OpenAI has reserved TSMC’s 1.6nm capacity to produce advanced AI chips

Many sources have repeatedly mentioned that the startup OpenAI has ambitions to organize, if not the production of chips for computing accelerators, then at least their development. According to Taiwanese media, among TSMC’s clients on the advanced 1.6-nm A16 process technology may be OpenAI, which hopes to have a ready-made design of its own chip by the time it is mastered.

Image Source: Intel

In any case, this is reported by the Economic Daily News. The main customer of chips produced by TSMC using A16 technology should still be Apple, due to already established practice and due to the latter’s need to improve its own mobile processors. TSMC expects to master the A16 technical process in 2026, but it traditionally refuses to discuss its relationships with customers publicly.

According to Taiwanese sources, OpenAI was actively discussing with TSMC the possibility of creating a specialized factory exclusively for its needs. However, after assessing the potential benefits, the plan to create a specialized enterprise was shelved. The development of chips for OpenAI is carried out by such American companies as Broadcom and Marvell Technology, in the case of the first of them, OpenAI may even be among its four largest clients. By the way, the first OpenAI chips can be produced by TSMC using a more mature 3nm process technology, and the choice of A16 technology has already been made for subsequent generations.

As TSMC noted earlier, the 1.6-nm A16 process, compared to its closest predecessor 2-nm N2P, provides an 8-10% increase in transistor switching speed at the same voltage, or reduces power consumption by 15-20% at the same performance level. In this case, the density of transistors can be increased by 10%.