Exclusive: China targets 70% advanced domestic silicon wafer use by 2026

Posted by DazzlingpAd134@reddit | hardware | View on Reddit | 46 comments

According to Nikkei Asia, China is working on more than 70% of the silicon wafers used by domestic chipmakers by the end of this year. This goal demonstrates China’s most aggressive attempt to localize key semiconductor supply chains.

People familiar with the matter pointed out that the Chinese government's goal has become an unwritten rule for domestic chipmakers to give priority to the use of local 12-inch wafers. Although some of the self-sufficiency targets have failed to meet the standards in the past, the industry believes that this is expected to be a success and become an important milestone in China's promotion of supply chain autonomy.

A senior executive in the semiconductor industry, who is familiar with the situation, told Nikkei Asia that in the future, foreign manufacturers will only retain about 30% of the market space. Some Chinese chipmakers are still pursuing more advanced processes, and the field still needs to rely on the support of leading foreign companies. He added that, however, in the field of mature processes and traditional chips, China's local silicon wafers have basically met the market demand.

According to two people familiar with the direct control of the plan, the mainland wafer giant is a material (688783. SH) is currently building new plants in Xi'an and Wuhan, and it is expected to add 700,000 new production capacity per month this year. One of them said that every Chinese customer is expanding wafer production capacity, and Eswell is the most positive of them, potentially accounting for nearly half of the overall expansion. Aiswei said that it has supplied a number of international customers, including Micron (MU.US), TSMC (TSM.US), Grofonde (GFS.US) and UMC.US. In addition, Samsung Electronics (005930) has a large production base in China. KS) with SK Hynix (000660. KS) is also certifying the products of Yeswei. In 2025, the revenue of Aiswei rose to 2.64 billion yuan (about 385 million US dollars), but it has not yet turned a profit.

SMIC (00981. HK, Huahong (01347. HK), as well as the head of China's top memory factory, Xin, and the Yangtze River, are the main customers of Weswei. Aiswei pointed out that its domestic wafers have become the default choice for the expansion of new domestic fabs. ( ec/da)

[-]

Zarmazarma@reddit

The edge in what, chip production? There's not a snowball's chance in hell. TSMC will still be dominant in 2030. They might have caught up by 2040, but no, 2030 is not remotely possible.

[-]

Pimpmuckl@reddit

Yep, it's hilarious how even in a sub like this, people have no idea how a cutting edge chip is made.

The US is, quite literally, not even present in the core chain.

The mirror is from Zeiss, Germany.

The EUV machine from ASML, Netherlands.

The chip is made by TSMC, Taiwan.

The reason Trump made a play for Intel is so those fabs can, one day, catch up to TSMC. And news flash, that doesn't change where the EUV machine comes from.

And the last thing people really are getting wrong is how insanely difficult EUV is. It's the hardest engineering challenge humanity has ever solved. Far harder than space travel. China needs to prove they can make it and commercialise it. And you can't do that on scale with a particle accelerator, the Dutch guys tried that, too.

[-]

dirtyid@reddit

It's hilarious how people drunk on muh EUV is blackmagic koolaid ignores the structure forces that likely going to make entire western semi uneconomic legacy stack in ~10 years. Western semi has no alternate path other than PRC failing, which is bad bet considering their LDP prototype is beating industry estimate timeline and resources being poured into fully indigenize semi.

ASML/EUV LLC consortium never seriously attempted synchrotron/accelerator, they crunched the numbers on synchrotron which was technically superior solution to LPP on paper, but went with LPP because suppose to be cheaper - west bad at infra. Only for LPP to be MAGNITUDE more expensive than projected. Single ASML EUV machines and NA are now the price of of PRC particle accelerators - PRC good at infra. Which again, on fundamental principles, significantly better than LPP (stronger light source, cheaper capex/opex, support multiple tools) than ASML LPP that is now path dependent and locked in terminal scaling:cost. It's exactly scale of SSMB that shares light source between tools that makes it scale. Good chance LPP is radial engine vs SSMB jet engine and ASML on borrowed time with dead/unviable tech tree.

Also the core chain that developed EUV also frankly moderate undertaking: 15k talent from AMSL, 3k from Zeiss, 1k from Cymer, with 5k suppliers. VS something like commercial aviation - Boeing/Airbus are ~100k with 50k suppliers. And ultimately EUV is just physics problem, have enough talent and matter of time before bruteforce solving precision, vs something like aviation which is as much regulatory at whims of geopolitics that paperwork from rival can delay for decades. There's also the consideration that every western semi player projected to have talent shortage in the 100k range, meanwhile PRC only player to be producing enough IC talent to basically replicate/indigenize entire semi supply chain with room for growth. Also entire industry chain for cheaper cleanrooms, fab/data center components, optical connects, energy prices, system appetite to run on commodity margins instead of every player in western semi inflating 50% margin. PRC has so many multipliers ready to stack when 0 turns to 1.

