orchestration layer built over years. Rao argues each new model (Opus 4 -> 4.5 -> 4.6 -> 4.7) brings a multiplier in token-processing efficiency, and that a dollar of compute goes further at Anthropic than anywhere else."
  },
  {
   "name": "Returns to frontier intelligence and recursive self-improvement",
   "timestamp": "13:02",
   "summary": "Rao argues intelligence is multidimensional (long-horizon tasks, tool use, speed), that new models unlock TAM, and that returns to the frontier are not slowing. With 90%+ of code written by Claude Code, models help build the next generation; Anthropic thinks of models as frontier-or-not rather than open-or-closed."
  },
  {
   "name": "Scaling laws and exponential vs linear thinking",
   "timestamp": "20:01",
   "summary": "Rao says pre-training, RL, and customer pain points all feed an internal loop, and that despite a skeptical, scientific-method culture, the scaling laws are not slowing down. He describes managing the business as scenarios with a low bar for updating priors rather than quarterly point estimates."
  },
  {
   "name": "Sourcing compute: near-term deals and the $100B+ commitments",
   "timestamp": "23:32",
   "summary": "Rao describes opportunistic near-term compute (a SpaceX/Colossus deal in Memphis) and large long-term deals—a five-gigawatt TPU deal with Google and Broadcom starting 2027 and up to five gigawatts of Trainium with Amazon, an over-$100B commitment. He frames compute as a layer cake assessed on price-performance, location, and timing."
  },
  {
   "name": "Platform vs application: AWS analogy and Claude Code",
   "timestamp": "28:22",
   "summary": "Rao says most of what Anthropic builds is platform (akin to early AWS), where customers building on it accrue even more value. Anthropic builds its own apps (Claude Code, Claude for Financial Services/Life Sciences/Security) when it can build ahead of model capability or demonstrate value, doing so in a partnership-oriented, level-playing-field way."
  },
  {
   "name": "Pricing, Jevons paradox, and margins on compute",
   "timestamp": "33:18",
   "summary": "Rao explains pricing has been stable across Haiku/Sonnet/Opus, the one big change being a price cut on Opus 4.5 that triggered a Jevons-paradox surge in consumption. He argues compute is not a software-style variable cost but a fungible resource, and Anthropic measures the return on the full compute envelope rather than per-customer marginal cost."
  },
  {
   "name": "Using Claude internally on the finance team",
   "timestamp": "38:51",
   "summary": "Rao describes producing statutory financial statements with Claude (human-checked), a library of 70+ finance skills, an automated monthly financial review that's 90-95% ready, and weekly revenue reports cut from hours to 30 minutes. He frames AI as a Jevons paradox for labor that makes talented people more productive rather than replacing them."
  },
  {
   "name": "Capital formation and investor misconceptions",
   "timestamp": "43:37",
   "summary": "Rao recounts raising through volatile moments (FTX during the Series D, DeepSeek news during the Series E), $75B raised since he joined plus $50B committed, and recurring investor skepticism about needing a frontier model, a small salesforce, and AI safety vs. business. He argues safety, interpretability, and alignment work has a downstream commercial benefit for enterprise trust."
  },
  {
   "name": "Mythos, safety, and government relationships",
   "timestamp": "56:50",
   "summary": "Rao explains the phased release of Mythos, a broadly capable model that spiked on cyber (finding 250 vulnerabilities in a codebase where a prior model found 22), as a template for responsible release. He describes Anthropic's America-first, regulation-has-a-role stance and working closely with the administration."
  },
  {
   "name": "Culture: collaboration, talent density, and transparency",
   "timestamp": "59:55",
   "summary": "Rao describes a culture with a real culture-interview bar, seven still-present co-founders, humility ('success is far from guaranteed'), rigorous debate followed by alignment, and biweekly transparent all-hands from Dario. He credits this culture with retaining talent—losing only two people to Meta's mega-offers while other labs lost dozens."
