Marvell says the business of connecting AI chips to each other could be worth $37 billion. That figure, laid out in a MarketBeat report on the company's investor communications, reframes an unglamorous corner of the silicon stack — the wiring, PHYs and switch silicon that link GPUs and accelerators inside data centers — as a market of its own.

The number landed with investors. Two stocks named alongside Marvell in the report rallied on the claim, according to MarketBeat, a sign that traders are treating interconnect not as an accessory market but as a direct way to price the AI buildout.

Why interconnect, why now?

AI clusters don't run on compute alone. Thousands of accelerators have to move terabytes of data between chips, and the silicon that manages those links — custom interconnect products, high-speed electrical and optical PHYs, switching fabrics — sits squarely in Marvell's portfolio. The company's $37 billion figure represents its own sizing of that opportunity.

The distinction matters for how the market reads the number. This is not a confirmed revenue figure from shipped product. It's a company estimate of a future addressable market, presented to investors as part of Marvell's AI narrative. Analysts and investors routinely treat such projections as roadmaps rather than booked business, and the MarketBeat report reflects that framing: the figure is what Marvell "says" the market could become, not what it bills today.

Who else moved?

According to MarketBeat, two additional stocks rallied in sympathy with Marvell's projection. The rally itself is the concrete market reaction — investors repriced the interconnect theme within a single trading cycle after the figure went public.

That kind of move is characteristic of the current AI hardware trade, where a single company's market sizing can lift an entire supply chain. Interconnect suppliers sit in a chokepoint position: no matter whose accelerators win, the chips still have to be linked.

What does the $37 billion figure actually rest on?

It rests on Marvell's own arithmetic about AI infrastructure spending, communicated through investor channels and reported by MarketBeat. The report did not break the figure into segments by process node, product family or customer, so the composition — how much is custom silicon versus merchant interconnect, electrical versus optical — remains the company's internal view.

For context on how the market prices such claims: AI infrastructure spending has scaled fast enough that interconnect, historically a single-digit percentage of a system's silicon cost, now represents a spend line large enough for a major vendor to anchor its investor story on it. Marvell's decision to lead with a $37 billion number signals where the company believes its data center revenue mix is heading.

The takeaway for the supply chain

The commercial picture is straightforward. If Marvell's sizing holds, interconnect becomes one of the larger identified opportunities in AI silicon — and one where a handful of suppliers, Marvell included, already hold design-win positions. If it doesn't, the figure still marks the boundary of what vendors are willing to promise investors in a competitive AI hardware market.

MarketBeat's report, and the rally that followed it, suggest investors are betting on the first reading. Whether the $37 billion materializes on schedule will depend on AI capital spending holding its pace and on interconnect suppliers converting projections into design wins over the coming quarters.