May 13, 2026 · 1h 22m · invest-like-the-best
Inside Anthropic's $100 Billion Al Compute Commitment | CFO Krishna Rao · Invest Like The Best
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of Invest Like The Best, Anthropic CFO Krishna Rao discusses the financial, strategic, and operational mechanics driving frontier AI development, detailing Anthropic's multi-chip hardware strategy, $100+ billion compute commitments, enterprise platform strategy, and unique corporate culture.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Patrick holds 24.3% of the talking time here. How this is scored →
Patrick as informed peer 4.8 Guest teaching 4.8 Guest disagreement 0.5 Patrick pushing back 1.4
speaking balance: gold is Patrick, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 37:15 Krishna dismissing traditional SaaS gross margin framing
Krishna directly rejects Patrick's suggestion to price models around static software gross margins, arguing that incremental variable COGS paradigms fail to represent fungible compute envelopes.
Hardest push from Patrick ▶ 44:05 Patrick questioning whether AI workflows are becoming subtly dystopianPatrick challenges the techno-optimist workflow narrative by pressing Krishna on whether humans taking direct orders from AI coordinators is unsettling and dystopian.
Biggest teaching moment ▶ 34:35 Krishna explaining Jevons paradox in Opus model pricingKrishna explains to Patrick why lowering token pricing on the flagship Opus model unlocked dramatic net revenue and token volume increases rather than hurting financial performance.
Patrick holds their own ▶ 4:14 Patrick detailing CUDA bare-metal hardware controlPatrick displays deep technical fluency by framing compute flexibility around bare-metal hardware abstraction, CUDA, and compiler optimization layers.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Patrick as informed peer | Guest teaching | Guest disagreement | Patrick pushing back | Why |
|---|---|---|---|---|---|---|
| Managing Compute Procurement and the Cone of Uncertainty | 6 | 5 | 1 | 2 | Patrick sets the stage by bringing up daily compute allocation and hardware bare metal mechanics, referencing CUDA. Krishna explains Anthropic's multi-chip orchestration across Amazon Trainium, Google TPUs, and Nvidia GPUs. | |
| Exponential Planning within the Cone of Uncertainty | 5 | 5 | 0 | 1 | Patrick asks about the practical mechanics of the 'cone of uncertainty'. Krishna explains the challenge of exponential planning and the internal floor on compute allocated to research. | |
| Measuring Model Efficiency and the Interconnected R&D Loop | 4 | 6 | 1 | 1 | Patrick inquires how Anthropic benchmarks efficiency gains. Krishna deconstructs the common 'car/sedan' analogy to explain that newer model leaps improve both token efficiency and raw capability simultaneously. | |
| Sponsor Announcements: Ramp, Rogo, and WorkOS | 4 | 6 | 1 | 2 | Following sponsor reads, Patrick asks why frontier returns are so high compared to using older cheaper models. Krishna reframes model capability from a 1D IQ score into multidimensional real-world enterprise execution. | |
| Recursive Self-Improvement and Product Velocity | 5 | 4 | 1 | 1 | Patrick probes whether recursive self-improvement widens the gap between frontier labs and open source. Krishna notes that Claude Code already writes over 90% of internal code and reframes the market as frontier vs non-frontier. | |
| Talent Density as an Accelerant for AI Research | 5 | 5 | 1 | 1 | Patrick asks if models will eventually eliminate the need for human researchers. Krishna counters by emphasizing Anthropic's identity as a scientific research lab where talent density directs discoveries. | |
| Exponential Decision-Making and Pattern Recognition | 4 | 5 | 0 | 1 | Patrick asks how an executive adapts to exponential capability curves. Krishna outlines scenario-based planning and using early coding adoption curves as an analog for wider business transformation. | |
| Price/Performance Optimization and Compute Absorption | 5 | 4 | 0 | 1 | Patrick asks about the trade-offs between cost, throughput, and speed in hardware procurement. Krishna details the granular matching of specific chip generations to distinct algorithmic workloads. | |
| Platform vs. Application Layer Strategy | 6 | 4 | 0 | 2 | Patrick raises the classic platform tension of building vertically vs enabling third parties. Krishna explains Anthropic's predominantly horizontal AWS-like platform strategy. | |
