Marvell Technology has put a $37 billion figure on the business of connecting AI chips to each other, a market forecast that moved not only its own shares but rallied at least two other stocks in the semiconductor supply chain.

The number matters because it comes from a company whose revenue depends directly on it. Marvell sells the silicon — custom interconnect and networking components — that links accelerators into the large clusters AI systems now require. When a supplier of that scale publishes a market sizing, investors treat it as a demand signal rather than a pure analyst exercise.

Why is connectivity suddenly a $37 billion question?

AI compute has shifted the bottleneck. Training and inference clusters now run tens of thousands of accelerators in parallel, and the chips are only as fast as the fabric between them. High-bandwidth links, switching silicon and custom PHYs have moved from a supporting role in the bill of materials to one of its largest line items.

Marvell's $37 billion claim, reported by The Globe and Mail, positions that fabric-building layer as a distinct and fast-growing business rather than a feature of the systems business. The forecast implies that as accelerator shipments scale, spending on the interconnect between them scales with them — and in some scenarios faster, because cluster efficiency depends on link bandwidth and latency.

The market reaction was immediate. Beyond Marvell itself, two other stocks rallied on the news, according to the report — a sign that investors read the forecast as a read-through for the wider connectivity supply chain rather than a single-company story.

What does the forecast tell buyers and rivals?

A supplier-published market number serves two audiences.

• For customers: it signals that Marvell intends to keep investing in the product families that serve this segment, and that pricing conversations will happen in a market the vendor believes is large and durable.
• For competitors: it draws a line in the sand. Broadcom, whose custom accelerator and networking franchises touch the same sockets, and the merchant networking incumbents now have a public figure to benchmark their own attach-rate assumptions against.

The distinction between a forecast and a booked figure is worth keeping sharp. The $37 billion is a market-size projection — how much will be spent on AI chip interconnect across the industry — not Marvell revenue guidance. Investors rallying connectivity names on the headline are effectively underwriting that projection before any vendor has reported order volumes against it.

Is the interconnect layer the next battleground?

The historical pattern in semiconductors argues for caution on any single number. Memory, packaging and advanced nodes have each produced large TAM claims over the past two years, and realized capacity and revenue have followed unevenly behind them. Interconnect demand is more tightly coupled to AI cluster buildouts than most segments, though, because every accelerator shipped needs a fabric around it to be useful.

That coupling is what makes Marvell's figure commercially significant. If hyperscaler capital expenditure on AI infrastructure continues at current scale, the interconnect layer grows as a forced consequence of accelerator shipments — not as a discretionary upgrade cycle. If cluster spending slows, the segment contracts just as mechanically.

The two unnamed stocks that rallied alongside Marvell suggest the market has already started pricing that linkage. Whether the $37 billion proves conservative or inflated will show up first in networking and custom-silicon revenue lines over the coming quarters, where Marvell and its rivals will have to convert the forecast into reported silicon shipments.