It's not just Infineon - it's called the European Semiconductor Manufacturing Company (ESMC) and is a joint venture by TSMC, Bosch, Infineon, and NXP, with TSMC being the majority (70%) shareholder.
I've met one of the engineers designing the piping for that plant. Hardest project to date for him and mainly because TSMC was setting the pace.
You're right, honest mistake. I checked this several times, thinking "surely there aren't two chip plants built at the same time, in roughly the same location with the same company involved?".
Infineon is betting big on the 800V dc power distribution that seems to become the new standard for AI data centers which is directly relevant for the chips that are made in this fab.
"Process node," in the sense of logic, RAM or Flash density, isn't relevant here. These are power devices. Silicon Carbide FETS and whatnot; bulk current switching and related devices. Not frontier process node logic devices.
That reality is carefully left unclear in all this "silicon sovereignty" narrative. It's a nice plant with new tooling cranking out 300mm wafers, but it's not the same game as a TSMC or Intel fab making cutting edge, high margin silicon, and there are a number of competitors making similar power devices around the world. And yes, the "AI" fluff is pure marketing nonsense; everyone needs lots of power devices for everything. Yes, they'll obviously seek lucrative contracts to supply exotic power devices for AI applications, but that stuff gets commoditized quickly.
Not directly. This fab is meant to primarily fabricate compound semiconductors which is Infineon's niche and is a major bottleneck for European industry today.
> Does everything need to have the word AI these days
Because Infineon's press release [0] for their compound semiconductor fab called out "AI".
Additonally, the "semiconductor" and "hardware" segment has now been rebranded has "AI" in a number of funds. By calling out something that's even tenuously tied to "AI", it allows funds that are contractually tied to investing in "AI" to purchase Infineon stock.
Also suitable for keeping an economy functioning and weaponry in war.
You might not be able to fabricate billion terraflop GPUs but at least the basics of survival will be able to be locally produced without scavenging washing machines for parts.
The (western) economy runs on sub 7nm phone, laptop and datacenter chips on which the white collar workforce produces value. Those are the ones that are also the most profitable since they have the highest margins. Europe doesn't have that.
Phones and laptops from a few years ago are not suddenly unusable.
Yes, the cutting edge is very nice, but any laptop past 2016 is useable for the average person. Even gpu inference on older process nodes is perfectly doable. The HPC space absolutely prefers newer chips but hasn't ripped out their 2018 chips in their clusters because they still deliver value.
And sure, the latest best things sells for higher margins for now, but with the way consumer prices are going, people may start choosing older still perfectly capable models that cost less.
The greater danger to a working economy is not absence of the absolute must cutting edge chips but lack of independence which this initiative seems to seek to curtail. Good for Germany.
> these are pretty low-tech chips only for industrial uses
I don't like this framing.
These aren't logic ICs but that doesn't mean they are useless or "easy".
Heck, the only countries with Gallium Nitride fabrication capabilities and knowhow are the US, China, Taiwan, Japan, South Korea, Germany, and India.
In fact, compound semiconductors and power electronics is one segment where Europe's China dependency is extremely high, as they have significant uses from automotive to PLCs to weapons systems, and China has already begun embargoing the EU's access to rare earth elements [0] and has begun enforcing sanctions against the EU's aerospace and UAV industry [1].
These are dual use technologies and a major reason why both the US and China heavily invested in compound semiconductor capacity in the late 2010s and early 2020s.
Edit: can't reply
> In 2024 Belgium closed its only semi fab, which had recently pivoted to GaN
THEMA Foundries is working in photonics, not GaNs [2]. The GaN initiative (BelGaN) failed and the only two buyers interested in buying out the property for GaN fabrication were a Chinese and Indian player [2].
>Heck, the only countries with Gallium Nitride fabrication capabilities and knowhow are the US, China, Taiwan, Japan, South Korea, Germany, and India.
So basically all major industrial powers? In which case I don't get the use of the world "only" here, as if the EU, the richest block in the world, deserve praise for doing something countries a lot less wealthier are doing.
> and China has already begun embargoing the EU's access to rare earth elements [0] and has begun enforcing sanctions against the EU's aerospace and UAV industry [1].
The way it's framed it sounds like China is some evil bad guy for doing that but that's standard practice that the EU and US also do. The EU also restricts ASML EUV machines to China and sanctions Chinese tech in their defense sector. Standard stuff.
It's actually the US that restricts ASML from selling EUV machines to China. The EUV light source itself is American technology developed and produced by Cymer in California. ASML was only permitted to acquire Cymer in 2014 under a strict technology sharing and export control agreement with the US government.
It's why, for instance, ASML's next generation research lithography machine is currently being installed in upstate New York and not somewhere in the Netherlands.
