AI startup Thinking Machines clinches capital and a major chip supply deal from Nvidia

Thinking Machines struck a multi-year partnership with Nvidia to receive a significant investment and procure at least 1 gigawatt of Nvidia's next‑generation processors, with Vera Rubin systems to be deployed starting early next year. Terms were undisclosed; Thinking Machines raised about $2 billion in a seed round at a $12 billion valuation and is seeking a larger round that could value it in the tens of billions. The deal materially expands the startup's training capacity and underscores Nvidia's growing role as both supplier and financier (Nvidia has invested ~$30 billion in OpenAI and $10 billion in Anthropic), though recent co‑founder departures and bubble-era comparisons introduce some downside risk.

Analysis

Nvidia’s expanding role as both supplier and financier creates a self-reinforcing demand loop for its accelerators that amplifies revenue visibility but also concentrates counterparty and reputational risk. That loop increases pricing power for Nvidia in the near term (allowing above-market ASPs and tighter supply discipline) while shifting margin risk onto the financed customers — a structure that can compress volatility in NVDA shares when capital markets are liquid and amplify downside when startup funding or model training economics deteriorate.

The realistic bottlenecks are not just wafer starts but system-level constraints: HBM memory, datacenter power provisioning, and high-performance interconnects. These are multi-quarter to multi-year frictions that create discrete timing mismatches between chip availability and customers’ ability to deploy at scale, producing episodic demand shocks for Nvidia and ancillary beneficiaries (switches, power gear, facility builders).

Competitors and hyperscalers have predictable responses: accelerate in-house silicon and diversify vendors, or push for bundled commercial terms (compute+financing). That dynamic increases the probability of regulatory scrutiny and commercial pushback over the next 12–36 months, which is the primary non-technical threat to the current profit-capture model.

For portfolio construction, treat Nvidia exposure as a convex event bet tied to the pace of large-scale model training deployments and capital markets liquidity for AI startups. Key catalysts to watch are quarterly guidance vs. buildout timelines, HBM pricing trends, and any public regulatory engagement regarding supplier-financier conflicts; adverse moves in these three can produce >20% drawdowns within months, while sustained training demand continuation supports upside measured in multiple points of datacenter TAM capture over 12–24 months.

Key Decisions for Investors

• Long NVDA equity or 12–24 month call spreads (e.g., buy 12m LEAP calls / sell higher strike to fund) — target size 4–6% net exposure; entry on pullback of 8–12% or on beat-and-raise confirmation. R/R: asymmetric upside if training demand stays robust; downside limited to premium paid for spreads.
• Pair trade: long NVDA / short AMD (equal notional) for 6–12 months to isolate Nvidia-specific financing + ecosystem moat. Rationale: capture pricing+ecosystem premium while hedging broad GPU cycle risk. Risk: AMD share gains or a broad semiconductor rally could negate hedge.
• Hedge catalyst risk with event-dated protection: buy 6–9 month NVDA puts sized to cover 25–30% of position ahead of major product ramps or regulatory hearings. Cost is insurance against fast drawdowns from supply/regulatory shocks.
• Opportunistic long positions in datacenter power/interconnect suppliers (selective, e.g., switch/PSU vendors) for 9–18 months — these lag Nvidia in the build cycle and can outperform on deployment acceleration. Size small (1–2%) and trim on 30–50% outperformance as deployments confirm.