AI Chip Demand Outpaces Packaging Capacity and Hiring

Stephen Engle reports “extreme bullishness” among suppliers gathered at SEMICON Taiwan about the continuation of the AI buildout. The optimism is not presented as a forecast from one category of company. It is visible, in Engle’s account, in the expectations of a maker of data-center storage controllers, in demand for the tools used by the world’s largest semiconductor fab, and in the order book of a Taiwanese advanced-packaging company.

That breadth is the basis for the language of a semiconductor “super cycle.” Phison, which makes NAND flash controllers that go into data centers, says the industry is in the first year of a 10-year cycle. Its estimate is based on its projections and its order book: the company is not merely describing current demand, but treating booked business and expected demand as evidence of a durable buildout.

TSMC offers a separate operational signal. Engle says the company’s need for chipmaking tools has doubled since the end of last year. The point is distinct from Phison’s forecast: Phison cites its own projections and orders for a component used in data centers, while TSMC’s increased tool requirement points to more demand for the equipment needed to make chips. Continued chip orders from customers including Nvidia, Engle adds, help sustain that demand.

The claims therefore describe pressure at different points of the supply chain rather than a single surge in demand for headline AI chips. Storage controllers, fabrication tools, and packaging capacity are all being pulled into the same investment cycle, according to the companies Engle spoke with.

The strongest tension in the bullish account is that suppliers expect a long expansion precisely while some cannot supply what their order books require. GPTC, an advanced-packaging company in Taiwan, says the super cycle will last at least until 2030. But its view is paired with a concrete capacity problem: it added a new factory that increased capacity by 50% and still says it cannot meet the needs represented in its orders.

That makes GPTC’s position different from both Phison’s and TSMC’s. Phison’s evidence is a long-range projection supported by orders. TSMC’s is sharply higher need for manufacturing tools. GPTC is reporting a more immediate mismatch between expansion and demand: even a substantial new factory has not brought available capacity into line with the business it has on hand.

Stephen Engle presents advanced packaging as one of the practical points at which a prolonged AI buildout has to be converted into physical output. GPTC’s forecast through 2030 is not simply a statement that demand will remain high. It is also a statement made while the company says its present expansion has not been enough to satisfy its order book.

The source does not establish whether this constraint is representative of the whole industry. It does show why broad confidence in a multi-year cycle does not mean supply can immediately catch up. The same orders that support a long-cycle outlook are, for GPTC, also evidence of a current production shortfall.

GPTC identifies manpower as its biggest problem. It has 200 open jobs, Engle says, and cannot fill positions it “desperately” needs despite the demand in front of it.

The hiring problem matters because GPTC has already taken a visible step to expand physical capacity. Its new factory added 50% capacity, yet the company still cannot meet its order-book needs; at the same time, it says it lacks the people needed for its operations. The source does not specify which roles are vacant or why they remain unfilled. It does make clear that adding a factory is not the only condition for increasing output.

This is the reservation inside an otherwise emphatically bullish picture. Semiconductor suppliers have experienced boom-and-bust cycles before, Engle notes, and that history has made some participants skittish about both the duration and breadth of the current expansion. Phison’s 10-year projection, TSMC’s doubled need for tools, and GPTC’s projection through 2030 all point toward continued demand. GPTC’s unfilled jobs and unmet order book show that confidence in demand can coexist with immediate limits on the ability to serve it.