Good news I guess, now waiting for them to sell their first transistor.

Probably the most meaningful and exciting news in the industry in decades! Not even considering the "shortages", this is something very significant! If you're into the stock market this is something to pay attention to. It's a completely different way of manufacturing logic and memory circuits at 2nm (and soon 1nm) resolution scales. It would only be more interesting if the process included the use of graphene!

This is pretty huge guys!

TesselatorIt's a completely different way of manufacturing logic and memory circuits at 2nm (and soon 1nm) resolution scales.

How so? Rapidus is implementing IBM Research technologies for their main processes. It's nanosheet Gate-All-Around - Intel, Samsung and TSMC do have equivalents.

TesselatorIt would only be more interesting if the process included the use of graphene!

That's unlikely if you're talking about using graphene as a semiconductor. Only in 2024 it was shown that it can actually be done, so commercialization is quite far away.

ncrsHow so? Rapidus is implementing IBM Research technologies for their main processes. It's nanosheet Gate-All-Around - Intel, Samsung and TSMC do have equivalents.

I'm just watching videos such as



There's also some RAPIDUS tech which uses a mold process rather than lithography and they're achieving similar results to 2 and 4nm lithographic processes with that now in the very early stages. Pretty interesting stuff. The new 2nm process from Rapidus, the progress in material research and the geopolitical implications make this sort of news very interesting - to me at least if not to the entire world. I think it's the most exciting and game changing news in decades!

ncrsThat's unlikely if you're talking about using graphene as a semiconductor. Only in 2024 it was shown that it can actually be done, so commercialization is quite far away.

Which is way I said it would be even more interesting if...

I think there is mixup between the Canon “stamp” based process that NIL is using for a non-euv 2nm process that is a fraction of the cost and power (but also size) of EUV from ASML, and the Rapidus 2nm that uses ASML EUV - both happening in Japan.

Rapidus is a bad idea. In terms of its business model, Rapidus is a pure foundry. Among the leading 3 semiconductor manufacturers, tsmc is a pure foundry, while samsung and intel are half-IDM and half-foundry. This shows that if you want to compete with the powerful tsmc, you need to at least have an IDM business model, generating profitable products through IDM, and using your own products to assist with process yield ramp-up, rather than relying on the goodwill of customers. You need to achieve this yourself first to help your potential foundry customers improve yield, which can enhance your own competitiveness. In fact, samsung only brought its gaa process into the foundry after years of IDM.

What makes Rapidus fundamentally different from the 3 leading semiconductor companies is its single-wafer production, which means that from the very beginning, Rapidus cannot match the cost control levels of the other 3, and is far inferior. Its hvm standard, in terms of process maturity, falls behind the risk production of intel and tsmc, even though the two companies define risk production differently.

Now Rapidus has already consumed a large amount of government funds with no sign of profitability. The total capacity of its fab is equivalent to 1/3 or less of a single fab of tsmc or intel. With less than a year left until its hvm target, the pdk status remains unclear, and no operational logic chips have been seen. Is such a lack of transparency typical of japanese industrial policy? If such funds had been invested in RAM production, where Japan was once strong and the difficulty is lower, it would have been profitable long ago.

For advanced semiconductors, not even samsung and intel dare to guarantee their advanced nodes, and tsmc is also extremely cautious, let alone technology that IBM(loser, out of the loop for years) itself was unwilling to invest in for mass production.

tsurugiRapidus is a bad idea. In terms of its business model, Rapidus is a pure foundry. Among the leading 3 semiconductor manufacturers, tsmc is a pure foundry, while samsung and intel are half-IDM and half-foundry. This shows that if you want to compete with the powerful tsmc, you need to at least have an IDM business model, generating profitable products through IDM, and using your own products to assist with process yield ramp-up, rather than relying on the goodwill of customers. You need to achieve this yourself first to help your potential foundry customers improve yield, which can enhance your own competitiveness. In fact, samsung only brought its gaa process into the foundry after years of IDM.

What makes Rapidus fundamentally different from the 3 leading semiconductor companies is its single-wafer production, which means that from the very beginning, Rapidus cannot match the cost control levels of the other 3, and is far inferior. Its hvm standard, in terms of process maturity, falls behind the risk production of intel and tsmc, even though the two companies define risk production differently.

Now Rapidus has already consumed a large amount of government funds with no sign of profitability. The total capacity of its fab is equivalent to 1/3 or less of a single fab of tsmc or intel. With less than a year left until its hvm target, the pdk status remains unclear, and no operational logic chips have been seen. Is such a lack of transparency typical of japanese industrial policy? If such funds had been invested in RAM production, where Japan was once strong and the difficulty is lower, it would have been profitable long ago.

For advanced semiconductors, not even samsung and intel dare to guarantee their advanced nodes, and tsmc is also extremely cautious, let alone technology that IBM(loser, out of the loop for years) itself was unwilling to invest in for mass production.