ASEAN doesn't have the capacity yet, but this is changing in 3-5 years as Malaysia, Singapore, Vietnam have gotten IP transfers from European, Korean and Japanese technology partners.
Neither does the rest of the Europe excluding an Ireland-UK-UAE JV called ChipX (but the team is largely located in the US, UAE, Malaysia, and Japan) and the portion of STMicro in France that was part of state-owned Thomson Semiconducteurs before it was privatized.
Neither do any of the major industrial states in the Americas like Canada, Mexico, or Brazil.
> that's standard practice that the EU and US also do
My point is that states need to build domestic capacity where possible. And the EU is not a state. France continued to protect their GaN fabrication IP closely (not even sharing it with Italy despite STMicro being a French-Italian JV), and same with Germany to a certain extent.
“Only for industrial uses” is kind of a crazy thing to say when you say it out loud, don’t you think? You might as well say “only for having real-world impact”.
You do not need a 2nm processor to run an engine, perform an ultrasound, drop a missile on someone's head, make 10 million parts in a factory, or fly a plane. The vast majority of the increased processing power we've developed since the mid-2010s gets pissed away rendering 30MB websites or generating AI cheating fruit husband videos. Every phone SoC release for the last 10 years combined has done less in the real, concrete world we live in than industrial controls hardware from 30 years ago.
We can vibecode SaaS junk that we will have all forgotten about this time next year at never-before-seen speed, but every single day, hundreds or thousands of times a day, you brush up against something that was built by an AB or Siemens PLC installed in 2004 that might be slated for decommissioning in a decade or so.
Designing a secure platform is possible within the EU[1].
[1]They need to use US EDA tools, And manufacture masks but maybe there are tricks they won't need to trust them - like inspecting critical parts of the masks.
It's not just high end CPUs that use the latest processes. Power, Performance and Area is important to all chips, including microcontrollers, FPGA, etc.
Design capacity doesn't imply fabrication capacity, as can be seen with Israel and India's comparative dominance in the chip design industry.
Design capacity (basically programming and logic design) is orthogonal to front-end fabrication (basically material science and chemical engineering) which is orthogonal to back-end/OSAT (basically materials science and metrology).
Only the US and Taiwan have domestic E2E capacity in all 3.
If I was Queen of Germany, I would put everything into a platform like Arduino or RaspPi which can be widely used across industries and education. Some amount of taxes will be allocated to buy these boards and guarantee demand for manufacturing. Then every citizen gets such a device in return. Basically mandatory purchase. The education sector can completely lean in on the platform, abundance will have it dominate DIY projects and may create additional demand as reference platform. Whoever doesn't need it, can sell it or give it away to charity. This will stabilize a critical industry and aid digital education, engineering and so on. I feel like people would be more easy about a tax, if they get a physical product in exchange.
Over time this program would be extended to include and bring back other critical industries and manufacturing capabilities, ultimately leading to citizens being able to choose their mandatory product to some extent and preference. For example it would be really cool to have a basic, but very robust and repairable sewing machine, 3D printer, ... which all aid survivability/adaptability of the collective in crisis, if widely distributed. These products would also set the baseline for quality and accessibility expectations.
Of course this goes hand in hand with a 4 day work week, so people can actually learn to appreciate their mandated crisis hobbies and indulge their family and friends doing so. And if all of this doesn't pan out economically, I would simply plunder and enslave a neighboring country <3
As I mentioned about this before [0], this is a compound semiconductor fab - a very critical bottleneck for European industry and a much more worrisome NatSec issue than sub-14nm logical chip fabrication or arguably even AI.
This is not directly related to AI or logical compute, so kvetching about GPUs, SoCs, TSMC, AI, and other buzzwords is dumb.
I would also like to see local PCB manufacturers - pcbway etc like. Modern production facilities might even be locally competitive given the amount of automation that can be had.
They should, but sadly it's extremely difficult for PCB board manufacturing to return to Europe.
EU has FTAs with Japan and SK, and others that dominate the segment like Taiwan, China, Thailand, Malaysia, Vietnam, and India have already unlocked public-private subsidized for the sector.
Additionally, the big players in the industry like ZDT, Unimicron, Nippon Mekatron, Foxconn, Compeq, TTM, and Flex have much stronger financial and political linkages in Asia or the Americas.
This fab itself is important, but was extremely difficult to stand-up and was largely a result of the supply chain issues that the automotive industry faced during zero covid, so it basically took 6 years to execute on this project. That lag-time is the biggest issue unless individual European states decide to take industrial policy their own hands, which becomes expensive very quickly.
Concentrating on building a niche in compound semiconductors as well as 2.5/3D packaging would probably be the best bet for the EU today, but I expect to see French-German industrial rivalry to undermine coordination.
