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, expanding its custom AI chip partnership while still using Nvidia GPUs and Amazon Trainium. Broadcom said AI semiconductor revenue reached $8.4 billion in Q1 FY2026, up 106% year over year, and management expects custom AI chips to exceed $100 billion in annual revenue by the end of 2027. The piece is broadly constructive for Broadcom and Alphabet, but it does not indicate Nvidia is losing meaningful share given its capacity is still largely sold out through 2027.

Analysis

The key market implication is not a wholesale shift away from Nvidia, but a broadening of the AI compute stack into a three-supplier regime. That matters because once model developers optimize across multiple chip architectures, pricing power starts to migrate from the incumbent hardware layer to the buyers that can arbitrage supply, latency, and inference economics. In other words, the announcement is more important for procurement leverage than for immediate share loss at NVDA.

Broadcom is the cleanest incremental beneficiary because custom silicon monetizes a longer runway of design wins with better visibility than merchant GPU demand. The second-order effect is that AVGO’s AI story should de-risk from a single customer narrative into a multi-hyperscaler repeat-order thesis, which typically supports a higher multiple as revenue durability improves. Alphabet gets a different benefit: TPU adoption can quietly deepen cloud stickiness and raise internal utilization of its own infrastructure, which could show up more in margin expansion than in obvious topline beats.

The bear case for Nvidia is more about duration than direction. If large model labs keep adding custom capacity, GPU demand growth can normalize faster than consensus expects, but the real inflection likely sits 12-24 months out rather than in the next quarter. Near term, the data still argues that Nvidia remains supply-constrained, so any selloff on this headline would likely be a positioning event, not a fundamental break.

The contrarian miss is that a balanced multi-chip strategy is actually bullish for AI capex overall: it lowers dependence on any one vendor and reduces the risk of compute bottlenecks slowing model training. That could extend the total addressable spend cycle across 2026-2027, with the biggest loser being whoever is priced for monopoly economics rather than ecosystem share.

Key Decisions for Investors

• Go long AVGO on any post-news pullback over the next 1-3 weeks; thesis is sustained custom-silicon mix shift and higher visibility into 2027 revenue, with upside if hyperscaler design wins compound faster than expected.
• Maintain a tactical long GOOGL vs. short NVDA pair for 3-6 months only if the market starts pricing TPU adoption as cloud margin expansion rather than pure capex; risk is NVDA remains supply-tight and outgrows the narrative.
• Do not short NVDA outright on this headline; the cleaner expression is a bearish call spread 6-12 months out, because the catalyst for actual share loss is still a later-cycle procurement reset, not current demand weakness.
• For event-driven accounts, buy AVGO Jan-2027 calls financed by selling nearer-dated upside; asymmetry favors long-dated optionality if custom chip revenue inflects toward management’s revenue target trajectory.
• Monitor GOOGL Cloud gross margin and capex commentary over the next two quarters; if TPU monetization lifts cloud margins without accelerating capex, increase exposure, as that would imply operating leverage rather than just revenue substitution.