Your analysis seems to have several fundamental misunderstandings about both Rapidus’ strategies and how the advanced semiconductor industry actually works in 2026.

1. “Pure foundry is a bad model / you need IDM like Samsung/Intel”
This is outdated thinking according to most CS and industry thinkers actually involved. TSMC became the most successful and profitable semiconductor company in history precisely as a pure-play foundry. The rise of the fabless model (NVIDIA, AMD, Apple, Broadcom, Qualcomm, etc.) is the dominant trend. Forcing an IDM model would require Rapidus to also design and sell its own high-volume products - which is something Japan does not currently have competitive strength in at leading-edge logic. Rapidus is correctly focused on being a foundry that serves global (and Japanese) designers.

2. Single-wafer production is a fatal cost disadvantage?
This is the opposite of the truth for what Rapidus is targeting. Single-wafer processing deliberately trades some raw throughput for:

• Dramatically shorter cycle times
• Much richer per-wafer process data (critical for AI-driven yield learning)
• Superior process control at the most advanced nodes
• Better suitability for high-mix, high-value chips (AI accelerators, HPC [High-Performance Computing], custom logic, etc.)

For the highest-end nodes, speed of learning and yield ramp can and often does matter more than absolute lowest cost per wafer in the early years. Rapidus is optimizing for the right variables for such a late entrant.

3. “No PDK, no logic chips, lack of transparency”?
This is factually incorrect as of late 2026:

• Pilot PDK has already been released to select customers.
• Working 2nm GAA transistors were demonstrated in 2025.
• Customer test-chip production is/was scheduled for 2026 and may have already shipped by now.
• Full HVM is targeted for the second half of 2027, which is still the official, publicly stated timeline.
• They have active partnerships with 17 design houses (including Synopsys, Cadence ecosystem players, etc. etc. etc.) - in discussions with 60+ potential customers.

Japanese industrial projects are often less noisy than American or Taiwanese ones, but the milestones are real and publicly reported by METI and the company. BTW, METI is Japan's Ministry of Economy, Trade and Industry. Worth knowing is: it's the successor to MITI, the old Ministry of International Trade and Industry, which was famous in the eighties for steering Japan's industrial rise. Same ministry, renamed.

4. “IBM is a loser who abandoned the technology”? What???
IBM never left advanced research. The 2nm nanosheet GAA technology Rapidus is industrializing originated from IBM Research as just one example. IBM chose not to run its own high-volume fab (a good and rational business oriented decision after exiting pure manufacturing years ago), but it is one of the strongest research organizations in the world on transistor architecture. Technology transfer + joint development is exactly how successful late "entrants" acquire capability and meaningful market share.

5. Government funding with no profitability yet...
This is industrial policy, not a normal startup. The United States (CHIPS Act), Europe, Korea, and China are all pouring massive public money into leading edge capacity for strategic reasons. Japan is doing the same because relying almost entirely on TSMC (geopolitically concentrated in Taiwan) is a national-security risk. Expecting Rapidus to already be profitable before HVM even begins is an absurd standard that no one applies to anyone else - not Intel Foundry, not Samsung’s advanced nodes, nor even TSMC’s early ramps on new nodes.

6. “Should have invested in memory instead...” :P
Japan already has significant strength in memory materials, equipment, and (via Kioxia) NAND. The critical strategic gap is leading-edge logic. That is exactly the gap Rapidus is designed to clobber. Diversifying into the one area where Japan is weakest is the likely the most rational "national strategy".

Rapidus is a high-risk, high-reward national project. It may or may not fully succeed commercially against/with TSMC or others, but dismissing it with 2020-era foundry "ideology" and outdated facts about its current progress is not serious analysis. Two days ago as this article highlights, they launched Rapidus CORE (Collaborative Open Rapid Ecosystem), a global framework to help customers design and mass-produce 2nm chips. The first piece is seventeen Design Solution Associates, including Synopsys, Cadence, Infosys, and Toshiba, who act as a bridge between customers and Rapidus, handling everything from architecture through tapeout and volume production coordination. The timing seems very deliberate: with mass production targeted for late fiscal 2027 and no committed volume customers yet, they're trying to lower the barrier to entry by making design support part of the package rather than just selling wafers. And as all gamers and CS historians should be aware, this has a been a winning strategy across several notable corporations in several of our favorite industries. Anyway, that's my story, and I'm sticking to it! :D If successful this could be a major win for Japan and even the globe - unless you think it's flat that is. :P

You're right about one thing though... Rapidus's biggest obstacle isn't making 2nm chips; it's convincing customers to bet their supply chain on a company that hasn't proven volume production yet. But if I know the Japanese, that won't be an issue for long - we should see that tension mostly evaporate by mid/late 2027 and be completely eliminated by about a year after that - just 16 to 22 months. 🙏

TesselatorYour analysis seems to have several fundamental misunderstandings about both Rapidus’ strategies and how the advanced semiconductor industry actually works in 2026.