  },
  {
   "name": "The frontier ahead: virtual collaborators and biotech optimism",
   "timestamp": "1:04:39",
   "summary": "Rao describes the frontier as a 'virtual collaborator' for the ~$40T of global knowledge work—context-aware, tool-using, memory-equipped—with Cowork growing faster than Claude Code did. He names diffusion stalling, scaling laws leveling, and losing the frontier as downside risks, and says biotech and accelerated drug discovery are what excite him most."
  }
 ],
 "entities": [
  {
   "name": "Krishna Rao",
   "type": "person",
   "salience": "central",
   "role_in_episode": "Guest; CFO of Anthropic; explains compute strategy, economics, pricing, capital formation, and culture"
  },
  {
   "name": "Patrick O'Shaughnessy",
   "type": "person",
   "salience": "central",
   "role_in_episode": "Host of Invest Like the Best; probes compute allocation, returns to the frontier, pricing, and the human implications of AI"
  },
  {
   "name": "Anthropic",
   "type": "company",
   "salience": "central",
   "role_in_episode": "The subject company; frontier AI lab whose compute, economics, models, and culture are the focus of the conversation"
  },
  {
   "name": "Claude Code",
   "type": "product",
   "salience": "central",
   "role_in_episode": "Anthropic's Claude-led coding product; now writes 90%+ of Anthropic's code and is the lead example of building ahead of model capability"
  },
  {
   "name": "Compute",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "The central resource of the episode; treated as a fungible, return-generating envelope rather than a software variable cost"
  },
  {
   "name": "Cone of uncertainty",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "Rao's core planning frame for buying compute across a range of one-to-two-year exponential-growth scenarios"
  },
  {
   "name": "Frontier intelligence",
   "type": "concept",
   "salience": "central",
   "role_in_episode": "The thesis that returns to being at the model frontier are high and not slowing, especially in enterprise"
  },
  {
   "name": "Scaling laws",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "Discussed as alive and well across pre-training, post-training, and reasoning; Anthropic founders authored the original papers"
  },
  {
   "name": "Mythos",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "A new, broadly capable Anthropic model that spiked on cyber capability and was released in a phased, safety-conscious way"
  },
  {
   "name": "Amazon",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "Cloud and chip partner (Trainium); deeply embedded collaboration and an up-to-five-gigawatt Trainium deal"
  },
  {
   "name": "Google",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "Cloud and chip partner (TPUs); counterparty to a five-gigawatt TPU deal starting 2027"
  },
  {
   "name": "NVIDIA",
   "type": "company",
   "salience": "discussed",
   "role_in_episode": "GPU supplier and partner; CUDA cited as part of Anthropic's drive toward bare-metal control"
  },
  {
   "name": "Broadcom",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Chip partner; co-counterparty alongside Google on the five-gigawatt TPU deal"
  },
  {
   "name": "Trainium",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Amazon's AI chip platform; one of the three Anthropic uses fungibly, with Trainium 2/3 generations cited"
  },
  {
   "name": "TPU",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Google's chip platform; Anthropic was an early at-scale adopter (3rd-gen) when peers thought it was crazy"
  },
  {
   "name": "Dario Amodei",
   "type": "person",
   "salience": "discussed",
   "role_in_episode": "Anthropic CEO; cited as a better revenue predictor than Rao, author of 'Machines of Love and Grace,' and leader of transparent biweekly all-hands"
  },
  {
   "name": "Cowork",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Anthropic's extension of Claude Code to general knowledge work; growing faster than Claude Code did at the same index point"
  },
  {
   "name": "Tom Brown",
   "type": "person",
   "salience": "mentioned",
   "role_in_episode": "Anthropic's Chief Compute Officer; his early-2024 vision on a walk reshaped Rao's expectations for the company"
  },
  {
   "name": "Opus",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Anthropic's top model family (4 -> 4.5 -> 4.6 -> 4.7); subject of the one major price cut and the Jevons-paradox consumption surge"
  },
  {
   "name": "SpaceX",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Near-term compute partner via the Colossus facility in Memphis, expanding consumer/prosumer capacity"
  },
  {
   "name": "Claude",
   "type": "product",
   "salience": "discussed",
   "role_in_episode": "Anthropic's model/platform; used internally to produce financial statements and monthly reviews, and sold to nine of the Fortune 10"
  },
  {
   "name": "Meta",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Cited as making huge compensation offers to AI talent; Anthropic lost only two people while other labs lost dozens"
