Did Anthropic, Alphabet, and Broadcom Just Say Checkmate to Nvidia?

Anthropic said it will begin deploying next-generation TPUs from Alphabet and Broadcom in 2027, reinforcing the custom AI-chip opportunity and Broadcom's outlook. Broadcom said AI semiconductor revenue was $8.4 billion in Q1 FY2026, up 106% year over year, and its CEO expects custom AI chips alone to exceed $100 billion in annual revenue by end-2027. The article argues Nvidia remains a core supplier, with analysts still modeling 79% revenue growth next quarter and 71% for the full fiscal year.

Analysis

This is less a "winner-take-all" moment than a capacity-arbitrage event. The marginal beneficiary is Broadcom, because custom silicon monetizes hyperscaler demand that Nvidia cannot fully absorb on a near-term basis; the key second-order effect is that every incremental TPU deployment validates a multi-vendor procurement model, which actually preserves pricing discipline across the stack rather than creating a true displacement threat. Alphabet’s upside is more opaque but potentially more levered if TPU usage translates into higher cloud attach and lower inference cost, improving Google Cloud margin mix over the next 6-12 months.

The market is likely underestimating how little this changes Nvidia's near-term sell-through. When a customer adds TPU supply, it usually reflects constraint management, not a preference shift, and Nvidia’s ecosystem still anchors training workflows, tooling, and migration inertia. The real risk to Nvidia is not lost share today; it is that custom silicon gradually compresses its long-duration pricing power in 2027+ if hyperscalers prove they can routinize workloads onto proprietary chips without material performance loss.

Contrarian takeaway: the move is arguably more bullish for Amazon and Intel than headline readers realize. Amazon Trainium becomes more credible as part of a diversified AI procurement stack, and Intel remains an option-value beneficiary if the market starts paying for any non-Nvidia accelerator path as strategic insurance. The overdone part of the current debate is the binary "Nvidia vs. custom chips" framing; the underdone part is that the AI capex pie may expand faster when buyers can source compute from multiple vendors and reduce single-supplier bottlenecks.

Key Decisions for Investors

• Stay long AVGO on pullbacks over the next 3-6 months; this is the cleanest way to own custom AI silicon monetization with better visibility than Alphabet, with upside if custom chip revenue re-rates as a standalone growth engine.
• Initiate a relative-value long GOOGL / short NVDA pair for 6-12 months: GOOGL benefits from TPU-driven cloud margin expansion while NVDA already discounts near-term scarcity; use a tight stop if NVDA guidance re-accelerates materially.
• Add AMZN as a secondary beneficiary basket name over 6-12 months: diversified accelerator sourcing supports Trainium relevance and reduces the risk of hyperscaler compute being captured entirely by AVGO/GOOGL.
• Avoid shorting NVDA outright; if expressing skepticism, use a call spread overlay or a modest underweight versus semis, because the base case remains strong 1-2 quarters out and the stock can grind higher on continued backlog visibility.
• Watch for any evidence that TPU deployments are being booked into Cloud revenue at Alphabet; if so, that creates a catalyst for margin surprise and could justify a tactical long into the next earnings window.