Broadcom vs Marvell: Custom AI Chip Valuation Showdown 2026

Now I have all the data needed. Writing the article now.

---

Broadcom leads the custom AI chip market with $10.8B in quarterly AI revenue, yet Marvell still trades at double its rival's forward earnings multiple.

Lead

The contest to design bespoke artificial intelligence silicon for the world's largest cloud providers has compressed into a two-company race — and produced one of the most pronounced semiconductor stock valuation divergences in the technology sector. Broadcom (AVGO), trading at $400.39 on July 12, 2026, with a market capitalization of $1.9 trillion, reported $10.8 billion in AI semiconductor revenue for its second fiscal quarter, a 143% year-over-year gain. Yet the market assigns Broadcom a forward price-to-earnings ratio of approximately 32x, while Marvell Technology (MRVL) — which generated $2.4 billion in total quarterly revenue — commands a forward P/E of roughly 66x, nearly double its larger rival. The Broadcom vs Marvell AI chips valuation gap has become the defining debate in tech sector investment positioning within semiconductors as 2026 approaches its second half.

Broadcom's Revenue Trajectory

Broadcom's AI semiconductor business has accelerated sequentially through fiscal 2026. The company posted $8.4 billion in Q1 FY2026 AI revenue, a 106% year-over-year increase, before reporting $10.8 billion in Q2, up 143%. Management has guided Q3 AI semiconductor revenue to approximately $16 billion, implying year-over-year growth in excess of 200%. Consolidated Q3 revenue is expected to reach $29.4 billion, up 84% from the same period a year earlier.

CEO Hock Tan has reiterated a target of exceeding $100 billion in AI semiconductor revenue for fiscal year 2027 — a figure that the Q3 guidance trajectory makes credible. The anchor for that forecast is a roster of six core hyperscaler customers whose long-term commitments provide multi-year revenue visibility. Google relies on Broadcom for its Tensor Processing Unit family, with a new next-generation TPU deal announced in April 2026. Meta deploys Broadcom-designed MTIA inference accelerators. Microsoft uses Broadcom silicon in its Maia AI program. Anthropic has expanded its Broadcom relationship to access approximately 3.5 gigawatts of compute capacity from 2027. OpenAI partnered with Broadcom to launch the Jalapeño Intelligence Processor. And in early July 2026, Apple committed more than $30 billion over the life of a deal extended through 2031, covering 15 billion U.S.-manufactured chips and triggering a $1.5 billion expansion of Broadcom's Fort Collins, Colorado facility.

Broadcom holds roughly 70% share of the custom AI chip ASIC co-design market, a position built over years of deep co-engineering relationships that carry substantial switching costs for customers.

Marvell's Case for a Premium

Marvell delivered fiscal year 2026 total revenue of $8.2 billion, growing more than 40% year over year, with its data center segment generating $6.1 billion — the growth engine of the entire business. The most recent quarterly report showed total revenue of $2.4 billion, up 27.6% year over year, with the data center unit at $1.83 billion, representing 76% of the company. The custom AI chip revenue run rate stands at approximately $1.5 billion annually across 18 cloud-provider design wins.

Marvell's design partnerships span the largest hyperscalers. Amazon Web Services built its Trainium AI training chip family with Marvell as co-designer; AWS has disclosed more than $225 billion in Trainium revenue commitments and is in early discussions to sell the chips externally to enterprise customers, a development that could materially expand Marvell's addressable volume. Microsoft's Maia AI accelerator also incorporates Marvell silicon. Meta has tapped Marvell for a new data processing unit. And in what may be the most strategically significant development of the competitive year, Google is reported to be in talks with Marvell to design new AI inference chips alongside its existing Broadcom TPU program — a potential break in the convention of single-partner ASIC relationships.

Management has raised forward guidance twice in 2026. Marvell now expects fiscal 2027 revenue of $11.5 billion, a 40% increase from the fiscal 2026 record, with data center the primary driver. Fiscal 2028 revenue is projected at $16.5 billion, implying another 43% gain, with the data center segment set to grow an additional 50% in that year.

The Valuation Argument

The semiconductor stock valuation gap between the two companies encodes a specific bet. Broadcom, at 32x forward earnings, is being priced as a large, diversified technology company with exceptional but maturing AI revenue growth. Marvell, at 66x, is being priced as a company in the early stages of a multi-year compounding cycle.

The Marvell premium is supported by a smaller base — incremental hyperscaler wins produce a proportionally larger revenue lift — and by the expanding customer list, which now includes all five of the world's largest cloud providers in some capacity. Marvell's stock gained 7% on July 9 on renewed enthusiasm for custom AI chip demand; the stock has returned roughly 187% year to date despite retreating from its all-time closing high of $316.43, set on June 4, 2026.

Broadcom's lower multiple reflects both its scale and its business composition. Software and networking infrastructure revenue dilute the AI-semiconductor growth rate at the consolidated level even as the chip division's absolute numbers are extraordinary. For tech sector investment mandates that require a cleaner AI silicon pure-play at the hyperscaler level, Broadcom's blended model presents a different risk-return profile than Marvell's more concentrated data center exposure.

Together, the two companies account for roughly 95% of the ASIC co-design market. Custom silicon now represents 27.8% of total AI server shipments in 2026, and that share is projected to rise as hyperscalers seek to escape the gross margin economics of Nvidia's GPU platforms.

Outlook