  },
  {
   "name": "Knowledge work",
   "type": "concept",
   "salience": "discussed",
   "role_in_episode": "The ~$40T annual market Anthropic aims to transform via a 'virtual collaborator'"
  },
  {
   "name": "Airbnb",
   "type": "company",
   "salience": "mentioned",
   "role_in_episode": "Rao's prior experience leading its pandemic financing; cited as an analog for clear thinking amid rapid change"
  }
 ],
 "claims": [
  {
   "holder": "Krishna Rao",
   "claim": "Compute is the single most important resource at Anthropic—the lifeblood and canvas on which everything is built—so misjudging how much to buy is existential.",
   "stance": "opinion",
   "conviction": 0.9,
   "subjects": [
    "Anthropic",
    "Compute"
   ],
   "timestamp": "3:53",
   "quote": "It is the most important thing in the company. It's like the canvas on which everything else gets built."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Buying compute is a knife-edge bet: too much and you go out of business, too little and you can't serve customers or stay at the frontier.",
   "stance": "assertion",
   "conviction": 0.85,
   "subjects": [
    "Anthropic",
    "Compute"
   ],
   "timestamp": "3:53",
   "quote": "if you buy too much compute, you go out of business.\nIf you buy too little compute, you can't serve your customers and you're not at the frontier."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic is the most efficient user of compute among the frontier labs, thanks to years of investment in a flexible orchestration layer across three chip platforms.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic",
    "Compute"
   ],
   "timestamp": "5:14",
   "quote": "We've invested in that over multiple years to be what I believe the most efficient users of compute amongst any of the frontier labs."
  },
  {
   "holder": "Krishna Rao",
   "claim": "A dollar of compute goes further inside Anthropic than anywhere else because it stresses the chips' limits and builds from the chip level up.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "Anthropic",
    "Compute"
   ],
   "timestamp": "6:47",
   "quote": "a dollar of compute inside our organization goes further than I think it does anywhere else"
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic will not let model-development compute fall below a floor, even at the cost of serving customers, because the returns to frontier intelligence are extremely high in enterprise.",
   "stance": "assertion",
   "conviction": 0.85,
   "subjects": [
    "Anthropic",
    "Frontier intelligence"
   ],
   "timestamp": "9:21",
   "quote": "There's a level of compute for model 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."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Each new model generation delivers a step-change multiplier in token-processing efficiency, unlike the car analogy where more power means worse fuel economy.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "Anthropic",
    "Opus"
   ],
   "timestamp": "10:44",
   "quote": "each one of those leaps to a new model has a multiplier in terms of how much more efficient it is at processing tokens effectively"
  },
  {
   "holder": "Krishna Rao",
   "claim": "The returns to frontier intelligence are not slowing down and are a core thesis of Anthropic's business, especially in enterprise.",
   "stance": "assertion",
   "conviction": 0.85,
   "subjects": [
    "Anthropic",
    "Frontier intelligence"
   ],
   "timestamp": "13:02",
   "quote": "that's a core thesis of our business set, especially in enterprise, the returns to frontier intelligence are not slowing down"
  },
  {
   "holder": "Krishna Rao",
   "claim": "Model intelligence is multidimensional—long-horizon tasks, tool use, speed—not a single IQ-like score, and published benchmarks are often saturated.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic",
    "Frontier intelligence"
   ],
   "timestamp": "13:02",
   "quote": "Intelligence for us is multidimensional. It's not just a score. In fact, we find that, yes, everyone publishes their model benchmark cards. We find that a lot of those benchmarks are saturated."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic categorizes models as frontier-or-not rather than open-or-closed, and is investing behind the thesis that frontier models capture the economic value.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "17:11",
   "quote": "We don't really think about models as closed or open. We think of them as frontier or not."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Talent density beats talent mass; the densest collection of AI research and inference-engineering talent, paired with the best models, is a winning combination.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "19:02",
   "quote": "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"
  },
  {
   "holder": "Krishna Rao",
   "claim": "Despite Anthropic's skeptical, scientific-method culture, the scaling laws are not slowing down.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "Scaling laws",