Let's see the state of this project in 5 years. Experience tends to show European projects dragging forever and then suddenly closing whilst funds mysteriously move into semi public companies with boards full to the brim with retired political figures
EDIT: Apparently it's a different plant.
It's not just Infineon - it's called the European Semiconductor Manufacturing Company (ESMC) and is a joint venture by TSMC, Bosch, Infineon, and NXP, with TSMC being the majority (70%) shareholder.
I've met one of the engineers designing the piping for that plant. Hardest project to date for him and mainly because TSMC was setting the pace.
These are different plants, that are situated in the same quarter.
https://www.openstreetmap.org/way/1191197175 vs. https://www.openstreetmap.org/way/583767292
You're right, honest mistake. I checked this several times, thinking "surely there aren't two chip plants built at the same time, in roughly the same location with the same company involved?".
Apparently there are.
> Hardest project to date for him and mainly because TSMC was setting the pace.
I can imagine their pace might be accelerated but better than endless discussions, shuffling and never shipping
Isn't that TSMC joint venture another plant that will open next year?
>"European" Semiconductor Manufacturing Company (ESMC)
>Looks inside
>It's all TSMC
FFS, we can't even come up with an original name, they just replaced Taiwan with European.
US didn't call TSMC's Arizona fab, the American Semiconductor Manufacturing Company (ASMC)
> The plant will produce chips for intelligent power management
> The company ... sought to capitalise on the massive AI investment boom
These chips are probably very useful and important, but I don't see what they have to do with AI. Does everything need to have the word AI these days?
Infineon is betting big on the 800V dc power distribution that seems to become the new standard for AI data centers which is directly relevant for the chips that are made in this fab.
Hopefully they can use that tech to expand the EV fast charger network once the AI bubble pops.
I think this fab was already in construction before the AI hype, this is just marketing.
Very bad marketing since the slop article doesn't even mention the process node of the fab.
"Process node," in the sense of logic, RAM or Flash density, isn't relevant here. These are power devices. Silicon Carbide FETS and whatnot; bulk current switching and related devices. Not frontier process node logic devices.
That reality is carefully left unclear in all this "silicon sovereignty" narrative. It's a nice plant with new tooling cranking out 300mm wafers, but it's not the same game as a TSMC or Intel fab making cutting edge, high margin silicon, and there are a number of competitors making similar power devices around the world. And yes, the "AI" fluff is pure marketing nonsense; everyone needs lots of power devices for everything. Yes, they'll obviously seek lucrative contracts to supply exotic power devices for AI applications, but that stuff gets commoditized quickly.
> don't see what they have to do with AI
Not directly. This fab is meant to primarily fabricate compound semiconductors which is Infineon's niche and is a major bottleneck for European industry today.
> Does everything need to have the word AI these days
Because Infineon's press release [0] for their compound semiconductor fab called out "AI".
Additonally, the "semiconductor" and "hardware" segment has now been rebranded has "AI" in a number of funds. By calling out something that's even tenuously tied to "AI", it allows funds that are contractually tied to investing in "AI" to purchase Infineon stock.
Investor relations is important as well.
[0] - https://www.prnewswire.com/news-releases/infineon-opens-the-...
there was a buzzword bingo and their application reached Bingo! first.
My understanding is that these are pretty low-tech chips only for industrial uses?
Also suitable for keeping an economy functioning and weaponry in war.
You might not be able to fabricate billion terraflop GPUs but at least the basics of survival will be able to be locally produced without scavenging washing machines for parts.
>Also suitable for keeping an economy functioning
The (western) economy runs on sub 7nm phone, laptop and datacenter chips on which the white collar workforce produces value. Those are the ones that are also the most profitable since they have the highest margins. Europe doesn't have that.
Phones and laptops from a few years ago are not suddenly unusable.
Yes, the cutting edge is very nice, but any laptop past 2016 is useable for the average person. Even gpu inference on older process nodes is perfectly doable. The HPC space absolutely prefers newer chips but hasn't ripped out their 2018 chips in their clusters because they still deliver value.
And sure, the latest best things sells for higher margins for now, but with the way consumer prices are going, people may start choosing older still perfectly capable models that cost less.
The greater danger to a working economy is not absence of the absolute must cutting edge chips but lack of independence which this initiative seems to seek to curtail. Good for Germany.
A drone will work with 30nm chips just fine, but it won't work with no chips at all
Europe's economy has a higher share in the industrial and agricultural sector. The US isn't the sole "western" benchmark.
https://en.wikipedia.org/wiki/List_of_countries_by_GDP_secto...
they may low-density, but low-tech?
> these are pretty low-tech chips only for industrial uses
I don't like this framing.
