TSMC calls AI demand endless as Micron warns it can meet only half of orders

Two earnings calls, about a month apart, describe the same shortage from opposite ends. TSMC sells the logic chips that go into AI accelerators. Micron sells the memory those accelerators need in growing volumes. Both are sold out. Only one sounds relaxed about it.

According to Ars Technica, TSMC's fourth-quarter revenue was $33.7 billion, up 25.5 percent from the same period a year earlier. Net income rose 35 percent to NT$505.7 billion, roughly $16 billion, beating analyst expectations. The company said it expects revenue growth of nearly 30 percent in 2026. It plans capital expenditure of $52 billion to $56 billion this year, up from $40.9 billion in 2025.

That spending number is the actual news. TSMC produces the vast majority of the world's most advanced semiconductors, and it makes chips for Apple, Nvidia, AMD and Qualcomm. When it lifts capex by more than a quarter, it turns other people's forecasts into concrete orders for equipment and factory space.

Wei: I talked to the cloud providers

CEO C.C. Wei addressed the bubble question directly during the earnings call. He said he had spoken to cloud service providers to check whether their demand was real before committing to the increase.

"I want to make sure that my customers' demand are real. So I talked to those cloud service providers, all of them," Wei said. "The answer is that I'm quite satisfied with the answer. Actually, they show me the evidence that the AI really helps their business."

He was more careful about the long horizon. "Can the semiconductor industry be good for three, four, five years in a row?" he asked. "I'll tell you the truth, I don't know. But I look at the AI, it looks like it's going to be like an endless, I mean, that for many years to come."

That optimism sits against months of public worry. Ars Technica notes that in November, Google CEO Sundar Pichai warned of "irrationality" in the AI market and said no company would be immune if a bubble bursts. OpenAI's Sam Altman acknowledged in August that investors are "overexcited" and that "someone" will lose a "phenomenal amount of money."

Wei's answer to that is not a forecast. It is a purchase order.

Micron: half to two-thirds

The memory side of the supply chain reads very differently. Tom's Hardware reported on 20 December that Micron, in its first earnings call since killing its Crucial consumer brand, posted record revenue of $13.64 billion for the first quarter of fiscal 2026, up nearly 57 percent year over year. The company attributed the growth to higher pricing and increased demand from AI data centers.

CEO Sanjay Mehrotra said supply constraints will "persist beyond calendar 2026" and that Micron is working on multi-year supply commitments. HBM, the high-bandwidth memory used in AI accelerators, takes three times as much wafer space as DDR5. Micron projects the HBM total addressable market will reach $100 billion by 2028, which would outpace the entire DRAM market in calendar 2024.

Even with capacity expansion underway, the company says it expects to meet only "half to two-thirds" of demand from key customers. Mehrotra said those customers are "concerned about long-term access to memory" and are lining up for multi-year contracts. Two fabs in Idaho are planned, with the first due to start producing chips in mid-2027. A New York fab is expected to break ground in early 2026, with production starting around 2030.

For anyone buying a PC, that timeline is the story. DDR5 prices are already surging. Most suppliers expect the shortage to continue into at least next year and likely beyond. GPU vendor Sapphire told Tom's Hardware prices will stabilize in the next six to eight months, while Kingston said prices will "continue to go up."

The two calls agree on the diagnosis and differ only on tone. TSMC is spending against demand it says it verified. Micron is spending too, and still telling customers to expect less than they asked for. Neither company is behaving like a business preparing for a downturn.

The trade backdrop adds a second variable. The TSMC earnings report landed the same day the US and Taiwan finalized an agreement cutting tariffs on Taiwanese goods to 15 percent from 20 percent. The deal commits Taiwanese companies to $250 billion in direct US investment, and TSMC is accelerating expansion of its Arizona fabrication facilities to match. Ars Technica reported those terms on 16 January.

So the constraint is not money, and it is not demand. It is wafers, fabs and time. Micron's first Idaho output is more than a year away and its New York site is roughly four years out. TSMC's capex jump buys equipment that takes years to install. The gap between what AI builders want and what the supply chain can hand them will not close in 2026.