    "Anthropic"
   ],
   "timestamp": "21:32",
   "quote": "from what we see, the scaling laws are not slowing down"
  },
  {
   "holder": "Krishna Rao",
   "claim": "Quarterly, point-estimate forecasting does not work for Anthropic; the business is too dynamic and requires scenario thinking with a low bar for updating priors.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "22:12",
   "quote": "this old like, well, we'll forecast once a quarter and we'll revisit this in three months at the next board meeting. That doesn't work for our business."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Most of what Anthropic builds is platform, akin to the early days of AWS, where customers building on the platform accrue even more value than the platform itself.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic",
    "Claude"
   ],
   "timestamp": "28:22",
   "quote": "most of what we're building is platform. We think that there's so many examples of where a platform can accrue a lot of value, but the customers who are building on that platform actually crew even more value."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic's strategy is mostly horizontal, building vertical apps only where it can add value, build ahead of model capability, or demonstrate platform value to the ecosystem.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "30:01",
   "quote": "I think of our strategy as mostly horizontal. We'll build vertical where we think we have some value to add or a perspective that's useful or a way to demonstrate to the market how we think about our platform adding value."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic's approach to a market scared of it is to be more partner-oriented than is typical—early access programs, listening to customers—even though releases still sometimes surprise people.",
   "stance": "opinion",
   "conviction": 0.7,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "31:16",
   "quote": "our approach to it is probably a little different and more partner oriented"
  },
  {
   "holder": "Krishna Rao",
   "claim": "Cutting the price of the Opus family triggered a Jevons paradox—consumption rose far more than expected—validating pricing for value over raising prices.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic",
    "Opus"
   ],
   "timestamp": "33:18",
   "quote": "Changing the pricing for Opus, you see this Jevons Paradox. We lowered the price of it, but the consumption went up way, way more than what you would have expected."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Treating compute as a per-customer variable cost wrongly forces Anthropic into a software paradigm; compute is a fungible envelope supporting all activities, and the right measure is the return on that whole envelope.",
   "stance": "opinion",
   "conviction": 0.85,
   "subjects": [
    "Anthropic",
    "Compute"
   ],
   "timestamp": "35:44",
   "quote": "This idea of a variable cost that's on the incremental to serve a customer, it doesn't really fit our business. It tries to maybe fit our business into a software paradigm. That's not the case."
  },
  {
   "holder": "Krishna Rao",
   "claim": "AI acts as a Jevons paradox for labor—making talented people far more productive and leading Anthropic to hire more, not fewer, people.",
   "stance": "opinion",
   "conviction": 0.75,
   "subjects": [
    "Anthropic",
    "Knowledge work"
   ],
   "timestamp": "42:07",
   "quote": "I think of it again like Jevin's paradox, but for labor, which is that we have people become incredibly more productive. We've hired a lot more people because of that, because there's no shortage of work to do."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic's investment in safety, interpretability, and alignment—made for mission reasons—has a downstream commercial benefit because enterprises entrusting sensitive workloads want a company they can trust.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "45:00",
   "quote": "When you have this investment that we made and will continue to make in safety, interpretability, alignment, that actually endures to the benefit of the enterprise customers as well"
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic raises large amounts of capital because of the cone of uncertainty rather than to fund losses, since the underlying business runs efficiently.",
   "stance": "assertion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "45:00",
   "quote": "the reason we raise this capital is more because of that cone of uncertainty than it is to fund losses in the business today"
  },
  {
   "holder": "Krishna Rao",
   "claim": "The hardest thing for investors to grasp is that compute is a fungible, self-reinforcing resource (inference in the morning, model development in the evening) rather than a separable variable cost.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic",
    "Compute"
   ],
   "timestamp": "49:13",
   "quote": "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."