| Selective Application Building and Vertical Showcases | 6 | 4 | 1 | 2 | Patrick asks if Anthropic worries that prospective customers fear them as a potential competitor. Krishna acknowledges the rapid pace of model breakthroughs but stresses Anthropic's partner-centric posture. | |
| Pricing Strategy, Jevons Paradox, and Model Economics | 6 | 6 | 1 | 2 | Patrick challenges why AI labs do not sharply raise token prices given tight compute constraints. Krishna explains that price cuts on Opus triggered Jevons paradox, drastically accelerating overall consumption. | |
| Compute Spend ROI as the Primary Financial Governing Metric | 5 | 6 | 2 | 2 | Patrick inquires about gross margin discipline. Krishna pushes back against fitting Anthropic into a traditional SaaS software COGS paradigm, framing compute spend ROI as the true governing metric. | |
| Deep Co-Engineering with Cloud Infrastructure Partners | 4 | 4 | 0 | 1 | Krishna walks through how Anthropic integrates directly with hardware teams like Amazon Annapurna Labs and uses Claude internally to generate corporate financial reviews. | |
| Sponsor Announcements: Vanta and Ridgeline | 5 | 4 | 1 | 3 | Patrick voices concern over whether humans passively taking orders from AI tools creates a slightly dystopian dynamic. Krishna counters with an optimistic view of labor productivity and resource reallocation. | |
| Fundraising History and Evolving Investor Skepticism | 5 | 5 | 0 | 1 | Patrick asks about investor sentiment evolution across funding rounds. Krishna recounts fundraising during the FTX liquidation and DeepSeek release while addressing early doubts about safety and enterprise sales. | |
| Safety Research as an Enterprise Commercial Advantage | 4 | 6 | 1 | 1 | Krishna explains how core safety investments in interpretability and alignment directly unlock enterprise trust for Fortune 10 customers dealing with mission-critical workflows. | |
| The Fungible Compute Paradigm vs. Traditional Software | 5 | 5 | 1 | 1 | Patrick asks what questions a skeptical investor should ask frontier labs. Krishna emphasizes tracking genuine customer ROI and the fungible daily reallocation of compute between inference and training. | |
| AI Public Perception, Societal Potential, and Risk Communication | 5 | 4 | 0 | 2 | Patrick cites poll numbers showing AI having low public favorability. Krishna reflects on the need for the industry to communicate tangible societal benefits while remaining transparent about risks. | |
| Mythos Model Rollout, Safety Protocols, and Government Relations | 5 | 4 | 0 | 1 | Patrick asks about the internal decisions behind the Mythos model rollout. Krishna details its phased release strategy given its outsized cybersecurity capabilities. | |
| Anthropic's Culture of Humility, Rigorous Debate, and Talent Retention | 4 | 5 | 0 | 1 | Patrick asks how Krishna would explain Anthropic's culture. Krishna describes the lack of internal fiefdoms, rigorous bi-weekly open Q&As with Dario Amodei, and industry-leading technical talent retention. | |
| The Frontier of AI and Virtual Collaborators | 4 | 5 | 0 | 1 | Patrick asks what the frontier feels like internally. Krishna outlines the transition from isolated tasks to persistent virtual collaborators with institutional context and agent fleets. | |
| Personal Leadership and Scaling Through Exponential Growth | 4 | 4 | 0 | 1 | Krishna shares personal leadership lessons, recounting an early walk with Chief Compute Officer Tom Brown that reset his priors on technological acceleration. | |
| Visionary Predictions and Holding Light and Shade | 5 | 4 | 0 | 2 | Patrick asks what factors could push Anthropic to the pessimistic lower bound of its compute planning cone. Krishna identifies customer enterprise diffusion bottlenecks and potential scaling law friction. | |
| AI Applications in Healthcare and Biotechnology | 4 | 4 | 0 | 0 | Krishna explains his long-term excitement for AI-driven biological discovery, and closes by sharing a personal story about his older brother's college decision. |
Statements from this episode (42)
Rao: Overbuying compute bankrupts AI labs, while underbuying ruins frontier competitiveness
“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.”
Anthropic CFO Rao spends 30% to 40% of his time on compute
“I would say I spend 30 or 40% of my time on compute even today.”
