Thinking Machines Lab inks massive compute deal with Nvidia

Thinking Machines Lab entered a multi-year strategic partnership with Nvidia that includes deploying at least 1 gigawatt of Nvidia Vera Rubin systems starting in 2027 and an undisclosed strategic Nvidia investment. Thinking Machines has raised more than $2 billion since its Feb 2025 founding and is valued at over $12 billion, though it has yet to release products and has seen several co-founder departures. The deal includes joint development of training and serving systems on Nvidia architecture and reinforces strong demand for AI compute amid Nvidia’s $3–4 trillion infrastructure forecast.

Analysis

At the architecture level, the current leader retains a two-sided moat: hardware performance advantages translate directly into software optimization investments and customer-specific stacks that raise switching costs for large training customers. That lock-in can convert into long-dated revenue visibility via reserved deployments and higher attach-rates for software/support, concentrating margin capture at the top of the stack while compressing opportunities for price-led competition in the near term.

The build-out of hyperscale training capacity creates tangible upstream demand outside GPUs themselves — power distribution, chillers, high-bandwidth interconnects, and rack-level systems become multi-year revenue streams and potential choke points. Expect increased activity in GPU leasing/resale markets and a bigger role for specialist cloud/co-location contracts as customers prefer opex over capex for risky, long-lead AI projects; these dynamics can shift who captures margin from hardware vendors to operators.

Key risks are execution and timeline slippage: product maturity, customer readiness, and talent turnover can easily push recognition beyond a 12–36 month horizon, while a credible alternative architecture (or in-house hyperscaler accelerator) would materially undercut premium pricing. Watch for early contract renegotiations, rising spot cloud capacity, or regulatory scrutiny around exclusive arrangements — any of these would compress the premium multiple assigned to the incumbent very quickly.

Key Decisions for Investors

• NVDA — Buy a directional, defined-risk exposure: enter a 9–18 month call spread (buy longer-dated call, sell a higher strike) sized 4–6% of NAV to capture continued multiple expansion from reserved deployments; set a hard stop at a 20% drawdown on the spread premium.
• Pair trade — Long NVDA / Short AMD (60/40 notional, beta-neutral): 12–24 month horizon to express top-end share consolidation while hedging broad semiconductor cyclicality; cut AMD short if it rallies >25% on beat-and-raise print or if NVDA misses materially.
• ORCL — Buy 12-month calls (or buy shares) to play the win for specialized cloud/co-location providers and middleware monetization as customers seek managed stacks; target 25–40% upside, limit premium risk to <2% NAV.
• Volatility structure — Implement a calendar/diagonal: sell near-term NVDA calls around high-IV events (earnings/announcements) and buy longer-dated calls to finance, collecting term premium while keeping upside optionality; use this tactically over weeks–months and size to keep max loss <3% NAV.