  },
  {
   "holder": "Krishna Rao",
   "claim": "The phased, expanding release of Mythos—focused on positive defensive use of its cyber capability—is a template Anthropic thinks could be used for future capable models.",
   "stance": "opinion",
   "conviction": 0.7,
   "subjects": [
    "Mythos",
    "Anthropic"
   ],
   "timestamp": "56:50",
   "quote": "we think that's a template that could be used for the future, but because of this one particular area, we wanted to be cognizant of that in how we released it"
  },
  {
   "holder": "Krishna Rao",
   "claim": "Regulation has a role to play and Anthropic takes an America-first stance, wanting the technology to support the US and democratic countries.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "58:39",
   "quote": "We are very America first in our approach. We want the technology to support the US as well as democratic countries around the world."
  },
  {
   "holder": "Krishna Rao",
   "claim": "Anthropic's distinctive culture—rigorous debate followed by real alignment, transparency, humility, and no fiefdoms—is empirically why it attracts and retains top talent despite not always paying the most.",
   "stance": "opinion",
   "conviction": 0.8,
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "59:55",
   "quote": "The culture underpins the reason why we've been able to attract and retain some of the best talent in the industry, because we don't always pay people the most."
  },
  {
   "holder": "Krishna Rao",
   "claim": "The frontier ahead is a 'virtual collaborator' for knowledge work—context-aware, tool-using, with memory and long-horizon ability—mirroring what has already happened in coding.",
   "stance": "prediction",
   "conviction": 0.75,
   "subjects": [
    "Anthropic",
    "Knowledge work"
   ],
   "timestamp": "1:04:39",
   "quote": "I think it is towards this vision or this goal of like a virtual collaborator. And so think of this as something that has context within your organization, that can use all of the tools that are specific to you"
  },
  {
   "holder": "Patrick O'Shaughnessy",
   "claim": "The returns to using a frontier model are so high that consumers and enterprises switch to the newest model immediately rather than riding cheaper older models.",
   "stance": "assertion",
   "conviction": 0.7,
   "subjects": [
    "Frontier intelligence"
   ],
   "timestamp": "12:32",
   "quote": "The second Opus 4.7 comes out, even me as a consumer, the thing you do is you switch it on, or GPT 5.5 comes out, you switch on the new one right away, like I want the best."
  },
  {
   "holder": "Patrick O'Shaughnessy",
   "claim": "It feels slightly dystopian that humans are starting to do what the AI tells them to do rather than directing it.",
   "stance": "opinion",
   "conviction": 0.5,
   "subjects": [
    "Knowledge work"
   ],
   "timestamp": "41:30",
   "quote": "if I really close my eyes, I'm just like doing what it tells me versus me telling it what to do. It's a really interesting just human dynamic that I'm curious for your take on."
  }
 ],
 "predictions": [
  {
   "holder": "Krishna Rao",
   "prediction": "Adding far more compute (2x, 5x, even 10x) overnight would now be deployed very rapidly across Anthropic's use cases, since it can spin up almost any type of compute quickly.",
   "timeframe": "unspecified",
   "subjects": [
    "Anthropic",
    "Compute"
   ],
   "timestamp": "27:05",
   "quote": "I would say today that getting a bunch more compute, I think it would be deployed very rapidly across those different use cases."
  },
  {
   "holder": "Krishna Rao",
   "prediction": "Anthropic's customers will continue to invest heavily in more tokens with each newer model, repeating the cycle of model leaps unlocking TAM.",
   "timeframe": "unspecified",
   "subjects": [
    "Anthropic",
    "Frontier intelligence"
   ],
   "timestamp": "13:02",
   "quote": "customers see that and then they invest really heavily in more tokens with the newer models. And we just have seen that cycle play out again and again."
