Nvidia is so spooked by Google’s sudden AI comeback that it’s posting on X to defend itself

Nvidia publicly rebutted a report that Google is pitching its TPUs to major customers like Meta after its stock fell more than 2.5% on the news, asserting its GPUs and new Blackwell architecture remain a generation ahead and the only platform that can run every AI model everywhere. The story gained traction as Google’s Gemini 3—trained on TPUs—received strong reviews, raising the prospect that large customers could shift some AI infrastructure away from Nvidia. Separately, Nvidia circulated a seven-page memo to analysts pushing back on Michael Burry’s accusations about stock-based compensation and accounting, escalating an investor feud that could influence sentiment and positioning among funds.

Analysis

Market structure: Google’s TPU traction is a targeted threat — winners are GOOGL (cloud/TPU revenue), META (lower marginal cost if on-prem TPUs reduce GPU cloud spend), and cloud service integrators; losers are marginal GPU cloud spots and some fabless GPU-dependent startups. If TPUs take even 5–15% share of specific training workloads over 12–36 months, Nvidia’s GPU pricing power for those workloads will soften, but GPUs remain dominant (>90% today) for general-purpose workloads.

Risk assessment: Tail risks include a sudden Meta public switch to TPUs (immediate 5–15% re-rating for NVDA), antitrust or export restrictions that fragment supplier access, or a software lock-in failure (CUDA losing developer mindshare). Time horizons: immediate (days) for sentiment swings and IV spikes, short-term (weeks–months) for customer announcements and Qs, long-term (quarters–years) for share reallocation and ecosystem lock-in; key hidden dependency is Nvidia’s software/SDK moat (CUDA, cuDNN).

Trade implications: Tactical long GOOGL/cloud exposure and hedged, reduced NVDA exposure is warranted: buy GOOGL exposure to capture TPU monetization while protecting NVDA exposure with put spreads rather than outright shorts. Use pair trades (long cloud vendor vs underweight NVDA) and volatility plays: buy 3–6 month protective NVDA put spreads sized to hedge 30–50% of existing NVDA delta; add GOOGL stock or 9–12 month call spreads on pullbacks >5%.

Contrarian angles: Consensus understates switching costs — CUDA/software ecosystem and third-party optimizers meaningfully slow adoption, so structural erosion of Nvidia revenue is likely gradual, not instantaneous. The market reaction (NVDA down ~2–3% on the report) is likely overdone in the short run; if Nvidia’s Blackwell demos or enterprise deals land in the next 60–90 days, rapid mean reversion is probable, creating short-term alpha opportunities.

Key Decisions for Investors

• Establish a 2–3% portfolio long in Alphabet (GOOGL/GOOG) over the next 2 weeks—prefer stock or 9–12 month call debit spreads—to capture TPU/cloud monetization; target 12-month upside ~20%, place stop-loss at -12% absolute or re-evaluate on quarterly cloud revenue miss.
• If NVDA position >4% of portfolio, reduce to 2–3% immediately and buy 3-month put spreads to hedge 50% of remaining delta (e.g., buy 10% OTM puts and sell 5% OTM puts sized to cost-fund); add incremental puts if NVDA falls >10% within 30 days.
• Implement a pair trade: long META (1.5–2% weight) vs short NVDA (1% weight, or buy inverse via options) for a 3–9 month horizon. Close if META underperforms NVDA by >15% over any 60-day window or if Meta confirms multi-quarter TPU deployment (then increase GOOGL and trim NVDA further).
• Monitor specific catalysts over the next 30–90 days: (a) Meta public infra announcements and capex guidance (if Meta confirms TPUs in-datacenter, increase GOOGL by +2%); (b) Nvidia enterprise win disclosures or Blackwell benchmarks (if NVDA posts outsized wins, buy-back NVDA up to prior weight within 2 weeks).