Gavin Baker, managing partner and CIO at Atreides Management and a closely watched AI investor on X, laid out what he called the “mega bull case” for AI infrastructure in a post over the weekend. Baker’s argument reframes who captures the economics of the AI buildout, and it lands at the center of a sharp cash flow inversion now unfolding between the chipmakers and the hyperscalers funding them.
Baker’s Mega Bull Case in His Own Words
Writing on X, Baker argued: “The mega bull case for AI infrastructure would be if market share shifted away from certain frontier labs with 90%+ inference margins toward cheaper models, whether open-source or closed… Margin dollars would effectively get redistributed from the frontier labs to AI infrastructure providers. The infra winners would be those with the lowest per token cost.”
Michael Burry has been circling the same theme, having similarly posted, “The AI race is shifting from bigger models to cheaper, smarter systems.”
Translation: if inference workloads migrate from expensive proprietary models toward cheaper open-source or vertically integrated alternatives, more of the industry’s margin dollars flow to the picks-and-shovels vendors. Baker singled out NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) CEO Jensen Huang’s aggressive open-source push as evidence Huang sees this dynamic playing out, adding that “with SpaceX and Meta being vertically integrated and possessing the #3 and #4 models respectively it is more possible than ever.”
A Major Cash Flow Transfer
The AI boom is quietly redirecting where the cash lands. According to a Bank of America summary circulated on X, NVIDIA, Micron Technology (NASDAQ:MU), Broadcom (NASDAQ:AVGO), and Applied Materials (NASDAQ:AMAT) are now expected to generate a combined $430 billion in free cash flow over the next 12 months, more than triple what they produced just two years ago.
The mirror image is unfolding at the companies footing the bill. Amazon (NASDAQ: AMZN), Alphabet (NASDAQ: GOOGL), Meta Platforms (NASDAQ: META), Microsoft (NASDAQ: MSFT), and Oracle (NASDAQ: ORCL) are projected to see combined free cash flow fall from a 2024 peak of roughly $250 billion to about $100 billion by the end of 2026, as they pour a projected $1.8 trillion into AI capex across 2026 and 2027. In short, the AI buildout is compressing cash flow at the hyperscalers while expanding it at the companies selling the chips, memory, networking gear, and equipment needed to keep the boom running.
The trailing numbers already tell the story:
| Company | Most Recent FY Free Cash Flow | Prior Year FCF |
|---|---|---|
| NVIDIA (FY2026) | $96.7B | $60.9B |
| Broadcom (FY2025) | $26.9B | $19.4B |
| Applied Materials (FY2025) | $5.7B | $7.5B |
| Micron (FY2025) | $1.7B | $0.1B |
| Alphabet Q1 2026 capex | $35.7B (+107% YoY) | |
| Amazon Q1 2026 capex | $44.2B | |
| Meta 2026 capex guide | $125B–$145B | |
Micron’s flip is the most vivid. Its Q3 FY2026 report showed revenue of $41.46 billion, up 345.7% year over year, non-GAAP EPS of $25.11, and free cash flow of $18.30 billion in a single quarter. NVIDIA’s Q1 FY27 pushed Data Center revenue to $75.25 billion, +92% YoY, with $119 billion in total supply commitments disclosed in its 8-K filing. Broadcom guided Q3 AI semiconductor revenue to $16 billion, up over 200% year-on-year, with backlog visibility extending into 2028.
What to Watch Next
The market has already priced part of the split. Micron is up 228.3% year to date, Applied Materials 123.9%, and Broadcom 10.9%. Meanwhile Microsoft is down 19.1% YTD, with investors watching the capex bill more closely. Yet forward multiples on the chipmakers have compressed sharply as earnings estimates race higher. Micron trades at a forward P/E of roughly 6, Broadcom at 21, NVIDIA at 24, and Applied Materials at 38.
The whole thesis hinges on hyperscaler ROI. If Alphabet’s $460 billion cloud backlog and Microsoft’s $37 billion AI run rate continue converting to revenue, capex holds and Baker’s redistribution thesis compounds. If token economics collapse before returns materialize, the same customers writing $50 billion checks to Jensen could pause.
For a longer look at how power, land, and grid buildout fit into this equation, our team’s AI infrastructure research maps the non-chip beneficiaries riding the same wave. For now, Baker’s point stands: the more open-source and vertically integrated models chip away at frontier lab margins, the more dollars land with the companies actually building the factories.
Contact [email protected] for any questions or corrections.
Top Gaining Stocks
Top Losing Stocks