Broadcom Just Named Its Next Customer to Pass Google: Anthropic. Here's Why That Matters More Than the Earnings Beat.

Broadcom reported fiscal Q3 adjusted EPS of $3.32 and revenue of $29.59B, ahead of consensus estimates of $3.24 and $29.36B, but shares fell after its $34.8B Q4 revenue outlook trailed the $35.03B consensus. Management expects Anthropic to become its largest XPU customer in 2027, with Ironwood TPU demand projected to rise from 1GW in 2026 to 5GW in 2027; OpenAI Jalapeno deployments are projected to grow from 1.3GW in 2027 to 5GW in 2028. Broadcom projected adjusted EPS above $30 in fiscal 2028 versus a $25 consensus estimate, supported by expanding custom AI-chip and networking demand.

Analysis

The key investable implication is not a single-customer ramp but a validation of Broadcom's custom-silicon model across multiple frontier labs. This expands AVGO's addressable market beyond one hyperscaler procurement cycle and creates a multi-year ASIC design-win backlog that is typically stickier than merchant GPU demand once software, networking, and data-center power architecture are qualified. The offset is concentration risk migrating from Alphabet to venture-funded AI labs whose capex ultimately depends on external financing and monetization rather than internally generated cash flow.

Near term, AVGO's price action indicates the market is discounting a high bar for quarterly delivery, not merely aggregate AI demand. That makes the stock vulnerable over days to weeks to any evidence that deployments are power-constrained, delayed by advanced packaging capacity, or recognized later than investors expect. Over the next 1-3 months, confirmation of shipment cadence and incremental custom-chip revenue should matter more than a modest top-line guide variance; failure to convert announced gigawatt-scale commitments into booked backlog would compress the AI premium quickly.

Custom XPUs are a credible competitive pressure on NVDA at the margin, particularly for inference and stable, repeatable workloads where customers can justify longer design cycles in exchange for lower cost per token. The contrarian view is that this is not necessarily zero-sum: NVDA remains advantaged in training flexibility and ecosystem lock-in, while AVGO's networking franchise can monetize GPU clusters as well. The more material risk to AVGO is that bespoke compute lowers AI-lab unit costs faster than end-demand rises, causing customers to optimize utilization rather than perpetually add capacity.

For the 6-18 month thesis, monitor whether Anthropic and OpenAI secure durable funding, enterprise revenue, and power commitments sufficient to support their deployment plans. A reduction in either lab's capex ambition would be disproportionately damaging because the market is likely capitalizing projected custom-silicon revenue well before full-volume shipment. Falsification signals are a downward revision to AI semiconductor revenue guidance, delayed volume ramps, or a material increase in customer-specific inventory or receivables.

Key Decisions for Investors

• Accumulate AVGO only on post-results weakness rather than chase a guide-driven rebound; target a 6-12 month holding period. The upside case is multiple expansion from independently confirmed ASIC backlog conversion, while the principal downside is a customer-capex reset that exposes concentration risk.
• Establish a modest long AVGO / short NVDA pair over a 3-6 month horizon only if AVGO discloses tangible custom-XPU shipment or backlog milestones while NVDA's forward demand commentary remains dependent on generalized hyperscaler capex. Use a tight risk limit: cover the NVDA short if NVDA reports accelerating non-hyperscaler demand or AVGO delays custom-silicon volume recognition.
• Maintain GOOG as a watch item rather than infer a negative read-through from customer-mix changes. Custom silicon can improve Alphabet's internal compute economics and cloud competitiveness; reassess if Google Cloud margin expansion or TPU external-adoption disclosures demonstrate that lower compute cost is being retained rather than passed through.
• Set alerts around frontier-lab financing, IPO documentation, power procurement, and advanced-packaging lead times over the next quarter. These are the missing verification points required before underwriting the advertised multi-year deployment curve as revenue rather than non-binding capacity intent.