1. “Pure foundry is a bad model / you need IDM like Samsung/Intel”
This is outdated thinking according to most CS and industry thinkers actually involved. TSMC became the most successful and profitable semiconductor company in history precisely as a pure-play foundry. The rise of the fabless model (NVIDIA, AMD, Apple, Broadcom, Qualcomm, etc.) is the dominant trend. Forcing an IDM model would require Rapidus to also design and sell its own high-volume products - which is something Japan does not currently have competitive strength in at leading-edge logic. Rapidus is correctly focused on being a foundry that serves global (and Japanese) designers.

You completely biased the meaning of the statement that IDM is obsolete. Most semiconductor design companies turned to fabless, and TSMC foresaw this trend, seizing the historical opportunity to become a top pure foundry company, these two events complemented each other.

For major fabless companies and TSMC, breaking away from the IDM business model was of vital importance. Does this have anything to do with Rapidus, another pure foundry that is completely unrelated to TSMC? All major fabless companies are customers of TSMC, not Rapidus.

Intel and Samsung, as companies that are still IDM, represent 2 out of all 3 advanced semiconductor players, proving the strong strength of the IDM business model in the field of semiconductor. I mean, if you are fabless, you can use TSMC to solve manufacturing issues, but if you want to manufacture advanced semiconductors outside of TSMC, learning from Intel and Samsung IDM is a practically viable solution.

That being said, Japan lacks the companies like Intel and Samsung in the design field, and there is no domestic demand for advanced semiconductors, so if they want to do advanced semiconductor, they can only operate as a pure foundry. Rapidus cannot directly compete with tsmc, Samsung, and Intel, while suffering from terrible yields, so it can only do single wafer production. This choice avoids all lucrative high volume customers and focuses on a small number of startups. It might be a clever choice, because Rapidus itself cannot obtain any single high volume customer.

provided that Japan wants to enter the advanced semiconductor field, limited by domestic conditions, it step by step chose a plan with weaker competitiveness and higher costs. This plan is extremely bold, featuring several main strategies that no semiconductor manufacturing company operating for many years would dare to invest in, such as global foundry, as a pure foundry, giving up on advanced semiconductors. Intel and samsung's manufacturing investments could only proceed under the condition of ensuring that they themselves can act as customers and the fab can generate predictable revenue.

Tesselator2. Single-wafer production is a fatal cost disadvantage?
This is the opposite of the truth for what Rapidus is targeting. Single-wafer processing deliberately trades some raw throughput for:

• Dramatically shorter cycle times
• Much richer per-wafer process data (critical for AI-driven yield learning)
• Superior process control at the most advanced nodes
• Better suitability for high-mix, high-value chips (AI accelerators, HPC [High-Performance Computing], custom logic, etc.)

For the highest-end nodes, speed of learning and yield ramp can and often does matter more than absolute lowest cost per wafer in the early years. Rapidus is optimizing for the right variables for such a late entrant.

And after investing $20 billion, what did Japan get in the end? A fab with single-wafer production and low volume, incapable of mass-producing products. What on earth is the use of this?

Japan has no strong fabless companies to support its industrial policy, is it possible that US fabless companies, having tsmc, samsung, and intel available, would be willing to choose Rapidus?

Tesselator3. “No PDK, no logic chips, lack of transparency”?
This is factually incorrect as of late 2026:

• Pilot PDK has already been released to select customers.
• Working 2nm GAA transistors were demonstrated in 2025.
• Customer test-chip production is/was scheduled for 2026 and may have already shipped by now.
• Full HVM is targeted for the second half of 2027, which is still the official, publicly stated timeline.
• They have active partnerships with 17 design houses (including Synopsys, Cadence ecosystem players, etc. etc. etc.) - in discussions with 60+ potential customers.

On its official website, the pdk was last mentioned 1 year ago, and these latest developments you mentioned were not brought up by them, such as a good pdk, or mentioning node performance and showcasing functional devices logic chips based on its node, which is what I mean by a lack of transparency.

TesselatorJapanese industrial projects are often less noisy than American or Taiwanese ones, but the milestones are real and publicly reported by METI and the company. BTW, METI is Japan's Ministry of Economy, Trade and Industry. Worth knowing is: it's the successor to MITI, the old Ministry of International Trade and Industry, which was famous in the eighties for steering Japan's industrial rise. Same ministry, renamed.

Japan's industrial policy did indeed lead the nation, but that was a long time ago. Given that today's Japanese elite class is not the elite class of that era, and bears responsibility for Japan's failed industrial and economic policies over the past 30 years, I cannot accept the view that equates the two. industrial policy is no silver bullet either.

