TSMC is entering mass production of its 2nm process this year, and analysts reckon high-performance computing rather than smartphones will fuel the growth. The 16 process is a next-generation 1.6nm-class fabrication node, marking the company's entry into the "Angstrom era."

If the rumours are true, it would mark only the second time Nvidia has gone straight to TSMC’s most advanced node, the last being the jump to 0.11 micron. Industry observers say the move reflects the insatiable demand for AI chips, with capital expenditure set to smash records as customers pile in.

While phones once drove leading-edge demand, the focus has shifted to AI and high-performance kit. AMD is expected to roll out 2nm HPC chips next year, and Nvidia is tipped to adopt A16’s backside power delivery technology for its future Feynman GPUs.

The economics are brutal. Reports claim Apple’s 2nm silicon already costs around $27,000 (€25,200), while Nvidia’s A16-based designs with backside power delivery could come in at more than $30,000 (€28,000). Still, the gains in density and performance are tempting enough that customers are willing to “pay for value.”

As the process shrinks, so do the tolerances. TSMC’s suppliers say verification requirements are tightening, with more reliance on system-level testing, automated optical inspection and burn-in. Inspection tools will need to detect flaws invisible to the naked eye, creating new opportunities for equipment makers in the semiconductor supply chain.

TSMC’s annual Games event is scheduled for 8 November, where the company is expected to boast about record revenues and profits as it rides the AI wave.