In an interview with Nikkei Asia, Keller said Tenstorrent, his AI startup, would not be reluctant to adopt Intel’s newest processes but added a warning.

“They still have a lot of work to do … to deliver a really solid technology roadmap,” Keller said.

He confirmed that Tenstorrent is talking to several players about 2nm production. “We are talking to TSMC, Rapidus and Samsung for 2-nm technologies,” he said. Intel is on the list, but only if it can show it is ready for outside customers.

Chipzilla has been pitching its 18A process as a breakthrough, saying it offers “25 per cent higher frequency at the same power and 36 per cent lower power at the same frequency versus Intel 3” as well as “more than 30 per cent density” improvements.

The marketing is strong, but Keller’s comments highlight the gap between the roadmap and what potential customers are willing to trust.

Keller is not suggesting Intel should be written off. He told Nikkei Asia that he remains optimistic about Intel Foundry’s future and that Tenstorrent will consider it seriously if the company can deliver. The key issue is whether Chipzilla can shift from building only its own chips to supplying external clients at real volume.

Tenstorrent, which develops RISC-V based AI hardware, is making 2nm a core part of its future plans. Keller said this is why he has approached every major foundry player in the business.

Keller’s CV includes stints at AMD, Job’s Mob, Intel and Tesla. Whenever he comments on process technology, the industry listens. The fact he is saying that Intel may be moving quickly with its nodes, but until it can prove that its foundry is ready for outsiders, it is not yet in the same league as its rivals, is bad news.