[-]

RogueHeroAkatsuki@reddit

To be fair even in chip production they may be ahead.

Sure, technologically they will not compete with TSMC but in contrast to USA they have extremely healthy energetic sector that can easily uphold more power hungry chips produced on worse manufacturing node.

Also China works like one huge corporation. If they will have decent enough technology then they will initiate production on scale that world never saw. In meantime US government will be debating if authors should pay AI giants for including their books in learning model.

[-]

Typical-Yogurt-1992@reddit

Furthermore, as Moore's Law slows down, a lag of two or three generations can be bridged by simply lowering the voltage.

In the case of AI servers, they don't necessarily need to operate at the same clock speeds as Western hardware. Power consumption is proportional to the square of the voltage. For example, if you downclock a GPU to half its speed, the power consumption drops to one-fourth, which theoretically doubles the efficiency.

(Of course, various bottlenecks exist, so it won't be exactly double in the real world, but by optimizing for clock speeds at half the Western standard, it is possible to significantly increase efficiency.)

While this approach requires doubling the production volume, I think everyone knows that China is exceptionally good at producing less advanced goods in massive quantities.

[-]

wpm@reddit

By 2040 China and Taiwan might not be separate countries anymore.

[-]

monocasa@reddit

Last I heard 2040 is actually what they're shooting for wrt invading Taiwan. The argument I heard is that's when they'll have enough carriers when you account for naval carrier doctrine saying that you only can plan on having half your carriers operational (the other half are in refit, training, etc), and ideally they want to have a spare carrier they can use to check in on other situations like the south china sea so they don't lose appreciable ground on other geopolitical pots.

[-]

hackenclaw@reddit

TSMC doesnt care, they will buy tools from anyone with the most advance tools to make chip. Its profit + staying the lead they are going after.

[-]

F9-0021@reddit

economics and technology in general. You're right that chip manufacturing will take longer, but that will come in time.

pixelpoet_nz@reddit

DJT's decision to restrict Nvidia GPU sales to China might well go down in history than the time he wanted to nuke a hurricane.

Great job, boys.

Well, I need to be honest - I have enough of America first and US big tech so I cheer on China to catch this race fast.

By 2030 China will have the edge, and by 2040 China will be the dominant global superpower. The US threw it all away by throwing all of the money at billionaire CEOs and the military industrial complex instead of investing in the future.

No_Sheepherder_1855@reddit

It’ll be pretty funny when all the billionaires leave the country with the rest of us holding the $40 trillion+ bag.

mictar@reddit

That's exactly what's going to happen. They have multiple citizenships and fickle loyalty and can easily move. For them it'll be like a change of clothes.

IBM296@reddit

By 2030 China will have the edge,

Not really. China will be lucky have a couple of working EUV machines by 2030. Not to mention ASML is already moving on to next gen High NA EUV machine.

Though China might have an edge by 2040 if it keeps on advancing at the current pace.

Restricting Nvidia GPU sales in 2026 didn't really have an effect. The ball was sent rolling in 2019 when Huawei was sanctioned and China unable to purchase EUV machines.

Now it's too late to stop China from achieving self sufficiency in high-end chip manufacturing.

Were they able to replicate EUV?

Not yet. Will probably do so by 2030..

HulksInvinciblePants@reddit

And by then ASML and TSMC will be on to the next. The real question is will these two ever stagnate to allow China to catch up.

abiotic_selection_2@reddit

You don't seem to understand the purpose of these chips.

They are needed to produce vehicles, factories, government infrastructure, household appliances, network infrastructure, computers for businesses and people, cameras, phones and also weapons and surveillance equipment

They want to be able to build all of these themselves without being reliant on foreign companies and foreign governments.

Incremental improvement dick waving about a process being 10 percent more power efficient is irrelevant because again they are trying to make stuff to use domestically, not compete in marketing charts against other companies to win market share.

Any semi modern process is more than good enough to build all of those things. Which means they can meet their needs, which is their goal.

The semiconductor race is arbitrary bullshit that is used for planned obsolescence or fomo 99 percent of the time, especially for the past ten years where performance and efficiency improve by 5-10 percent every 2 years and maybe 2-3x over a decade, instead of by 2x a year and 1000x over a decade like in the 90s

ju2au@reddit

The most advanced process is currently 2nm. I think that we are running out of nm to go any lower.

Honestly it would have helped the general public loads if these fabs used truthful feature sizes for their nodes.

Might as well call them something completely random at this point, say, the two-steve-process.

