A chip company's chief executive has forecast that demand for the chip his company sells will grow roughly fivefold. On the call about Intel's third-quarter results, Lip-Bu Tan said the global server CPU market could expand from about $40 billion today to more than $200 billion within five years. His engine for all this is the AI agent, meaning software that carries out tasks with minimal human involvement.
The forecast is easy to wave off as earnings-call optimism. The reasoning behind it is more interesting than the number.
A market estimate, not a revenue promise
Tan was describing the size of the whole market, not what Intel itself expects to book. The report is explicit on this point. The $200 billion figure is a potential market, and nothing in it guarantees Intel revenue of that size, or of any particular size.
It is a useful distinction, since a rising tide has a way of being described in the first person by whoever owns a boat.
Why CPUs, in Tan's telling
The argument rests on energy. Tan said GPUs are well suited to the computationally heavy part of the job, the reasoning done by the models themselves. CPUs, he argued, handle some of the work of executing commands more efficiently.
He put numbers on it. By his account, a single model query run on a GPU can consume several watt-hours of energy, while a query on a CPU takes only several joules.
| Processor | Where Tan says it fits | His energy claim per query |
|---|---|---|
| GPU | Compute-heavy model reasoning | Several watt-hours |
| CPU | Some command-execution operations | Several joules |
Two caveats. These are the CEO's claims on an earnings call, and the report does not say which workloads he was comparing. The units also differ, watt-hours against joules, which makes the gap sound dramatic without telling you what was actually measured.
Two multipliers
Tan tied the market's growth to two things. One is the growing number of digital agents. The other is the number of agents that a single processor serves. In his forecast, both push up demand for server CPUs and the dollar size of the market.
The report does not explain how the two interact. More agents per processor would, taken alone, suggest fewer processors per agent, so the forecast presumably depends on the first number growing faster than the second. That is a reading, not something Tan was quoted as saying.
The Classiq deal
The other topic on the call was Intel's purchase of Classiq, a company that makes software for quantum computing. Its platform lets developers write quantum algorithms without tying them to a specific hardware architecture. Tan described it as a complement to Intel's own quantum hardware and software work.
The larger pitch is a product lineup. With Classiq, Intel says it can offer a computing platform that combines CPUs, GPUs, application-specific integrated circuits (ASICs) and quantum processors (QPUs). Tan said the deal lets Intel become a full participant in the quantum market, earning revenue from both hardware and software.
What the report does not say
- The price. The purchase cost of Classiq is not given.
- The timeline. There is no date by which the quantum business is expected to start producing revenue.
- The basis for the forecast. Beyond Tan's reasoning about agents and energy use, no modelling behind the $200 billion figure is described.
So for now the $200 billion is a number said aloud on a call, and the quantum revenue is a sentence in the future tense. A forecast for a whole market is the easiest kind to make and the hardest to invoice.