Marvell Technology told investors this week that its custom-silicon business for AI data centers is on track to reach approximately $30 billion in annual revenue by fiscal 2031, a figure it simultaneously raised its fiscal 2029 custom outlook from more than $10 billion to more than $12 billion. The number, confirmed in Marvell's own Investor Day materials, puts a concrete price on the portion of the AI chip market Nvidia does not control — and on a segment Marvell says now spans all four major U.S. hyperscalers.

CEO Matt Murphy framed the outlook in direct terms: "We have increased our fiscal 2029 revenue outlook from more than $10 billion to more than $12 billion, and we are on a path to approximately $30 billion by fiscal 2031 at the midpoint of our long-term target range." That is a company-stated target, not booked revenue, and it covers the custom business specifically rather than Marvell's total revenue.

The $30 billion headline obscures a structural split. Murphy told attendees the fiscal 2031 figure should be seen as "largely balanced between the existing XPU programs we have coming and then XPU attach" — the custom accelerators themselves versus the networking, memory, storage and connectivity silicon that surrounds them in a rack. His qualifier, "we still don't know exactly," signals the mix is a planning assumption, not a contracted breakdown.

Is Marvell actually competing with Nvidia?

Not directly — and the evidence points the other way. In March 2026, Nvidia and Marvell announced a strategic partnership built around Nvidia's NVLink Fusion platform, under which Marvell supplies custom XPUs and networking silicon designed to work with Nvidia's interconnect architecture rather than against it. Nvidia also made a direct investment in Marvell, Reuters reported.

Industry analyst Jacob Bourne of eMarketer captured the dynamic: "So Nvidia can maintain its dominant position while also expanding the scope and utility of the AI semiconductor sector." Marvell's biggest growth story is partly a growth story for Nvidia's own ecosystem, not a replacement for it.

Why hyperscalers want their own chips

The demand driver is straightforward. Custom accelerators tuned to specific workloads such as ad-ranking inference or large-model training can cut power per unit of compute, avoid merchant-chip margins, and reduce dependence on a single supplier's allocation decisions. Google's TPU line has reportedly priced its newest generation well below comparable Nvidia silicon, while Amazon has structured multibillion-dollar arrangements around its own accelerator roadmap.

Marvell does not build its own branded AI chip. It positions itself as the design and manufacturing partner that lets any hyperscaler build a TPU-style or Trainium-style chip of its own.

What is confirmed versus projected

• Fiscal 2029: more than $12 billion custom revenue — a company-stated outlook, raised in the same presentation.
• Fiscal 2031: approximately $30 billion — the midpoint of a long-term target range, not realized revenue.
• Revenue mix: "largely balanced" between XPU products and XPU attach — a company characterization; Murphy said the exact split is undetermined.
• Customers: Marvell claims shipments to all four major U.S. hyperscalers, but has not named all four in disclosed materials — a company claim, not a confirmed customer list.

Broadcom operates the larger of the two independent custom-AI-chip franchises, but the two companies have tended to win different hyperscaler customers rather than bid for the same programs. AMD, meanwhile, forecasts AI chip demand will outpace supply well into 2028, a backdrop that benefits every credible designer, custom or merchant.

What the forecast leaves open

Murphy also made a sweeping infrastructure claim at Investor Day: "Every model so far that's ever been created has been trained using Marvell connectivity." That is a company claim about supply-chain ubiquity, not a measured share statistic.

The trade-offs for hyperscalers are real. A custom chip that misses its design target, or arrives a generation late relative to Nvidia's cadence, can strand billions in committed fleet spending — part of why Google and Amazon keep buying large volumes of Nvidia GPUs alongside their custom fleets.

With five fiscal years of runway, the $30 billion target will likely be revised at least once more, up or down, before fiscal 2031 arrives — and Wall Street will be watching the gap between the confirmed fiscal 2029 figure and the 2031 midpoint as the clearest test of whether custom silicon becomes a distinct budget line or stays a footnote to the merchant-GPU market.