Anthropic builds custom compilers to maximize chip-level hardware flexibility
“We're building our own compilers. We're really building things from the chip level up in order to have that customization and that flexibility”
Rao: Anthropic could generate billions in revenue from internal employee compute
“If we were to say to our employees, you can't use our models anymore. We could serve billions of dollars of revenue with that 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 unc…”
Anthropic sacrifices customer capacity to maintain absolute minimums for model training
“There's a level of compute for model development that we will not go below, right? So even if it means it's harder to serve customers, or we have to do kind of unnatural things when it comes to that, we want to continue to make that long-term investment in dev…”
Rao: Every Opus generation multiplies token processing efficiency
“In our case, we actually see both improvements, huge improvements in capability, but also in model efficiency. And so if you look at going from Opus, you know, four to four, five, four, six, and now four, seven, You know, each one of those leaps, they're not e…”
Rao: More efficient inference directly increases reinforcement learning efficiency
“If we're doing reinforcement learning on the model, it's basically inference within a sandbox with a reward function, right? And so if the model's better at more efficient inference, that RL is more efficient as well.”
Rao: Anthropic grew run-rate revenue from $9B to over $30B in four months
“We started the year with about nine billion dollars of run rate revenue, and we ended the quarter with, you know, north of thirty billion dollars of run rate revenue.”
Rao: Over 90% of Anthropic's code is written by Claude Code
“Right now within the company, you know, 90 plus percent of Our code is actually written by cloud code, right? A lot of cloud codes code is written by cloud code.”
Rao: Anthropic shipped 30 product and feature releases in January
“So we had 30 different product and feature releases in January.”
Rao: Anthropic does not train models on enterprise customer data
“We don't train on customer data in the enterprise side. On the prosumer side, it's only if you opt in.”
Anthropic explicitly denies an internal slowdown in AI model scaling laws
“From what we see, the scaling laws are not slowing down.”
Anthropic saw massive revenue surges following Claude Sonnet capability jumps
“Where, you know, starting with around Sonnet three, five, three, six, we started to see this really remarkable jump in capability, which was then followed By adoption and usage and revenue.”
Anthropic Uses AI Coding Adoption to Forecast Broader Economic AI Impact
“And, you know, 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. And so we kind of look at pattern recognition in our own business to try to predic…”
Anthropic partners with SpaceX for compute at Colossus facility in Memphis
“We announced a partnership with SpaceX for their Colossus facility in Memphis. We're really excited about that. It's gonna allow us to continue to expand, especially on the consumer and prosumer side.”
Anthropic signs 5GW TPU deal with Google and Broadcom starting in 2027
“So last month, you know, we signed a five gigawatt deal with Google and with Broadcom for TPUs starting in twenty-twenty-seven.”
Anthropic signs a $100 billion deal with Amazon for 5GW of compute
“We also signed a deal with Amazon for Tranium for up to five gigawatts as well. It was an over a hundred billion dollar commitment, and a lot of that computer's actually Like already landing and will land in the rest of this year into next year.”
Rao: Anthropic is compute-constrained across use cases today
“So the answer is, you know, we're constrained kind of across those use cases internally today.”
Rao: Anthropic focuses on platform layer rather than vertical applications
“Most of what we're building is platform, and 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 accrue even more value. We think that's where, what we're setting up…”
Rao: Anthropic builds first-party apps when it has foresight into future model capabilities
“That said, like, we will also build our own applications on that same platform where a couple things are true. Number one, if we feel like we have, you know, a vision into where the models are going, and we can kind of demonstrate that and create customer valu…”
Cutting Anthropic's Opus prices triggered a massive demand surge via Jevons paradox
“The changing the pricing for Opus actually, you know, you see this Jevons paradox, right? Like we lowered the price of it, but the consumption went up way, way more than what you would have expected.”
Anthropic CFO: Traditional software incremental cost paradigms do not fit AI labs
“And so th this idea of a variable cost that's like on the incremental to serve a customer is, is, is a little bit like it doesn't really fit our business, right? It tries to maybe fit our business into like a software paradigm, but that's not the case.”
