Jaimin Rangwalla of Coatue says Anthropic's annualized revenue jumped from $9bn in December to $30bn, with OpenAI and Anthropic together near $60bn.
Rangwalla says Coatue now follows gigawatts across the AI supply chain, since memory tightness has lasted longer than past cycles with supply deals guaranteed through 2029 and 2030.
Rangwalla predicts an individual's semiconductor footprint could expand about tenfold as agents launch agents, a shift he calls one of AI's most underappreciated unlocks.
Coatue's CIO says AI's profits go to shortage sellers
Rangwalla argues AI is still early, and the money is in the shortage. Demand from enterprises, individuals and agents is outrunning the supply of memory, power, chips and optics. Sellers of those inputs earn outsized profits, while the big buyers spending on them see their multiples compress. Coatue therefore tracks gigawatts and the whole supply chain rather than a single stock.
Takeaways
AI adoption is still early despite huge valuations. Private valuations are enormous, with OpenAI above $800bn and SpaceX at $1.25tn, yet revenue ramps are faster still. Anthropic's annualized revenue rose from $9bn in December to $30bn, and AI spending is not a one-for-one labor swap.
Profits flow to sellers of the shortage. Rangwalla splits the market into sellers of the shortage, such as memory, power and semiconductor makers, and buyers such as Microsoft, Amazon and Meta. Buyers' multiples have compressed as capital spending rises, while memory sellers beat estimates by 30% to 40% a quarter.
Coatue follows gigawatts across the supply chain. Coatue bet early on NVIDIA but now tracks gigawatts as the atomic unit of AI growth, along with inputs, buyers and users. Memory tightness has outlasted past cycles, with supply agreements guaranteed through 2029 and 2030, and optical components are a surprise breakout.
Agents launching agents will multiply chip demand. He calls this one of the most underappreciated unlocks and expects an individual's semiconductor footprint to grow about tenfold. Boris Cherny runs thousands of agents overnight. The CPU-to-GPU ratio could flip to one GPU for four CPUs, with Intel, AMD and Arm named as CPU players.
The real bottleneck is many bottlenecks at once. Rangwalla argues the constraint is that so many bottlenecks exist simultaneously, including memory, power and optics. TSMC, one of Coatue's largest holdings, is raising prices, and some companies have lifted earnings three to seven times within a year or two.
Sentiment is unreliable and volatility is constant. Sentiment is basically a coin flip as a predictor. S&P 500 earnings growth for 2026 is about 15%, accelerating to 18%. Some Coatue stocks fall 5% to 10% for no clear reason, so the firm uses hedges, puts and shorts alongside fundamentals.
AI should create jobs and open IPOs to retail. OpenAI, SpaceX and Anthropic already rank among the top 30 companies, and public listings would give retail investors access. He expects jobs to spread out rather than vanish, with new business formation near highs, though training electricians takes two or three years.
Risks Rangwalla names
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
A DeepSeek-style efficiency breakthrough is the risk he names. He thinks it would be good for AI in the long run and points to Jevons paradox.
Markets see crises roughly every ten years, and a 30% decline has happened before, so Coatue accepts volatility as part of the job.