TSMC is spending like crazy and it still says the AI crowd will be left queuing.
The foundry giant said on its earnings call that capital expenditure is expected to hit $56 billion in 2026, with the cash going toward new fabs and upgrades to existing lines. Even with that lot on the table, TSMC says it will not keep up with demand from the AI binge tearing through the industry.
TSMC is warning that everything from GPUs, CPUs and memory to voltage regulators, ICs, cabling and basic materials is tight.
As Nvidia, AMD, and the Fruity Cargo Cult Apple renew wafer orders, TSMC expects constraints to persist through 2027. It is pushing a global buildout across Japan, Taiwan and the US, with 3nm lines planned in multiple regions.
TSMC president and chief executive C.C. Wei said: “In addition to all the new fabs, we continue to convert 5-nanometer tools to support 3-nanometer capacity in Taiwan. We are also leveraging our manufacturing excellence to drive greater productivity across our fab locations and increase wafer output.”
He said the company is focusing on capacity optimisation across nodes, which includes flexible capacity support among the N7, N5 and N3 nodes.
“We are using multiple levers to do everything we can, wherever we can, however we can to maximise the support to all our customers across all platforms. Also, let me emphasise that while the capacity is tight, we do not pick-and-choose or play favourites among our customers,” Wei said
TSMC says it has not bolted on extra capacity at existing plants yet, but once a node reaches peak output, the company expects upgrades to kick in. The message is that it will squeeze every tool it can before customers start screaming louder.
Wei laid out the 3nm expansion plan in more detail, pointing to Tainan, Arizona and Japan as the next big stepping stones.
Wei said: “Based on our assessment, to meet the strong demand in AI applications, we are stepping up our CapEx investment to increase our N3 capacity. Thus, we are now executing a global capacity plan to support the robust multiyear pipeline of demand for 3-nanometer technologies, which are used by smartphone, HPC/AI (including HBM base dies), automotive, and IoT customers.”
In Taiwan, the company is adding a new 3-nanometer fab to its GIGAFAB cluster at the Tainan Science Park. Volume production is scheduled for the first half of 2027.
“In Arizona, our second fab will use 3-nanometer technologies. Construction is already complete, and volume production will begin in the second half of 2027. In Japan, we now plan to use 3-nanometer technology in our second fab and volume production is scheduled in 2028,” he said.
With TSMC getting hammered by supply limits, chip buyers are spreading risk across rival foundries. Tesla is reportedly working with TSMC and Samsung on next-gen AI silicon while lining up a Terafab partnership with Intel.
Chipzilla is tipped to land major customers by the end of 2026 if its 14A process convinces anyone it can deliver at scale. Samsung is seeing more foundry interest, yet its attention is locked on DRAM like HBM and LPDDR where the AI crowd is hoovering supply.
