Nvidia and Broadcom split $750B AI capex wave with different chip strategies

Nvidia Corp. and Broadcom Inc. both posted record AI semiconductor quarters, yet their paths to capturing the $750 billion infrastructure buildout are diverging in ways that matter for investors.

"Nvidia sells the platform hyperscalers default to, while Broadcom sells the custom silicon those same customers use when they want to design around Nvidia," said Stacy Rasgon, an analyst at Bernstein.

Nvidia's Q1 FY2027 revenue hit $81.6 billion, up 85% year over year, with data center alone contributing $75.2 billion. The company guided Q2 to $91.0 billion. Broadcom's Q2 FY2026 revenue reached $22.2 billion, up 48%, with AI semiconductor revenue of $10.8 billion growing 143%. Chief Executive Officer Hock Tan guided Q3 AI revenue to $16.0 billion, implying growth exceeding 200%.

The divergence comes down to architecture. Nvidia sells merchant GPUs paired with its CUDA software platform, capturing 75% gross margins. Broadcom designs custom ASICs (application-specific integrated circuits) for a handful of hyperscalers, trading margin breadth for customer lock-in. Both models produce record revenue, but they carry different risk profiles as the AI capex cycle matures.

Merchant vs. Custom — Two Ways to Win

Nvidia's Blackwell platform and upcoming Vera Rubin architecture target the broadest possible market: cloud providers, sovereign AI projects, enterprise deployments and automotive customers. The company's $80 billion new buyback authorization and $119 billion in supply commitments suggest management sees no demand ceiling. The risk is circularity. Nvidia has committed more than $40 billion to AI equity deals in 2026, including a $30 billion stake in OpenAI, and is in talks to backstop as much as $250 billion in financing for OpenAI's data center leases, according to the Wall Street Journal. The Bank for International Settlements warned in June that opaque private arrangements among chip makers, hyperscalers and AI labs create financing structures that could amplify a credit correction.

Broadcom's model is narrower by design. The company targets three to four hyperscalers building their own accelerators, with a stated goal of exceeding $100 billion in AI semiconductor revenue by 2027. VMware contributed $7.18 billion in recurring software revenue, providing a cushion Nvidia lacks. Broadcom's adjusted EBITDA margin of roughly 68% reflects the efficiency of a custom silicon model where design costs are amortized across committed volumes from a concentrated customer base.

The Valuation Question

Nvidia trades at roughly 35x forward earnings, reflecting the market's willingness to pay for its 75% gross margins and dominant CUDA software platform. Broadcom trades at about 20x forward earnings with a $0.65 quarterly dividend, appealing to income-oriented investors who want AI exposure with less valuation risk. The gap in multiples captures a fundamental disagreement about whether Nvidia's platform moat justifies the premium or whether Broadcom's custom ASIC model will capture a larger share of incremental AI compute spending as hyperscalers diversify away from merchant silicon.

The next test comes when both companies report their next quarterly results. Nvidia's $91.0 billion Q2 guidance, excluding any China data center revenue, sets a high bar. Broadcom's $16.0 billion AI revenue target will show whether custom ASIC growth can sustain its trajectory. For investors, the choice is not binary — both companies benefit from the same secular trend — but the risk profiles differ. Nvidia offers higher margin and platform lock-in with greater exposure to a potential financing unwind. Broadcom offers lower valuation and software diversification with concentration risk in a handful of hyperscaler customers.

This article is for informational purposes only and does not constitute investment advice.