Animal Spirits · · 27 min
Talk Your Book: The Anthropic IPO
With Derek Yan
Summary
Derek Yan of CraneShares joins Animal Spirits to discuss Anthropic and the broader AI investment case. He argues that Anthropic's enterprise penetration is high, with more than half of the Fortune 500 as clients, yet spend is only about 0.1% of those companies' expenses, leaving room for large growth toward $60-100 billion ARR. He cites a reported GAAP profit in Q2 and roughly 5x annual gains in compute efficiency as evidence that AI economics can work, while acknowledging Anthropic lost some share to open-source models before regaining it. On infrastructure, he sees the risk shifting from overbuilding to underbuilding, with constraints from HBM and TSMC supply, energy, local permitting, and financing costs. He expects value to migrate across the AI stack over time, and he favors a rules-based, rebalancing approach. He also points to physical AI, robotics, and robotaxis, starting with industrial settings, as the next growth phase.
Key takeaways
- Derek Yan (bullish on Anthropic): Anthropic's enterprise spend penetration is tiny, so its dollar spend from large clients could grow 10-20x.
- Derek Yan (bullish on Anthropic): Anthropic is already GAAP-profitable and its compute efficiency is rising fast, so profitability is not a concern.
- Derek Yan on AI infrastructure & data centers: AI data center risk has shifted from overbuild to underbuild, with capacity owners gaining margin.
- Derek Yan (bullish on Taiwan Semiconductor Manufacturing Company): TSMC's fundamentals are strong; the post-earnings selloff was driven by Anthropic slowdown rumors, not TSMC.
- Derek Yan (bullish on Anthropic): Anthropic lost some share to open-source models this year but is regaining it based on recent data.
What was said, by company and theme
12 views from 1 speaker. Every quote is checked against the episode's transcript (how).
Anthropic
Across the shows: 114 speakers in other episodes, 171 bullish and 87 bearish calls, past 30 days
Where everyone stands on Anthropic
Derek Yan · KraneSharesBullish
Anthropic's enterprise spend penetration is tiny, so its dollar spend from large clients could grow 10-20x.
Anthropic's clients include more than half of the Fortune 500, yet those companies spend about 0.1% of their total expense on Anthropic, and the top 10 Fortune 500 clients about 1%. Derek argues the penetration is there but dollar spend is far below what people assumed. If penetration and spending keep rising, Anthropic could see 10x or 20x more spending from these companies. He concludes AI spending is only beginning and that this path could support $60-100 billion ARR. CraneShares holds Anthropic in its fund.
“Their revenue, they have clients that are more than half of the Fortune 500. But those Fortune 500s spend part of the total expense less than 0.1 into Anthropic. So the penetration is there, but the dollar spend was way lower than people thought.”at 5:57
Derek Yan · KraneSharesBullish
Anthropic is already GAAP-profitable and its compute efficiency is rising fast, so profitability is not a concern.
Derek says Anthropic reached a GAAP profit in Q2, which counters the view that AI companies can only lose money. He cites compute efficiency rising about 5x over a year. From that he infers Anthropic can generate roughly $2 of revenue per $1 spent on compute. He concludes profitability is not an issue if the efficiency trend continues. The argument depends on the trend persisting. CraneShares holds Anthropic in its fund.
“Using the GAAP standard, they achieve profit in Q2... that efficiency have increased 5x over a year. So which means they can probably generate like $2 per $1 spent on compute. So the profitability is not an issue if this trend continue.”at 7:19
Derek Yan · KraneSharesBullish
Anthropic lost some share to open-source models this year but is regaining it based on recent data.
Derek acknowledges Anthropic lost some market share to open-source models earlier this year. He says recent data shows it is gaining that share back. He uses this to argue that the leadership dynamic is hybrid rather than winner-take-all. CraneShares holds Anthropic in its fund.
“Anthropic this year, they have some moment where they are losing some market share to open source model. But if you look at like recent data, they're gaining market share back.”at 22:39
Taiwan Semiconductor Manufacturing Company
Across the shows: 6 speakers in other episodes, 5 bullish and 0 bearish calls, past 30 days
Where everyone stands on Taiwan Semiconductor Manufacturing Company
Derek Yan · KraneSharesBullish
TSMC's fundamentals are strong; the post-earnings selloff was driven by Anthropic slowdown rumors, not TSMC.
TSMC reported phenomenal revenue growth and raised its capex guidance. Management showed strong confidence in delivering the data centers being built. Even so, the stock fell after earnings because of rumors that Anthropic's revenue could slow. Derek calls the fundamentals really strong and says the volatility reflects uncertainty from rumors. The sell-off looks like sentiment, not a problem with TSMC's business.
“TSMC just delivered, phenomenal revenue growth, further CapEx guidance... Even with that, the equity was down post earnings because of the rumors that Anthropy [Anthropic] is gonna, their revenue can potentially slow down.”at 18:35
Robotics & autonomous vehicles
Where everyone stands on Robotics & autonomous vehicles
Derek Yan · KraneShares
Physical AI is the next phase of AI growth, with intelligence rather than hardware now the bottleneck.
Derek says physical AI is the next phase of AI growth, and CraneShares has recently focused its private investments there. He argues the same framework used to train large models over the past decade can now train robotic models that learn from real-world data. Hardware is already advanced, since humanoids can run marathons or do kung fu. So the bottleneck is intelligence, which he says is reaching a turning point. Real-world data from sources like games can help train robots. Physical AI needs different capital than data centers, since bodies and actuators require their own investors.
“So we feel like physical AI is the next phase of AI growth.”at 4:35
Derek Yan · KraneShares
Industrial manufacturing will adopt robots first; homes need much higher intelligence and safety.
