Broadcom signs long-term deal to develop Google’s custom AI chips

Broadcom signed a long-term deal with Google to develop and supply custom AI chips and components for next-generation AI racks through 2031, and a separate agreement with Anthropic to provide about 3.5 GW of AI compute capacity on Google's processors starting in 2027. Financial terms were not disclosed; Broadcom shares rose ~3% in extended trading. Reuters highlights rising demand for Google TPUs as an alternative to Nvidia GPUs; Anthropic said its run-rate revenue surpassed $30B in 2026 (from ~$9B at end-2025) and is committing $50B to bolster U.S. computing infrastructure.

Analysis

Large, multi-year committed supply relationships change unit economics for ASIC/SoC vendors more than headlines imply: design-win revenue shifts cashflow from lumpy transactional sales to predictable, higher-margin engineering and recurring manufacturing revenue. Expect FCF conversion to improve materially (order-of-magnitude: a few hundred basis points over 18–36 months) as R&D recovery is amortized against longer-lived contract revenue and warranty/maintenance tails shrink.

The competitive response will be multi-modal. GPU incumbents are likely to defend share through pricing incentives for reserved GPU capacity, deeper software bundling, and preferential cloud discounts — a process that pressures spot pricing for GPU rentals within 6–18 months and compresses mid-cycle GPU ASPs. Conversely, vendors with long-term supply or design partnerships gain bargaining power with foundries and can lock in capacity, which creates a sustained barrier to entry for smaller ASIC challengers.

Key risks are execution (performance parity and software tooling), foundry throughput timing, and regulatory scrutiny around vertical supply relationships; any of these can defer commercial revenue by 6–24 months. Catalysts to track: vendor-level margin guidance shifts, reserved-capacity disclosures by large cloud customers, and material changes in GPU spot pricing. Contrarian angle: the market may overstate an immediate GPU displacement — software ecosystem and model optimization favor incumbents, so expect a gradual share reallocation rather than a binary collapse of GPU economics.

Key Decisions for Investors

• Overweight AVGO (2–3% net exposure, 12–24 month horizon): buy stock to capture improving FCF/margin profile from recurring ASIC design wins. Risk: integration/scale delays; target +25–40% upside vs ~-15–20% downside on execution failure.
• Paired position — Long AVGO / hedge with NVDA protection (net 1–2%): purchase AVGO shares and buy NVDA 12-month puts ~25–30% OTM (~cost 3–6% notional) to protect against an accelerated GPU share loss. Reward: asymmetric capture if ASIC adoption accelerates; cost-limited hedge against broader AI drawdown.
• Tactical GOOGL cloud exposure (1–2%, 9–12 month trade): buy GOOGL call spread to play improved cloud gross margins and differentiated hardware stack monetization. Expect 15–30% upside if cloud per-CPU economics improve; cap losses to option premium.
• Event-driven NVDA hedge/short (small, tactical): if GPU spot prices or cloud reservation cadence show clear declines over next 3–12 months, scale into NVDA short or buy deeper 6–12 month OTM puts. This is a risk-on directional hedge — high payoff if market re-prices GPU monopoly but limited size given the incumbent software moat.