Broadcom Inks Major AI Chip Deals With Google and Anthropic

Broadcom announced a new five-year deal to develop Google's next-generation TPUs and will supply networking and rack components to Google through 2031, while Anthropic secured access to 3.5 GW of TPU compute capacity starting in 2027. Broadcom reported fiscal Q1 FY2026 revenue of $19.3B (+29% YoY) and adjusted EPS of $2.05 (+28% YoY); CEO Hock Tan projected AI chip revenue above $100B in 2027 and Q1 AI semiconductor revenue was $8.4B. The agreements materially deepen Broadcom's role in AI data-center infrastructure and support upside to revenue and margins, implying favorable upside for Broadcom and related data-center suppliers.

Analysis

Broadcom’s deeper co‑design role with hyperscalers is not just another customer win — it signals a structural shift in data center procurement from commodity accelerators to vertically integrated subsystem suppliers. That transition raises switching costs (firmware, rack integration, power/cooling optimization) and converts episodic silicon sales into multi‑year subsystem annuities, which should support a higher multiple contingent on execution.

A move toward purpose‑built ASIC stacks has clear second‑order supply‑chain effects: increased demand for specialized nodes, advanced packaging and test capacity, and longer lead times that favor vendors with secured foundry/packaging slots. The consequence is a bifurcation of the accelerator market — flexible, high‑margin training GPUs remain valuable, but inference and rack‑level workloads will increasingly gravitate to lower‑cost, higher‑efficiency custom silicon and network ASICs, pressuring OEM gross margins that rely on generic GPU SKUs.

Principal risks are technological and macro. Rapid software pivots (sparsity, extreme quantization, or compiler breakthroughs) could restore the flexibility advantage of GPUs inside 12–36 months, while a capex pullback or foundry bottleneck could delay revenue conversion and leave multiple expansion vulnerable. Near‑term catalysts to watch: hyperscaler capex cadence, foundry utilization reports, and the next two quarters of data‑center bookings from major silicon vendors as they will indicate whether this is early innings or peak enthusiasm.

Key Decisions for Investors

• Buy AVGO via a 12–18 month call spread (buy 10–20% OTM calls / sell farther OTM calls) sized to 2–4% portfolio risk. Rationale: captures multi‑year design lock and subsystem annuity re‑rating while capping premium paid. Target 30–100% return if revenue conversion accelerates; max loss = premium paid. Monitor foundry utilization and Broadcom product booking cadence as exit signals.
• Pair trade — Long AVGO equity (or near‑ATM 12m calls) / Short NVDA via a 12m put spread (10–20% OTM) in a 1:0.5 size ratio. Rationale: hedge upside training risk for Nvidia while expressing conviction that inference share will migrate to ASICs. Time horizon 6–18 months. Risk: NVDA training demand surprise; cap pair size accordingly and buy tail protection (long NVDA calls) if downside breach levels trigger volatility spikes.
• Long Alphabet (GOOGL) cloud exposure through 9–15 month calls or modest equity add. Rationale: hyperscaler that secures vertically integrated ASIC+rack supply benefits from lower TCO offerings to customers and reduced unit economics volatility. Watch quarterly cloud gross margin and committed capacity disclosures for catalyst events.
• Tactical short/vol trade on NVDA — buy a 6–9 month put spread ~15% OTM to capture near‑term re‑pricing if ASIC adoption shows up in bookings. Rationale: defined downside with limited premium in a high‑vol underlying; this is a volatility play rather than a fundamental short. Limit position size to <2% portfolio and exit on sequential data‑center revenue misses or a spike in GPU bookings that invalidate the displacement thesis.