  },
  {
   "holder": "Krishna Rao",
   "prediction": "An additional $50 billion of capital will come in the future from the Amazon and Google deals signed last month.",
   "timeframe": "unspecified",
   "subjects": [
    "Anthropic",
    "Amazon",
    "Google"
   ],
   "timestamp": "45:00",
   "quote": "We have another $50 billion that will come in the future from the Amazon and Google deals that we inked last month."
  },
  {
   "holder": "Krishna Rao",
   "prediction": "The five-gigawatt TPU deal with Google and Broadcom will begin in 2027, with much of the over-$100B of compute landing this year and next.",
   "timeframe": "starting 2027",
   "subjects": [
    "Anthropic",
    "Google",
    "Broadcom",
    "TPU"
   ],
   "timestamp": "23:32",
   "quote": "Starting in 2027, we also signed a deal with Amazon for Trainium for up to five gigawatts as well. It was an over $100 billion commitment and a lot of that computer is actually already landing and will land in the rest of this year and the next year.",
   "due": "2027-01-01"
  },
  {
   "holder": "Krishna Rao",
   "prediction": "AI will accelerate drug development and discovery so that diseases incurable at diagnosis may have a cure found within a patient's lifetime, including via 10x-100x lab throughput.",
   "timeframe": "unspecified",
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "1:12:55",
   "quote": "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 might not die of that disease."
  },
  {
   "holder": "Krishna Rao",
   "prediction": "Over the long term, the opportunity from AI will be significantly greater than the risks and downsides, though the path will not be perfectly smooth.",
   "timeframe": "long term",
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "52:43",
   "quote": "I do think that over the long term, the opportunity is going to be significantly higher and greater than some of the risks and the downsides that will happen. But that doesn't mean it's going to be perfectly smooth on the curve."
  }
 ],
 "facts": [
  {
   "statement": "Anthropic started the year with about $9 billion of run-rate revenue and ended the quarter at $30 billion of run-rate revenue.",
   "subjects": [
    "Anthropic"
   ],
   "value": "$9B -> $30B run-rate revenue",
   "timestamp": "13:02",
   "quote": "We started the year with about $9 billion of run rate revenue, and we ended the quarter with the $30 billion of run rate revenue.",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic is a customer of three chip platforms—Amazon's Trainium, Google's TPUs, and NVIDIA's GPUs—used fungibly.",
   "subjects": [
    "Anthropic",
    "Amazon",
    "Google",
    "NVIDIA"
   ],
   "value": "3 chip platforms (Trainium, TPU, GPU)",
   "timestamp": "5:14",
   "quote": "we are customers of Amazon's Trainium chip, Google's CPUs and NVIDIA's GPUs",
   "source": "Krishna Rao"
  },
  {
   "statement": "Over 90% of Anthropic's code is now written by Claude Code, much of it Claude Code writing its own code.",
   "subjects": [
    "Anthropic",
    "Claude Code"
   ],
   "value": "90%+ of code",
   "timestamp": "17:11",
   "quote": "right now, within the company, 90 plus percent of our code is actually written by Claude Code",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic had 30 different product and feature releases in January.",
   "subjects": [
    "Anthropic"
   ],
   "value": "30 releases in January",
   "timestamp": "17:11",
   "quote": "We had 30 different product and feature releases in January.",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic announced a partnership with SpaceX for its Colossus facility in Memphis to expand consumer and prosumer capacity.",
   "subjects": [
    "Anthropic",
    "SpaceX"
   ],
   "timestamp": "23:32",
   "quote": "We announced a partnership with SpaceX for their Colossus facility in Memphis.",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic signed a five-gigawatt TPU deal with Google and Broadcom, plus up to five gigawatts of Trainium with Amazon, an over-$100B commitment.",
   "subjects": [
    "Anthropic",
    "Google",
    "Broadcom",
    "Amazon"
   ],
   "value": "5GW + up to 5GW; >$100B",
   "timestamp": "23:32",
