Skipping the queue
Every other foundry in this cohort’s Track A built its advanced-node capability by climbing a ladder — mature nodes first, leading edge later, over decades. Rapidus is trying something close to unprecedented: a startup with no prior advanced-node manufacturing history, attempting to reach 2-nanometer gate-all-around production by 2027, essentially skipping the intermediate rungs entirely [1]. Whether that is ambitious or reckless depends entirely on whether the specific shortcuts Rapidus is taking — IBM’s technology transfer chief among them — can substitute for the decades of accumulated yield learning every other leading-edge fab operator has behind it.
What has actually been demonstrated
The concrete, verifiable technical milestone as of this briefing: Rapidus demonstrated working 2-nanometer gate-all-around transistors in mid-2025, and entered the prototyping phase for the full 2nm process that year [3]. That is a real result — gate-all-around transistor structures at 2nm are genuinely difficult to fabricate correctly at all, let alone reproducibly. It is also, importantly, a prototyping milestone, not a yield or volume-production milestone. The gap between “we made one that works” and “we can make enough of these, reliably and profitably, to run a foundry business” is exactly the gap every other company in this cohort’s foundry track has spent years or decades closing.
The IBM partnership is the whole strategy, not a detail
Rapidus’s technical approach depends directly on IBM’s 2-nanometer-class process expertise, transferred through a formal partnership; IBM is reportedly opening a new office in Chitose, Hokkaido — where Rapidus’s fab is located — specifically to support the collaboration [3]. This is not incidental technical assistance. It is the mechanism by which Rapidus is attempting to compress a multi-decade learning curve into a few years, and it means Rapidus’s success is partially a bet on how transferable IBM’s process knowledge actually is outside IBM’s own research environment — a question with no clean historical precedent to check it against.
The money and the customers
Funding has arrived in stages consistent with the ambition. Rapidus secured ¥267.6 billion (reported as roughly $1.7 billion) from the Japanese government and 32 private-sector companies including Canon, Fujitsu, NTT, SoftBank, and Sony Group [2]. Separately, the Japanese government approved 631.5 billion yen (roughly $3.97 billion) in additional subsidies [1]. On the customer side, Rapidus states it is in active discussions with more than 60 potential customers, with IBM, Tenstorrent, and Fujitsu named as lead or early confirmed clients [4].
What would actually validate the bet
Sixty potential customers in discussion and a handful of named early partners is a promising pipeline, not a proof of business viability. The number that will actually settle whether Japan’s skip-the-queue strategy worked is a 2027 yield figure on production-volume 2nm wafers, compared honestly against what TSMC, Samsung, and Intel were achieving on their own 2nm-class processes at an equivalent point in their own ramps. Rapidus is not a company to dismiss — the technical milestone and the customer interest are both real — but it is also not yet a company that has answered the one question that actually determines whether “Japan’s 2nm foundry” becomes a durable fixture of the industry or a well-funded, well-intentioned experiment that never scales past its first customers.
Why Japan is trying this at all
Japan’s semiconductor industry once led the world in DRAM manufacturing before losing that position to Korea and Taiwan over the course of the 1990s and 2000s — a history this cohort’s companion history briefing on the 1986 US-Japan Semiconductor Agreement covers directly. Rapidus is best understood as Japan’s explicit attempt to re-enter the leading edge rather than remain permanently confined to the materials and equipment layers where Japanese companies such as Shin-Etsu, SUMCO, and Tokyo Electron still hold genuine global leadership. That context matters for judging the scale of government commitment involved: a combined package north of $5.5 billion in public and private funding is a large sum for a single startup, but it is a modest one set against the scale of what Japan lost when its DRAM industry collapsed, and against what TSMC and Samsung have each spent reaching the position Rapidus is trying to reach from a standing start.
The realistic range of outcomes
Given the technical demonstration achieved so far and the scale of external support committed, the plausible outcomes by 2028 span a wide range: a genuinely competitive, yield-mature third leading-edge foundry option outside Taiwan and Korea; a smaller, specialty-focused fab serving a handful of loyal early customers like Tenstorrent without ever reaching TSMC-competitive scale; or a costly, technically partial result that nonetheless advances Japan’s broader semiconductor policy goals even if Rapidus itself never becomes commercially self-sufficient. All three are consistent with what is publicly known today, which is the honest reason this briefing does not pick one as more likely than the others.