Rao: Anthropic is the only AI lab on all three major clouds and chip architectures
“We are, you know, the only model that's on all three clouds today, we're the only language lab that's using all three of these chip platforms.”
Rao: Anthropic engineering is deeply embedded with Amazon's Annapurna Labs for Trainium
“Our teams are deeply embedded with the Annapurna Labs team. You know, we are, you know, really good users of Tranium. We've spent a lot of time and energy and worked closely with the team and internally we plan capacity together.”
Anthropic uses Claude to generate statutory financial statements across all legal entities
“Today, you know, 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.”
Anthropic built 70 finance skills to automate 90% of its monthly reviews
“We now have a library of skills for Claude that are specific to finance. I think last, last I checked, there were over 70 of them that everyone can kind of access through this kind of common repository. And on top of that, we built an MFR, a monthly financial …”
Anthropic CFO: AI creates a Jevons paradox for labor, increasing hiring
“And there's a little bit of this I think of it again like Jevin's paradox, but for labor, which is that we have people who become incredibly more productive. We actually, we've hired a lot more people because of that, because there's no shortage of work to do,…”
Rao: Anthropic Reached Nearly $1B Run Rate by Late 2024 Series E
“At the end of 20, 24, we raised the Series E. You know, the business had scaled to, you know, close to a billion dollars of run rate revenue, but the day of our first close was the day of the DeepSeek news came out.”
Anthropic sells to 9 of the Fortune 10 enterprises
“Like we now sell to nine of the fortune 10, all of those enterprises are entrusting us with, you know, customer information, with their data, they're interacting with their employees, sometimes even interacting with their customers as well.”
Anthropic had $250M revenue run rate when Krishna Rao joined
“Well, when I joined the business, it had about two hundred fifty million of run rate revenue, and the plan was to get to a billion, and I said, great, in what year?”
Anthropic dynamically swaps the same chips between inference and model training daily
“We, You know, run workloads on one day in the morning on a chip for inference, and in the afternoon or evening we use it for model development.”
Rao: Compute fungibility separates AI economics from traditional software and manufacturing
“That paradigm does not exist in a company like a software company or a factory, right? If you can't repurpose, 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 compani…”
Anthropic's annualized net dollar retention rate is currently over 500%
“Our net dollar retention rate is over 500% on an annualized basis.”
Anthropic CFO signed two $10M+ enterprise commitments during a 20-minute Uber ride
“On the way here, I was in, in, in an Uber, and I signed two double-digit million dollar commits, like, in the car ride, which was, like, 20 minutes.”
Anthropic's Mythos model found 250 security vulnerabilities versus 22 in prior models
“We had an open source code base that, you know, a prior model found 22 security vulnerabilities in, and Mythos then found 250.”
Anthropic adopts a strict 'America first' geopolitical stance on AI development
“We are very like America first in, in our approach. We want the technology to support the U S as well as, you know, democratic countries around the world. And that's one of the reasons why we've been working closely with the administration on something on some…”
All seven Anthropic co-founders and most early employees remain at the company
“All seven of the co-founders are still at the company. The vast majority of the first, you know, 20 to 30 employees are still at the company.”
Anthropic lost only two employees to Meta's aggressive AI talent poaching efforts
“When, you know, Meta and others were out, you know, with these huge packages for some of the technical talent in, in, across the large language labs, I think we lost two people and other labs lost dozens.”
Rao: Enterprise AI frontier is long-horizon virtual collaborators with organizational context
“I think it is towards this vision or this goal of like a virtual collaborator. And so think of this as, you know, something that has context within your organization that can use all of the tools That are specific to you, whether they be homegrown tools or too…”
Anthropic executes daily product shipping using fleets of internal AI agents
“Even our product development today is not done by, like, one product manager with two engineers shipping something over three months. It's shipping daily, and there's a fleet of agents that are working across the company on a specific task, so everyone kind of…”
Anthropic CFO: Enterprise use cases are currently lagging behind frontier model capabilities
“The use cases are playing catch up to the model capability.”
Rao: Anthropic's AI is accelerating clinical study reporting and drug development paperwork
“A lot of what we're doing today is helping to speed up The drug development process, right? A lot of the paperwork and clinical studies reports and things like that that are needed to be done. AI and our solutions in particular are helping to rapidly accelerat…”