   "quote": "Last month, we signed a five gigawatt deal with Google and with Broadcom for TPUs.",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic is the only model lab that is on all three clouds and using all three of these chip platforms.",
   "subjects": [
    "Anthropic"
   ],
   "timestamp": "37:22",
   "quote": "we are the only model that's on all three clouds. Today, we're the only model that has large language lab that's using all three of these chip platforms.",
   "source": "Krishna Rao"
  },
  {
   "statement": "The biggest pricing change Anthropic made was bringing down the price of the Opus family when it launched Opus 4.5.",
   "subjects": [
    "Anthropic",
    "Opus"
   ],
   "timestamp": "33:18",
   "quote": "The biggest pricing change we made was to bring down the price of the Opus family when we launched Opus 4.5.",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic can produce the statutory financial statements for all of its legal entities using Claude (human-checked).",
   "subjects": [
    "Anthropic",
    "Claude"
   ],
   "timestamp": "38:51",
   "quote": "all of our legal entities, we can produce the statutory financial statements using Claude. And yes, the human checks it, but all of those financial statements are produced with Claude",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic's finance team has a library of over 70 Claude skills specific to finance, plus a monthly financial review skill that produces a review that's 90-95% ready.",
   "subjects": [
    "Anthropic",
    "Claude"
   ],
   "value": "70+ finance skills; MFR 90-95% ready",
   "timestamp": "38:51",
   "quote": "Last I checked, there were over 70 of them",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic is a little over five years old, with its first dollar of revenue three years ago this past March and its first real frontier model in March 2024.",
   "subjects": [
    "Anthropic"
   ],
   "value": "~5 years old; first revenue March 2023; first frontier model March 2024",
   "timestamp": "33:18",
   "quote": "We only had a frontier model for real for the first time in March of 2024",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic has raised $75 billion since Rao joined the company two years ago, with another $50 billion committed from the Amazon and Google deals.",
   "subjects": [
    "Anthropic"
   ],
   "value": "$75B raised + $50B committed",
   "timestamp": "45:00",
   "quote": "We've raised $75 billion since I joined the company. We have another $50 billion that will come in the future from the Amazon and Google deals that we inked last month.",
   "source": "Krishna Rao"
  },
  {
   "statement": "When Rao joined, Anthropic had about $250 million of run-rate revenue; he was closing the Series D, and by end of 2024 raised the Series E at close to $1B run-rate.",
   "subjects": [
    "Anthropic"
   ],
   "value": "~$250M run-rate at join; ~$1B at Series E",
   "timestamp": "47:51",
   "quote": "When I joined the business, it had about 250 million of run rate revenue and the plan was to get to a billion and I said, great, in what year?",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic now sells to nine of the Fortune 10.",
   "subjects": [
    "Anthropic"
   ],
   "value": "9 of the Fortune 10",
   "timestamp": "45:00",
   "quote": "we now sell to nine of the Fortune 10",
   "source": "Krishna Rao"
  },
  {
   "statement": "Anthropic's net dollar retention rate is over 500% on an annualized basis.",
   "subjects": [
    "Anthropic"
   ],
   "value": ">500% net dollar retention",
   "timestamp": "50:40",
   "quote": "Our net dollar retention rate is over 500 percent on an annualized basis.",
   "source": "Krishna Rao"
  },
  {
   "statement": "On the way to the interview, Rao signed two double-digit-million-dollar commitments during a roughly 20-minute Uber ride.",
   "subjects": [
    "Anthropic"
   ],
   "value": "2 double-digit-$M commits in ~20 min",
   "timestamp": "51:29",
   "quote": "I was in an Uber and I signed two double digit million dollar commits in the car ride, which was like 20 minutes",
   "source": "Krishna Rao"
  },
  {
   "statement": "Mythos found 250 security vulnerabilities in an open-source codebase where a prior model had found 22.",
