Catalyst Alert: Alphabet Just Made Broadcom's Stock a Must-Buy

Broadcom extended its TPU partnership with Alphabet and added a major new demand driver: Anthropic access to 3.5 gigawatts of TPUs starting in 2027, with TPUs also supplied directly by Broadcom. Anthropic already has a $21 billion TPU order this year, and Broadcom reaffirmed that TPU sales should carry gross margins similar to the rest of its semiconductor business. The deal reduces a key bear case around losing Alphabet's TPU business and supports Broadcom's $100 billion custom AI chip revenue target for fiscal 2027.

Analysis

The incremental bull case is not just “more TPU demand,” but that Broadcom is converting what looked like a one-customer concentration risk into a multi-year capacity lock-in business with better visibility and pricing power. If large-scale AI inference shifts toward custom accelerators, the value chain tilts from merchant GPU dominance toward ASIC design, packaging, and supply orchestration — areas where Broadcom is structurally stronger than a pure merchant silicon model. That lowers the probability of a near-term multiple compression event tied to customer insourcing fears and supports a longer duration cash flow profile.

The second-order read-through is more negative for NVDA than the headline suggests, even if not in a direct replacement sense. Every major AI platform operator that commits to custom silicon is signaling a desire to reduce dependence on one GPU vendor, which could cap upside in incremental share gains over the next 12-24 months, especially for inference workloads where economics matter more than flexibility. The likely loser is not “AI spending” but the composition of that spending: more capex migrates into non-NVIDIA accelerators, where Broadcom monetizes design wins and systems integration rather than being forced into price competition.

The key risk is that this remains a future-demand story until 2027, and the market may be ahead of fundamentals if AI capex pauses or Anthropic’s ramp proves less elastic than implied. A meaningful reversal would come from TPU volumes slipping, a rival ASIC supplier winning a marquee platform, or evidence that custom chips cannibalize cloud economics faster than expected. Near term, the catalyst path is cleaner for AVGO than for GOOG: each additional design win extends the duration of the re-rate, while any slowdown in hyperscaler capex would hit the entire group, with NVDA most exposed to sentiment compression.

Consensus may still be underestimating how sticky Broadcom becomes once it owns the design-to-deployment workflow for custom silicon. The market often prices semis as if customer concentration is a binary risk, but in practice the more embedded the engineering and supply chain, the more the customer becomes locked in by switching costs and time-to-market constraints. That suggests the bear case around Broadcom losing TPU share was not just wrong, but structurally misframed.

Key Decisions for Investors

• Long AVGO into the next 1-3 months on any post-news consolidation; the thesis is a multiple expansion from reduced customer-loss risk, with downside limited unless AI capex rolls over broadly.
• Initiate a relative value long AVGO / short NVDA pair for 3-6 months; the trade expresses custom-ASIC monetization versus merchant-GPU dependency, with AVGO benefiting from design-in stickiness and NVDA more exposed to mix shift.
• Buy AVGO call spreads 6-12 months out to capture a continuation re-rate without paying full premium; best if the market begins to model upside to 2027 custom AI revenue guidance.
• Use GOOG as a lower-beta secondary long only if you want exposure to TPU ecosystem optionality; the direct beneficiary is modest, so keep size smaller than AVGO and avoid treating it as the primary expression.
• Avoid chasing NVDA on broad AI headlines over the next several weeks; the risk/reward worsens if more hyperscaler customers follow the custom-chip path and incremental AI capex shifts away from merchant GPUs.