TSMC Sees 'Endless' AI Demand as Micron Warns DRAM Supply Falls Short

Two earnings calls, two very different positions in the same supply chain. TSMC sells the logic chips that AI accelerators are built on. Micron sells the memory that feeds them. Both reported record numbers. Only one said it can keep up.

The numbers TSMC put on the table

According to Ars Technica, TSMC reported fourth-quarter net income of NT$505.7 billion, roughly $16 billion, up 35 percent year over year and ahead of analyst expectations. Revenue reached $33.7 billion, a 25.5 percent increase from the same period a year earlier. The company guided for nearly 30 percent revenue growth in 2026 and said it plans capital expenditure of between $52 billion and $56 billion this year, up from $40.9 billion in 2025. That spending number is the real signal.

TSMC makes chips for Apple, Nvidia, AMD and Qualcomm, and turns out the vast majority of the world's most advanced semiconductors. When it raises capex, it acts on what customers have told it about future orders, not on a hunch.

CEO C.C. Wei said he did the checking himself. "I want to make sure that my customers' demand are real. So I talked to those cloud service providers, all of them," he told analysts on the call. "The answer is that I'm quite satisfied with the answer. Actually, they show me the evidence that the AI really helps their business." Wei hedged on the long view. Asked whether the semiconductor industry can stay strong for three, four or five years in a row, he said plainly: "I'll tell you the truth, I don't know." He then added that AI looks like it will be "endless" for many years to come, a phrase that travelled quickly.

Memory is the bottleneck nobody can fix quickly

Micron's call, reported by Tom's Hardware, lands on the same problem from the other end of the board. In its first quarter of fiscal 2026, the company posted record revenue of $13.64 billion, up nearly 57 percent year over year, with what it described as significant margin expansion. It credited higher pricing and increased demand from AI data centers. Then came the caveat.

CEO Sanjay Mehrotra said Micron expects supply constraints to persist beyond calendar 2026 and that the company is working on multi-year supply commitments. Even with manufacturing expansion under way, Micron said it expects to meet only half to two-thirds of demand from its key customers. Some of that gap is structural. High-bandwidth memory, the stacked DRAM that sits next to AI accelerators, needs three times as much wafer space as DDR5, according to Tom's Hardware. Micron projects the HBM total addressable market will hit $100 billion by 2028, which would outpace the entire DRAM market as it stood in calendar 2024. Every wafer shifted to HBM is a wafer not making the memory that goes into a PC or a phone.

The expansion timeline does not help much either. Micron is building two fabs in Idaho, with the first expected to start producing chips in mid-2027. A New York fab is expected to break ground in early 2026, with production starting somewhere around 2030. Neither arrives in time for a buyer trying to lock in supply this year.

Mehrotra said customers are "concerned about long-term access to memory" and are lining up for multi-year contracts. That is how buyers behave when they expect scarcity to last, not a glut.

The DRAM shortage is continuing to cause a surge in DDR5 prices. 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.

What the two calls say together

Put side by side, the reports describe an AI buildout that is constrained on the memory side and unconstrained, for now, on the logic side. TSMC can raise capex and commit tens of billions because it trusts the demand it sees. Micron can post record revenue and still tell customers it cannot serve them fully. That asymmetry matters for anyone reading chipmaker earnings as a single AI sentiment indicator. Record revenue at a memory maker is not purely a demand signal. It is also a price signal, produced by shortage. Micron's own framing supports that reading: the company said it is "disappointed" it cannot meet demand across all market segments.

The bubble question hangs over both calls, and neither company settled it. Ars Technica noted 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. In August, OpenAI's Sam Altman acknowledged investors are "overexcited" and that someone will lose a phenomenal amount of money. Wei's answer was to say he verified demand with cloud providers before signing off on the spending increase. Mehrotra's is to sign multi-year contracts.

There is a political layer too. The TSMC earnings landed the same day the US and Taiwan finalized a trade agreement cutting tariffs on Taiwanese goods to 15 percent from 20 percent, according to Ars Technica. The deal commits Taiwanese companies to $250 billion in direct US investment, and TSMC is accelerating expansion of its Arizona fabrication facilities to match. That does not resolve the geographic concentration of leading-edge production, but it changes the arithmetic on where the next round of capacity gets built.

One quarter of records does not prove a multi-year trend. But the two calls do show something narrower and more useful: the AI trade is no longer a single story about demand. It is a story about which parts of the supply chain can physically respond, and how long buyers are willing to wait.