   "subjects": [
    "Mythos"
   ],
   "value": "250 vs 22 vulnerabilities",
   "timestamp": "56:50",
   "quote": "we had an open source code base that, a prior model found 22 security vulnerabilities in, and Mythos then found 250",
   "source": "Krishna Rao"
  },
  {
   "statement": "All seven of Anthropic's co-founders and the vast majority of the first 20-30 employees are still at the company.",
   "subjects": [
    "Anthropic"
   ],
   "value": "7 co-founders; first 20-30 employees retained",
   "timestamp": "59:55",
   "quote": "It is one where all seven of the co-founders are still at the company. The vast majority of the first 20 to 30 employees are still at the company.",
   "source": "Krishna Rao"
  },
  {
   "statement": "When Meta and others made huge compensation offers to technical talent, Anthropic lost two people while other labs lost dozens.",
   "subjects": [
    "Anthropic",
    "Meta"
   ],
   "value": "2 lost vs dozens at other labs",
   "timestamp": "59:55",
   "quote": "I think we lost two people, and other labs lost dozens.",
   "source": "Krishna Rao"
  },
  {
   "statement": "There is roughly $40 trillion of knowledge work done annually in the world, which Anthropic aims to make more productive.",
   "subjects": [
    "Knowledge work",
    "Anthropic"
   ],
   "value": "~$40 trillion/year",
   "timestamp": "1:04:39",
   "quote": "There's something like $40 trillion of knowledge work done annually in the world.",
   "source": "Krishna Rao"
  },
  {
   "statement": "Cowork is growing faster than Claude Code was when indexed to the same point in time.",
   "subjects": [
    "Cowork",
    "Claude Code"
   ],
   "timestamp": "1:04:39",
   "quote": "Co-Work is growing faster than Claude Code was if you index them to the same point in time.",
   "source": "Krishna Rao"
  },
  {
   "statement": "Weekly revenue/compute-utilization reports that used to take hours now take about 30 minutes using Claude.",
   "subjects": [
    "Anthropic",
    "Claude"
   ],
   "value": "hours -> 30 minutes",
   "timestamp": "38:51",
   "quote": "what used to take hours to produce a weekly report for what's driving revenue or what's driving our compute utilization now comes down to 30 minutes",
   "source": "Krishna Rao"
  }
 ],
 "notable_quotes": [
  {
   "speaker": "Krishna Rao",
   "timestamp": "3:53",
   "quote": "It is the most important thing in the company. It's like the canvas on which everything else gets built.",
   "why": "Crystallizes the entire episode's thesis that compute is the foundational resource on which Anthropic is built."
  },
  {
   "speaker": "Krishna Rao",
   "timestamp": "3:53",
   "quote": "if you buy too much compute, you go out of business.\nIf you buy too little compute, you can't serve your customers and you're not at the frontier.",
   "why": "The memorable knife-edge framing of why compute procurement is the hardest decision at the company."
  },
  {
   "speaker": "Krishna Rao",
   "timestamp": "17:11",
   "quote": "right now, within the company, 90 plus percent of our code is actually written by Claude Code",
   "why": "Striking, quotable data point on recursive self-improvement that listeners would remember."
  },
  {
   "speaker": "Krishna Rao",
   "timestamp": "50:40",
   "quote": "Our net dollar retention rate is over 500 percent on an annualized basis.",
   "why": "An extraordinary growth metric that anchors the 'returns to frontier intelligence' argument."
  },
  {
   "speaker": "Krishna Rao",
   "timestamp": "51:29",
   "quote": "I was in an Uber and I signed two double digit million dollar commits in the car ride, which was like 20 minutes",
   "why": "Vivid anecdote conveying the speed and scale of enterprise demand."
  },
  {
   "speaker": "Krishna Rao",
   "timestamp": "59:55",
   "quote": "Our competitors are incredibly capable and success is far from guaranteed.",
   "why": "The laptop-sticker line that captures Anthropic's culture of humility and urgency."
  }
 ],
 "extracted_by": "agent",
 "prompt_version": "episode-extract@3",
 "_file": "invest-like-the-best-2026-05-13-krishna-rao-anthropic-s-cfo-on-managing-compute-scaling-to-3.json"
}