Tesselator4. “IBM is a loser who abandoned the technology”? What???
IBM never left advanced research. The 2nm nanosheet GAA technology Rapidus is industrializing originated from IBM Research as just one example. IBM chose not to run its own high-volume fab (a good and rational business oriented decision after exiting pure manufacturing years ago), but it is one of the strongest research organizations in the world on transistor architecture. Technology transfer + joint development is exactly how successful late "entrants" acquire capability and meaningful market share.

IBM certainly did not dare to invest in such a project, but it is a different story if there is easy money from the government, which is also the purpose of their R&D. Do not forget that samsung, trapped in 3gaa with lower density and persistently low yields for years, did not ask IBM for their 2nm technology either.

Tesselator5. Government funding with no profitability yet...
This is industrial policy, not a normal startup. The United States (CHIPS Act), Europe, Korea, and China are all pouring massive public money into leading edge capacity for strategic reasons. Japan is doing the same because relying almost entirely on TSMC (geopolitically concentrated in Taiwan) is a national-security risk. Expecting Rapidus to already be profitable before HVM even begins is an absurd standard that no one applies to anyone else - not Intel Foundry, not Samsung’s advanced nodes, nor even TSMC’s early ramps on new nodes.

I noticed that you are assuming Rapidus can eventually be profitable, which is very bold. The reason why tsmc, intel, and samsung are willing to bear the massive investments in r&d and fab construction is certainly not to end up with a single-wafer production node patronized only by startups, but because IDM exists, and tsmc has Apple.

volume customer to lower risks and assist in improving yields, while traditional hvm(not single wafer) keeping production costs close to industry standards.

If we calculate based on Rapidus's alleged plan of 25k wspm producing for 4 years, after the 5th year of its 2nm node hvm, it produces 25k * 12 * 4 = 1,200k wafers. Calculated at $20k per wafer, it generates only $24b in 5 years.

Tesselator6. “Should have invested in memory instead...” :P
Japan already has significant strength in memory materials, equipment, and (via Kioxia) NAND. The critical strategic gap is leading-edge logic. That is exactly the gap Rapidus is designed to clobber. Diversifying into the one area where Japan is weakest is the likely the most rational "national strategy".

If Japan's goal is to produce an advanced semiconductor looks like tsmc node in a short time to claim Japan as a technologically advanced country, regardless of the health of the node itself, then they can achieve this.
However, maintaining node leadership requires continuous capital investment, investing in an unprofitable advanced node every 2-3 years is a financial money pit with no practical significance.

TesselatorRapidus is a high-risk, high-reward national project. It may or may not fully succeed commercially against/with TSMC or others, but dismissing it with 2020-era foundry "ideology" and outdated facts about its current progress is not serious analysis. Two days ago as this article highlights, they launched Rapidus CORE (Collaborative Open Rapid Ecosystem), a global framework to help customers design and mass-produce 2nm chips. The first piece is seventeen Design Solution Associates, including Synopsys, Cadence, Infosys, and Toshiba, who act as a bridge between customers and Rapidus, handling everything from architecture through tapeout and volume production coordination. The timing seems very deliberate: with mass production targeted for late fiscal 2027 and no committed volume customers yet, they're trying to lower the barrier to entry by making design support part of the package rather than just selling wafers. And as all gamers and CS historians should be aware, this has a been a winning strategy across several notable corporations in several of our favorite industries. Anyway, that's my story, and I'm sticking to it! :D If successful this could be a major win for Japan and even the globe - unless you think it's flat that is. :P

You're right about one thing though... Rapidus's biggest obstacle isn't making 2nm chips; it's convincing customers to bet their supply chain on a company that hasn't proven volume production yet. But if I know the Japanese, that won't be an issue for long - we should see that tension mostly evaporate by mid/late 2027 and be completely eliminated by about a year after that - just 16 to 22 months. 🙏

Meanwhile, keeping a node competitive requires expanding production scale, which demands even higher capital investment. Japan's debt-to-gdp ratio is the highest in the world, and the government recently spent a vast amount of money to maintain the yen exchange rate, which seems to merely provide funds for investors shorting the yen. intel spent $90 billion over 5 years. How much money can Rapidus actually put out when losing money on every node?

Rapidus is a risky idea but the high level of government promotion / investment and joint private investment by the JDM coalition of companies gives it a path to viability.
Selfishly, having another option for leading edge manufacturing is a really good thing in my opinion given the catastrophic levels of dependence on a single company, as amazing as their historical performance has been. It’s going to take a looong time but I hope Rapidus does well and brings Japan back to leading edge semi manufacturing (along with NIL/Canon).
The low throughput is one of the bigger challenges I see for Rapidus among many others (advanced packaging, stacking etc.)