Meta in talks to spend billions on Google’s chips, The Information reports

Meta is reportedly in talks to spend billions to use Google’s TPU AI chips in its data centers from 2027 and to rent TPUs from Google Cloud as soon as next year, a move that could diversify Meta’s accelerator supply away from Nvidia. Google is pitching TPUs as a cheaper, more security-focused alternative and has discussed targeting roughly 10% of Nvidia’s revenue with its TPU business; the report notes Meta has committed $600 billion to U.S. infrastructure and AI data centers over the next three years. Reuters could not verify the report and the companies did not comment, leaving timing and scale uncertain.

Analysis

Market structure: This move would incrementally weaken Nvidia’s near-monopoly on accelerator revenue and pricing power, creating a multi-vendor equilibrium where Google can realistically target ~5–10% of high-end accelerator spend within 2–4 years. Expect margin compression pressure on GPU pricing in spot/inventory markets and increased price competition for long-term supply contracts, particularly for inference-optimized workloads that TPU economics favor.

Risk assessment: Key tail risks include underperformance of TPUs on Meta’s training workloads (technical mismatch) and heightened antitrust scrutiny if hyperscalers consolidate cross-licensing, any of which could swing markets ±20% for NVDA/GOOGL in 3–12 months. Immediate effects will show up in option vol and sentiment (days–weeks), procurement and contractual shifts will play out over 12–36 months, while the strategic ecosystem lock-in (CUDA vs TPU) is a multi-year battleground.

Trade implications: Near-term tradeable items are differential volatility and spread exposures: buy GOOG call exposure with 9–18 month tenor and hedge NVDA risk with limited-cost put spreads over 3–6 months; pair trades (long cloud infra, short pure-play GPU) capture structural share reallocation while capping downside. Also reweight energy/utilities exposure modestly (1–2%) to capture incremental datacenter power demand over 2–3 years.

Contrarian angles: The market may underappreciate CUDA’s software moat — historical CPU/GPU platform shifts took multiple years and required parity-plus performance and developer buy-in; hence NVDA downside may be overdone absent concrete share losses >5–10%. Conversely, a confirmed large Meta TPU deal could catalyze rapid re-rating of GOOG vs NVDA in 30–90 days, so size and option tenor should be asymmetric to reflect binary outcomes.

Key Decisions for Investors

• Establish a 3% long position in GOOGL (class A or C) with a 12–24 month horizon; complement with a 12–18 month 20% OTM call spread (pay limited premium) to lever upside if Google announces TPU commercial traction within 90 days. Add to position on pullback >10% from entry; trim if TPU adoption announcements fail in next 6 months.
• Buy a protective NVDA 3–6 month 15%/30% OTM put spread sized at 1.5% of portfolio to hedge semiconductor cyclicity and potential share loss; increase notional to 3% if Meta formally confirms TPU contract or Google publishes a >$5B revenue target for TPUs. Close spread after 20% realized P&L or at 6 months.
• Initiate a pair trade: long 1.5–2% META equity vs short 1.5% NVDA (equal-dollar) for 6–12 months to capture diversification benefit to Meta and potential NVDA re-rating. Scale long-META by +50% on confirmation of TPU rental contracts for 2025–2027, exit short-NVDA if NVDA reports >10% upside to consensus guidance.
• Rotate 2–3% from pure semiconductor hardware exposure into cloud-infrastructure and power plays: add +1–2% GOOGL and +1% to utilities/infrastructure ETF exposure (e.g., XLU or IGF) to capture datacenter build power demand over 2–3 years; reduce direct semiconductor cyclicality by the same amount.