The problem is ASML / western stack is locked into LPP for future, which itself was only EUV bet that cost magnitude more than projected with lethal scaling costs that PRC may leapfrog if they sort out SSMB. The only western alternative is JP nanoimprint (maybe FEL), but white paper first principles, SSMB economics makes LPP or NIL none competitive.

Eastern_Ad6546@reddit

which is what? tsmc won't even take high-na or hyper-na machines.

Most of the development is now on packaging which china isn't behind in.

oursland@reddit

They got the same performance out of advanced DUV with the equipment they have, which was a known process that wasn't pursued by others for commercialization when EUV was available. This is why ASML's DUV equipment is now restricted from China.

EUV via particle accelerators has been demonstrated in China, but it's unknown if they've managed to employ this in lithography.

vhu9644@reddit

Wait, there is a demonstration of SSMB? Do you have a link?

zdy132@reddit

It's had two prototype/demonstation and is on the third implementation now iirc.

The english news on this is sparse, especially after 2023, which is to be expected. There are some Chinese industrial news and web discussions on the third stage. It looks like the construction started in 2023. And considering the last one took 6.5 years to complete construction, and this one is still a research purposed proof of concept, SSMB still has a long way to go.

The Tsinghua SSMB team doing IPAC presentations in a few weeks. It's useful to dig through researchgate of their SSMB team, some of the leads release papers like every few months. Far from industry expert, but if you have access to the studies, throw them into deep research and ask for technologic readiness level assessment. SSMB seems to be at TRL4, i.e. past proof of principle / lab validation, probably component integration or integrated system design. Seems like they're putting XiongAn facility together I'd expect first SSMB first light 2027/2028.

Thanks for the information, I didn't think of searching for the specific researchers.

Best of luck to them, hopefully we will have great competitions in the coming decade.

Oh interesting.

Them getting any EUV source would probably drastically accelerate their progress though, as it means they can parallelize the research into all the things that interface with it more effectively.

The landscape does look interesting, with them trying SSMB and LDP

Yeah competition would be nice. I've been trying and failing to buy a 5090 at MSRP. It would be great to have more alternatives in the future.

benefit420@reddit

Umm no they didn't? Do you have a source that shows DUV is equivalent to EUV? Also I heard yields are less than 30%.

Immersion DUV unlocks 7nm, although yields are under 50%.

That's 7nm. China can't go to 5nm without EUV.

And for below 1nm, companies need ASML's even more advanced High NA EUV machines.

Lianzuoshou@reddit

Chinese SMIC Achieves 5 nm Production on N+3 Node Without EUV Tools

StickiStickman@reddit

China can't go to 5nm and below without EUV.

Meanwhile last year:

SMIC Achieves 5 nm Production on N+3 Node Without EUV Tools

Brilliant_Run8542@reddit

Through quad patterning you can get close. Yields will be in the toilet though

WhoTheHeckKnowsWhy@reddit

no, but they did managed to whip DUV down to a usable 5nm equivalent.

Indeed, a lot of really upset people who seem to think that the process is the most important thing, when it's the product. No one really gives a damn about the lithography technology, it's the chips that matter. China has them now and are on target to match existing capabilities in very short order.

Wait_for_BM@reddit

Functionality of a product depends the actual logic that implements a design not the "node size". Without the necessary IP you can't implement the design. That's something they need to expand.

"Node size" defines density, performance - power/speed and density. The product might not be the "best", but it might be good enough. They are at the chock point because of geopolitical reasons and will have to work around that. They can get to a limited node size with older process just that the cost/yield might not be there.

they dont even need to rush EUV, they will destroy the DUV market first and the Western semiconductor will force to live with just EUV's profit.

That is a huge chunk of profit, investing R&D is to shrink semiconductor further is getting more expensive too. That means the leading node companies will have less fund affecting their progress further due to reduce profit from DUV + more expensive to advance further.

At some point, these semiconductor companies will be handing their IP on a silver place for money due to loss income.

kongweeneverdie@reddit

Nobody know. Now their chips at 5nm level.

CanLongjumping9360@reddit

Technically, China is 8 years behind in foundry and 5 years behind in memory. This represents a significant gap in the semiconductor industry. In fact, without subsidies, all Chinese semiconductor companies would go bankrupt. Moreover, DUV equipment will be essential in the future, but access is currently blocked.

ML7777777@reddit

The biggest losers in the shift towards a fully domestic semiconductor supply chain continues to be Japan. Unless they really hit it off with Rapidus, Japanese supply companies will continue to lose customers and sales. Largest percentage of silicon wafers used in China come from Japan.

pendelhaven@reddit

https://archive.ph/SwofN

allahakbau@reddit

National security baby