Japan’s Rapidus is accelerating the global race toward next-generation chipmaking, and it has now secured significant interest from major American tech players for its ambitious 2nm process. The Tokyo-based firm, which only announced its ‘2HP’ 2nm node last year, is moving beyond theory and into real-world partnerships. According to Rapidus president Atsuyoshi Koike, American companies are already preparing prototypes using the technology, with production trials expected to start as early as next year. 
That’s not a small step – it signals a turning point for Japan’s semiconductor aspirations, which have long been overshadowed by Taiwan’s TSMC, Korea’s Samsung, and America’s Intel.
The two standout partners so far are IBM and Tenstorrent. IBM’s collaboration is unsurprising: the company has been embedded in Rapidus’ 2nm journey since day one, offering advanced packaging expertise, patents, and joint R&D muscle. But Tenstorrent’s inclusion is particularly eye-catching. Led by chip industry legend Jim Keller – who previously reshaped Intel and AMD’s CPU architectures – Tenstorrent is pushing aggressively into AI with its RISC-V based processors. Aligning with Rapidus suggests both a bet on Japanese engineering quality and a willingness to diversify beyond the usual foundry giants.
It’s worth noting that rumors swirl about other interested giants, including NVIDIA, though nothing official has been confirmed. Even so, the fact that Rapidus has secured names like IBM and Tenstorrent at such an early stage underscores both credibility and demand. For perspective, Intel has been loudly advertising its comeback in advanced process technology, yet even long-standing customers have been hesitant to commit. Samsung, meanwhile, continues to battle yield problems and skepticism over its timelines. By contrast, Rapidus appears to be winning mindshare – and potentially market share – before mass production has even begun.
Still, challenges remain. Critics argue that calling Rapidus a ‘leading chip firm’ is premature. Unlike TSMC or Intel, Rapidus doesn’t yet operate high-volume fabs, and much of its momentum relies on government backing, strategic alliances, and promises of future execution. Skeptics point out that Japan’s semiconductor industry has stumbled in the past, with decades of underperformance and reliance on foreign IT companies eroding its global standing. To them, the current optimism feels like déjà vu – bold announcements without proven silicon.
Supporters counter that this time really is different. Japan is uniting multiple heavyweights: not only IBM and Tenstorrent, but also specialized material suppliers like Namics and Nagase, both of which already provide advanced packaging materials for TSMC’s Apple supply chain. With ASML still the sole provider of EUV lithography machines globally, no player can act alone; success comes from alliances. Rapidus, backed by the Japanese government and powered by international partnerships, might be better positioned than past initiatives to close the gap.
Looking ahead, Rapidus expects to deliver process design kits (PDKs) to customers by the first quarter of 2026. If the timeline holds, mass production could begin in late 2026 or early 2027. That would place the company nearly neck-and-neck with TSMC and Intel on 2nm – an extraordinary leap for a newcomer. The central priority, as Koike emphasizes, is not simply speed but producing a reliable, high-quality product. And in a market where trust, yields, and consistency matter as much as raw nanometers, execution will determine whether Rapidus emerges as a serious challenger or another ambitious footnote in chip history.
One thing is clear: competition in the semiconductor world is heating up, and Japan is back in the conversation. Whether Rapidus can sustain this momentum into tangible 2nm wafers remains to be seen, but for now, it has given Intel and Samsung something to worry about – and customers something new to consider.
2 comments
You don’t just borrow IBM patents and print 2nm. This stuff is atomic scale engineering, takes years of iteration
Tbh I bet Rapidus gets it together faster than Intel, real question is who ends up dead last, Samsung or Intel?