Applied Materials Wants to Double the Machines by 2028

Applied Materials reported fiscal third-quarter revenue of $9.12 billion on August 13, 2026 — up 25% year over year and 15% sequentially — with earnings of $3.50 a share, ahead of Wall Street. The semiconductor systems segment alone contributed $7.04 billion, against $5.56 billion in the same quarter last year.

Then CFO Brice Hill told analysts something more consequential than the beat: the company is targeting the capacity to build twice its current quarterly semiconductor-system volume by 2028, and has already started groundwork on a further manufacturing buildout meant to carry demand into 2030.

Fourth-quarter guidance is $10.25 billion, plus or minus $500 million.

The stock fell anyway.

What a doubling actually commits to

Semiconductor capital equipment is not a business you scale on a quarter's notice. A deposition or etch system is a multi-ton instrument with a supply chain of specialized optics, vacuum components, precision motion stages, and control electronics that themselves have year-plus lead times. Doubling output means signing for that supply chain now, hiring and training the technicians who assemble and install it, and building the floor space it sits on.

Companies do not make that commitment on a forecast they consider soft. They make it when the order book has already outrun the factory, and when the customers placing those orders are large enough that a missed delivery slot costs more than the overbuild risk.

Hill's language — groundwork for an additional buildout beyond the doubling, aimed at 2030 — is the part worth reading twice. That is not a company hedging a cycle. That is a company treating the current level of demand as a new floor.

The demand is not general. It is specific.

Applied attributed the strength to leading-edge logic, DRAM, and advanced packaging. Those three categories are the physical substrate of AI compute, and each is constrained in a different way.

Leading-edge logic is the GPU and accelerator die itself. DRAM — particularly the high-bandwidth stacks that sit beside those dies — has been the binding constraint on accelerator shipments for most of two years; memory, not compute, is what determines how large a model you can serve per package. Advanced packaging is the assembly step that bonds logic and memory into a single unit, and it is the newest bottleneck: the industry spent decades optimizing the transistor and comparatively little time optimizing the interposer.

Applied sells tools into all three. That is why its revenue mix has been rotating toward the AI buildout faster than a diversified equipment vendor's normally would.

It is also why the doubling target is a statement about packaging as much as about wafers. The number of process steps per finished AI package has gone up, and each new step is a tool sale.

Why the stock fell on a record

Shares slipped despite the beat and the raise, which is now a familiar pattern across the AI supply chain. It has been the reaction to hyperscaler capex prints, to memory guidance, and now to equipment.

The market is no longer grading whether the buildout is happening. It is grading who pays for it, and on what terms. A record quarter accompanied by a commitment to double capacity reads two ways at once: as confidence in durable demand, or as a supplier absorbing cycle risk on behalf of customers who can cancel. Equipment vendors have been on the wrong side of that trade before — every prior semiconductor upcycle ended with tool makers holding capacity built for orders that stopped arriving.

The bull case is that this cycle is structurally different because the demand is coming from a small number of extremely well-capitalized buyers with multi-year contractual commitments, not from a fragmented consumer-electronics market. The bear case is that a small number of buyers is exactly what concentration risk looks like.

Both readings are defensible from the same print. The stock reaction is the market declining to resolve them.

The geopolitical line item nobody guides on

Applied's growth is happening while its access to one of the largest equipment markets on earth remains politically constrained. Export controls have reshaped what it can sell into China, and the mature-node capacity Chinese foundries are adding — SMIC crossed $3 billion in quarterly revenue for the first time this quarter at 93.7% utilization, without a leading-edge node — is capacity that will need tools from someone.

The company does not guide on policy, and it should not. But a doubling of Western equipment output landing at the same time Chinese foundries expand mature-node capacity produces a specific outcome: two parallel equipment markets, with different customers, different tool generations, and different suppliers. That is a structurally less efficient industry than the one that existed in 2018, and the inefficiency is priced into everyone's capex.

The read

Three things are true of this print.

The revenue is real and the mix is the right mix — leading-edge logic, DRAM, and packaging are the three places the AI buildout physically bottlenecks, and Applied sells into all of them.

The capacity commitment is the actual news. A supplier does not double manufacturing output on a two-year horizon, and pre-fund a further expansion for 2030, unless its customers have told it something more specific than "demand looks good." Equipment order books are the earliest honest signal in the chain — earlier than hyperscaler capex, which is announced, and earlier than chip revenue, which is recognized.

And the stock's decline on a record quarter is the market's consistent message this year: it will pay for the buildout, but it has stopped applauding it. Every participant now gets asked the same question — not whether the demand exists, but who is left holding the assets when the ordering slows.

Applied Materials just answered that it intends to be holding twice as many.