TSMC Calls AI Demand Endless, Micron Warns Memory Shortage Runs Past 2026

Two chipmakers, two earnings calls, one supply chain. TSMC posted record fourth-quarter net income of NT$505.7 billion, about $16 billion, up 35 percent year over year, according to Ars Technica's report on the 16 January call. Revenue reached $33.7 billion, a 25.5 percent increase from the same period a year earlier. The company told investors to expect nearly 30 percent revenue growth in 2026, and said it will spend between $52 billion and $56 billion on capital expenditures this year, up from $40.9 billion in 2025.

That is a lot of money for a company that says it cannot see the end of the cycle.

"All in all, I believe in my point of view, the AI is real, not only real, it's starting to grow into our daily life," CEO C.C. Wei said on the call, according to Ars Technica. "So another question is 'can the semiconductor industry be good for three, four, five years in a row?' 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."

What TSMC is actually selling

TSMC manufactures chips for Apple, Nvidia, AMD and Qualcomm. Ars Technica notes the company produces the vast majority of the world's most advanced semiconductors, and describes it as a linchpin of the global electronics supply chain. When TSMC raises spending, the companies designing AI accelerators are effectively voting with their order books. TSMC does not build chips on spec. It builds what customers book.

Wei said he verified that demand before signing off on the capex increase. "I want to make sure that my customers' demand are real. So I talked to those cloud service providers, all of them," he said on the call, per Ars Technica. "The answer is that I'm quite satisfied with the answer. Actually, they show me the evidence that the AI really helps their business."

That claim matters because the bubble question has been loud. In November, Google CEO Sundar Pichai warned of "irrationality" in the AI market and said no company would be immune if a potential bubble bursts, Ars Technica reported. OpenAI's Sam Altman acknowledged in August that investors are "overexcited" and that "someone" will lose a "phenomenal amount of money."

Wei's answer is not a forecast. It is a procurement decision. TSMC is committing tens of billions of dollars to capacity on the basis of conversations with the buyers, not on the basis of a market model.

The timing of the earnings report was not accidental in political terms either. Ars Technica reported the results landed the same day the US and Taiwan finalized a trade agreement that cuts tariffs on Taiwanese goods to 15 percent, down from 20 percent. The deal commits Taiwanese companies to $250 billion in direct US investment, and TSMC is accelerating the expansion of its Arizona fabrication facilities to match.

Micron's version of the same story

If TSMC is the optimistic half of the earnings season, Micron is the supply-side warning. In its first earnings call since killing the Crucial consumer brand, Micron reported record revenue of $13.64 billion for the first quarter of fiscal 2026, up nearly 57 percent year over year, according to Tom's Hardware's write-up of the 20 December call. The company attributed the growth to higher pricing and increased demand in AI data centers, and reported "significant margin expansion."

Then CEO Sanjay Mehrotra said something less cheerful for anyone building a PC. Micron expects supply constraints to "persist beyond calendar 2026," and is working on multi-year supply commitments. Even with capacity expansion under way, Micron anticipates it will only be able to meet "half to two-thirds" of demand from its key customers.

"Concerned about long-term access to memory" is how Tom's Hardware described the customer posture, with buyers lining up for multi-year contracts to secure supply.

The reason is structural. HBM requires three times as much wafer space as DDR5, according to Tom's Hardware, and Micron expects strong growth in HBM revenue. The company projects the HBM total addressable market will hit $100 billion by 2028, which would outpace the entire DRAM market in calendar 2024.

Micron is building. Two fabs in Idaho are in progress, with the first expected to start producing chips in mid-2027. A New York fab is proceeding as well. Micron says it expects to break ground in early 2026, with production starting somewhere around 2030. That timeline is the point. The shortage is a 2026 and 2027 problem. The relief is a 2030 problem.

The price signal is already visible

The consumer end of the market is where this becomes concrete. The DRAM shortage is continuing to cause a surge in DDR5 prices, Tom's Hardware reported. Most suppliers agree the shortage will continue into at least next year and likely beyond, though some suggest prices will level off soon. GPU vendor Sapphire says prices will stabilize in the next six to eight months, while Kingston suggests prices will "continue to go up" in the future.

Two vendors, two directions. That is what an unresolved shortage looks like from the retail side.

Micron still supplies DRAM for PCs and phones despite closing the Crucial brand, and expects PC shipments to continue to grow even with constrained supply. The company said it is "disappointed" it cannot meet demand across all market segments. Disappointment is doing some work in that sentence: unmet demand at record prices is a specific kind of corporate regret.

Put the two calls side by side and the picture is not really contradictory. TSMC sells the logic that goes into AI accelerators. Micron sells the memory that feeds them. Both are running at capacity. Both are spending heavily. One says demand looks endless, the other says it can serve at most two-thirds of its best customers through next year.

The difference is where the constraint bites. TSMC's capex number, $52 billion to $56 billion, is a bet that the demand Wei heard about on those cloud calls is durable enough to justify building for it. Micron's half-to-two-thirds figure is the admission that even with that kind of spending across the industry, memory supply does not turn on quickly.

Ars Technica framed the TSMC result as a signal: when TSMC reports strong demand and ramps up spending, it indicates that the companies designing AI chips expect years of continued growth. Tom's Hardware framed Micron's result as grim, at least for PC enthusiasts. Both framings are accurate, and they describe the same supply chain from opposite ends.

What neither call settles is whether the AI buildout is a durable industrial shift or a cycle that will look obvious in hindsight. Wei said he does not know. Mehrotra said customers are worried about access, not about whether they will need the memory. For now, the order books are doing the talking, and the fabs will not be finished until the end of the decade.