Derek says manufacturing and industrial settings come first. Apptronik is based in Austin and works with local manufacturers, including auto companies. Industrial work is low-hanging fruit because it is tangible, high-value, and safer for robots to navigate than homes, where robots must deal with kids and dogs. Home robots would need GPT-4 or 5-level intelligence and further safety before deployment.
“I would say that's something is low hanging fruits because that's tangible, high value, and it's an environment that's really safer in a way for robotics to navigate compared to our home.”at 15:57
Derek Yan · KraneShares
Robotaxis are the first autonomous-driving adoption because their unit economics now work.
Derek says robotaxis are definitely the first adoption of autonomous driving. Autonomous driving has been promised for years, but the unit economics did not work before. Deploying autonomous driving as a robotaxi now gives a very high ROI, and that makes further deployment economic. He points to Waymo, Neura, and Elon Musk's efforts as players seeing the same opportunity. He expects penetration over the next three to five years to look very different from the past decade.
“I think robotaxis is definitely the first adoption... the challenge for that is the economic, the unit economy really doesn't work. But now... deploying autonomous driving as a robotaxi is gonna make it... The ROI is very high.”at 25:12
AI: overall outlook
Where everyone stands on AI: overall outlook
Derek Yan · KraneShares
AI's long-run addressable market is huge, with knowledge-worker penetration still very low.
Derek calls the long-term picture a total addressable market. He cites about $40 trillion in knowledge-worker economic value, where AI penetration is still at a very low point. From that, he argues investors should focus on where long-term margins and profits will be made and who will capture them.
“But the long term picture is, I think, a total addressable market. Look at AI, the $40,000,000,000,000 knowledge workers, all those penetration at very low point.”at 9:25
Derek Yan · KraneShares
AI value is migrating from chips to infrastructure, cybersecurity, and applications, so the winners will shift.
Derek says AI value today sits mostly in certain hardware and infrastructure, but it is broadening to data monitoring, model infrastructure, cybersecurity, and then applications. Applications are agent-centric now and may become more physical-AI-centric later. He argues the value migration means the leaders will change over time. That is why he prefers a rules-based approach that rebalances across layers instead of holding a concentrated set of names.
“We feel like the value can migrate. Today is more... Mostly just certain hardware or infrastructure, but it's broadening out to, say, data monitoring, say, model infrastructure, say, the cybersecurity, then to application.”at 11:44
AI infrastructure & data centers
Where everyone stands on AI infrastructure & data centers
Derek Yan · KraneShares
AI data center risk has shifted from overbuild to underbuild, with capacity owners gaining margin.
Earlier concern was overbuilding and a dot-com-style bubble, but Derek sees the risk now as underbuilding because demand is very large. Constraints come from HBM and TSMC supply, energy, local government pushback, and capital, including private credit and high-yield markets. Value can migrate from hardware to infrastructure owners whose margins improve, and GPU rental prices are rising. The constraints may slow overall progress, but he expects the long-term opportunity to remain large. Financing costs differ by player: hyperscalers borrow cheaply while new clouds face higher costs from investors demanding higher returns.
“The previous concern was really like, are we overbuilt? ... But like right now, the more the concern is, are we under built? Because demand is so large, but you have so much constraint.”at 8:19
Private markets for individual investors
Where everyone stands on Private markets for individual investors
Derek Yan · KraneShares
Direct cap-table ownership beats SPVs, which add counterparty risk and that companies dislike.
Derek describes two routes to private shares: direct investment through primary financing, cap-table positions, tender offers, or secondaries, and SPVs that give second-tier access through a counterparty. He says SPVs carry counterparty risk, and some companies are unhappy with them. CraneShares has so far used only direct investment. Institutional investors find cap-table ownership the most transparent way to express the allocation.
“So I think there's some counterparty risk and some company is really not happy with that. So we haven't really used SPV in our current investment.”at 21:43
Frontier AI labs & model race
Where everyone stands on Frontier AI labs & model race
Derek Yan · KraneShares
Model leadership is concentrating into about two players, with challengers winning lower-value work.
Derek says model leadership is consolidating from more than 10 players to about two. The dynamic is also hybrid: several leaders remain, and challengers can still gain share and capture value in lower-cost or lower-value jobs. He contrasts this with physical AI, where there are no leaders yet and leadership could change quickly.
“I think the leadership is getting more concentrated. Just think of the model space. It's getting more concentrated into just two players from 10 plus players.”at 22:28
Across the shows
What other episodes Chorus covered said about the same companies in the 30 days up to this one.
- Anthropic114 speakers · 171 bullish, 87 bearish calls
- Taiwan Semiconductor Manufacturing Company6 speakers · 5 bullish, 0 bearish calls
Topics
- AI addressable market
- AI margins and compute efficiency
- data center buildout constraints
- enterprise spend penetration
- foundry demand and capex
- model leadership concentration
- model market share
- physical AI and humanoid robots
- private market access structures
- robotaxi adoption
- robotics first adoption in manufacturing
- value migration across AI stack
More from Animal Spirits
- What the Hell Just Happened?! (EP. 490) · November 11, 2026
- Drowning Financially (EP. 489) · November 4, 2026
- Talk Your Book: Diversifying the AI Trade · November 2, 2026
- Why Baby Boomers Have All the Money (EP. 488) · October 28, 2026
- Talk Your Book: Beyond Bonds in Retirement · October 26, 2026
Ask your AI about this episode
Chorus gives Claude, ChatGPT and other assistants every view from this episode and every other show Chorus follows. Ask who agrees, who changed their mind, and what else was said about the same companies.
Chorus reports what speakers said. It isn't investment advice. How Chorus works · About Chorus