The global AI boom is pushing the semiconductor industry into a supply crunch that even record-breaking investment can’t immediately fix. TSMC, the world’s most important contract chipmaker, says that despite spending tens of billions of dollars to expand and upgrade its manufacturing footprint, it still expects to fall short of demand as the AI super cycle accelerates.
During its latest earnings call, TSMC revealed that capital expenditures are projected to reach $56 billion in 2026. That massive budget is being funneled into new chipmaking facilities and upgrades to existing production lines, all aimed at increasing wafer output and preparing for the next wave of advanced-node chips.
Even with that scale of spending, TSMC says the industry-wide AI demand surge is simply too large to satisfy quickly. The shortages aren’t limited to just one category. The squeeze touches core AI building blocks like GPUs, CPUs, and memory, while also impacting critical supporting components and materials such as voltage regulators, integrated circuits, cabling, and other supply chain essentials needed to assemble high-performance systems.
TSMC expects the tightness to persist well beyond next year. With major customers like NVIDIA, AMD, Apple, and others continuing to renew and expand wafer orders, the company is anticipating continued constraints into 2027, not just 2026. That outlook matters because advanced manufacturing slots at leading-edge nodes are foundational for everything from AI servers and accelerators to flagship smartphones and next-generation automotive computing.
To address the gap, TSMC is expanding 3nm manufacturing across multiple regions. The company is planning new capacity in Taiwan, the United States, and Japan, with corresponding timelines that show just how long it takes to bring new leading-edge production online. In Taiwan, a new 3nm fab is being added to the company’s GIGAFAB cluster in Tainan Science Park, with volume production targeted for the first half of 2027. In Arizona, TSMC’s second fab will also use 3nm technology; construction is complete and volume production is expected in the second half of 2027. In Japan, the company now plans to use 3nm in its second fab as well, with volume production slated for 2028.
Alongside new facilities, TSMC is also squeezing more output from what it already has. The company says it is converting some 5nm tools to support 3nm capacity in Taiwan, improving productivity across fabs in all locations, and optimizing capacity across multiple nodes with flexible support among N7, N5, and N3. TSMC CEO C.C. Wei also emphasized that while capacity remains tight, the company is not favoring certain customers over others.
The pressure on TSMC is also reshaping customer strategies. With supply constraints limiting how quickly demand can be satisfied, chip designers and system builders are increasingly looking to diversify manufacturing plans. Tesla, for example, is working with both TSMC and Samsung for next-generation AI chips while also preparing its own Terafab approach in partnership with Intel. At the same time, Intel is rumored to be aiming for major customer wins with its upcoming 14A process technology, and Samsung is reportedly seeing increased interest in its foundry services—though it remains heavily focused on DRAM production such as HBM and LPDDR, which are in especially high demand due to growing interest in more capable AI systems.
For the broader tech world, the key takeaway is clear: pouring money into new fabs helps, but advanced semiconductor manufacturing can’t expand overnight. With AI demand hitting nearly every corner of the industry, from data centers to devices and automotive platforms, the next few years are likely to remain defined by tight capacity, long lead times, and fierce competition for cutting-edge production. TSMC’s global expansion should bring meaningful relief, but the company is signaling that the wait will extend through 2027 and, for some new capacity, into 2028.
