Marvell has put a $37 billion figure on the business of connecting AI chips to each other, according to a report carried by Yahoo Finance. The forecast, issued by the company itself, frames high-speed interconnect silicon as one of the largest addressable markets inside the AI infrastructure buildout — and the number moved markets the same day.

Two other stocks rallied alongside the news, Yahoo Finance reported. The report did not name the two companies in its headline, but the market reaction signals that investors treat the interconnect layer of AI clusters as a monetizable segment in its own right, not merely an accessory to GPU compute.

What is actually being claimed?

The $37 billion figure is a market-size projection from Marvell, not a confirmed revenue figure. It belongs in the same category as other vendor-published total addressable market (TAM) estimates: useful for sizing a segment, but produced by a company with a direct commercial interest in that segment's growth. No independent analyst estimate is cited in the report to corroborate the number.

What is confirmed is the direction of demand. AI accelerators only deliver useful performance when thousands of them communicate at high bandwidth, and the silicon, optics and switching fabric that make that possible scale roughly in proportion to accelerator shipments. Every incremental GPU or custom accelerator deployed in a hyperscaler data center pulls interconnect content with it.

Marvell competes in this space through its data infrastructure portfolio, which includes interconnect and connectivity silicon sold to cloud operators building custom AI platforms. That positions the company to benefit directly if its own forecast proves accurate.

Why interconnect is the choke point

The economics of AI clusters make the connection layer strategically critical. Training and inference workloads split across thousands of accelerators stall on communication bottlenecks, so operators buy high-bandwidth, low-latency links aggressively even at premium prices. That demand pattern has turned interconnect silicon into a scarcer, higher-margin corner of the supply chain than many commodity logic segments.

A vendor forecasting a $37 billion opportunity in this layer is effectively saying that the connective tissue of AI data centers will, within the forecast horizon, approach the scale of businesses that once defined the semiconductor industry. Whether the horizon is five years or ten matters enormously for discounting that figure, and the report does not specify the timeframe Marvell used.

What should readers watch?

Three filters apply when evaluating this figure:

• Attribution: the $37 billion is Marvell's own projection, not third-party analysis.
• Timeframe: no explicit period is given in the report, so the number cannot be annualized or compared to current revenue bases without assumptions.
• Market reaction: the rally in two related stocks is real and immediate, but short-term price moves are not evidence the forecast is right.

The bullish case rests on a structural fact: interconnect demand is derivative of accelerator deployment, and accelerator deployment continues to rise across hyperscalers. If Marvell's forecast direction holds even at half the stated magnitude, connectivity silicon remains one of the fastest-growing segments vendors can sell into — and competitive intensity from other interconnect suppliers will be the variable to watch next.