Anthropic Just Announced Huge News for Alphabet and Broadcom

Anthropic said it will use multiple gigawatts of next-generation TPUs starting in 2027, reinforcing demand for Alphabet’s Google Cloud TPU revenue and Broadcom’s custom AI chip business. Broadcom said its AI semiconductor division generated $8.4 billion last quarter, up 106% year over year, and expects custom AI chip revenue to exceed $100 billion annually by the end of 2027. Alphabet’s Google Cloud revenue rose 48% year over year last quarter, signaling continued strong AI-driven growth for both companies.

Analysis

The market is underpricing how quickly TPU adoption can shift AI capex from a single-vendor architecture to a two-layer stack: training economics remain driven by CUDA dominance, but inference and large-scale reserved workloads are now becoming a negotiated procurement market. That matters because once a frontier lab locks multi-year capacity, the value migrates from one-off chip sales to contracted utilization, which improves revenue visibility for the silicon vendor and the cloud host while compressing pricing power for the incumbent GPU leader at the margin.

For AVGO, the second-order win is not just chip volume; it is the software-defined, custom-silicon relationship that becomes stickier as workloads are ported. If multiple hyperscale customers follow, Broadcom’s AI mix can re-rate from "growth segment" to "platform annuity," and the stock could deserve a higher multiple than a cyclical semiconductor name. The hidden risk is concentration: any delay in customer rollout, power delivery, or packaging capacity would hit sentiment sharply because expectations are now anchored to a 2027 step-up rather than near-term revenue.

For GOOGL, the TPU story is more strategic than monetization. TPU-based compute increases cloud differentiation and creates a pull-through effect for storage, networking, and higher-margin managed services, but it also signals that Alphabet is willing to subsidize AI infrastructure to win workload share. That can depress near-term margin optics even as it strengthens competitive positioning; the key question is whether cloud growth can keep outpacing capex normalization over the next 2-4 quarters.

NVDA is not losing the AI trade, but its monopoly narrative is fading into a premium-vs-good-enough debate. The consensus still assumes every incremental AI dollar goes to GPUs, yet a growing share of inference workloads may increasingly migrate to custom silicon where latency, power efficiency, and negotiated cost matter more than absolute flexibility. The contrarian setup is that the "winner-takes-most" view is now too simplistic: the real market share fight may be in cloud economics, not model performance.

Key Decisions for Investors

• Stay long AVGO vs. NVDA via a 3-6 month pair trade; thesis is relative multiple expansion for custom-silicon annuity cash flows versus increasing competitive share leakage in non-core workloads. Use a tight stop if NVDA re-accelerates cloud demand commentary.
• Add to GOOGL on any post-capex selloff over the next 1-2 quarters; upside is cloud re-rating as TPU utilization improves, while downside is mainly margin compression if AI infrastructure spend outruns monetization.
• Sell NVDA upside calls 6-9 months out against core longs to finance exposure elsewhere; this is a way to express view that consensus growth assumptions are too linear while still retaining upside if GPUs remain dominant longer than expected.
• For more aggressive exposure, buy AVGO Jan-2027 calls or call spreads; the long-dated structure matches the 2027 TPU revenue inflection and limits theta risk if deployment timing slips.
• Avoid chasing broad AI beta here; prefer quality relative-value expressions, because the next leg is likely a share-shift trade, not an indiscriminate multiple expansion trade.