These aren't logic ICs but that doesn't mean they are useless or "easy".
Heck, the only countries with Gallium Nitride fabrication capabilities and knowhow are the US, China, Taiwan, Japan, South Korea, Germany, and India.
In fact, compound semiconductors and power electronics is one segment where Europe's China dependency is extremely high, as they have significant uses from automotive to PLCs to weapons systems, and China has already begun embargoing the EU's access to rare earth elements [0] and has begun enforcing sanctions against the EU's aerospace and UAV industry [1].
These are dual use technologies and a major reason why both the US and China heavily invested in compound semiconductor capacity in the late 2010s and early 2020s.
Edit: can't reply
> In 2024 Belgium closed its only semi fab, which had recently pivoted to GaN
THEMA Foundries is working in photonics, not GaNs [2]. The GaN initiative (BelGaN) failed and the only two buyers interested in buying out the property for GaN fabrication were a Chinese and Indian player [2].
[0] - https://www.reuters.com/world/china/eu-firms-brace-more-shut...
[1] - https://www.scmp.com/economy/global-economy/article/3351292/...
[2] - https://www.semiconductor-today.com/news_items/2025/apr/belg...
In 2024 Belgium closed its only semi fab, which had recently pivoted to GaN.
>Heck, the only countries with Gallium Nitride fabrication capabilities and knowhow are the US, China, Taiwan, Japan, South Korea, Germany, and India.
So basically all major industrial powers? In which case I don't get the use of the world "only" here, as if the EU, the richest block in the world, deserve praise for doing something countries a lot less wealthier are doing.
> and China has already begun embargoing the EU's access to rare earth elements [0] and has begun enforcing sanctions against the EU's aerospace and UAV industry [1].
The way it's framed it sounds like China is some evil bad guy for doing that but that's standard practice that the EU and US also do. The EU also restricts ASML EUV machines to China and sanctions Chinese tech in their defense sector. Standard stuff.
> EU also restricts ASML EUV machines to China
It's actually the US that restricts ASML from selling EUV machines to China. The EUV light source itself is American technology developed and produced by Cymer in California. ASML was only permitted to acquire Cymer in 2014 under a strict technology sharing and export control agreement with the US government.
It's why, for instance, ASML's next generation research lithography machine is currently being installed in upstate New York and not somewhere in the Netherlands.
> So basically all major industrial powers
ASEAN doesn't have the capacity yet, but this is changing in 3-5 years as Malaysia, Singapore, Vietnam have gotten IP transfers from European, Korean and Japanese technology partners.
Neither does the rest of the Europe excluding an Ireland-UK-UAE JV called ChipX (but the team is largely located in the US, UAE, Malaysia, and Japan) and the portion of STMicro in France that was part of state-owned Thomson Semiconducteurs before it was privatized.
Neither do any of the major industrial states in the Americas like Canada, Mexico, or Brazil.
> that's standard practice that the EU and US also do
My point is that states need to build domestic capacity where possible. And the EU is not a state. France continued to protect their GaN fabrication IP closely (not even sharing it with Italy despite STMicro being a French-Italian JV), and same with Germany to a certain extent.
“Only for industrial uses” is kind of a crazy thing to say when you say it out loud, don’t you think? You might as well say “only for having real-world impact”.
And 3nm CPUs, GPUs, phone SoCs don't have "real world impact"?
One could say they have an even bigger real world impact.
You do not need a 2nm processor to run an engine, perform an ultrasound, drop a missile on someone's head, make 10 million parts in a factory, or fly a plane. The vast majority of the increased processing power we've developed since the mid-2010s gets pissed away rendering 30MB websites or generating AI cheating fruit husband videos. Every phone SoC release for the last 10 years combined has done less in the real, concrete world we live in than industrial controls hardware from 30 years ago.
We can vibecode SaaS junk that we will have all forgotten about this time next year at never-before-seen speed, but every single day, hundreds or thousands of times a day, you brush up against something that was built by an AB or Siemens PLC installed in 2004 that might be slated for decommissioning in a decade or so.
Germany doesn't design high end SoCs or x86 processors so why would they build a fab for them?
Designing a secure platform is possible within the EU[1].
[1]They need to use US EDA tools, And manufacture masks but maybe there are tricks they won't need to trust them - like inspecting critical parts of the masks.
It's not just high end CPUs that use the latest processes. Power, Performance and Area is important to all chips, including microcontrollers, FPGA, etc.
Bad take.
Design capacity doesn't imply fabrication capacity, as can be seen with Israel and India's comparative dominance in the chip design industry.
Design capacity (basically programming and logic design) is orthogonal to front-end fabrication (basically material science and chemical engineering) which is orthogonal to back-end/OSAT (basically materials science and metrology).
Only the US and Taiwan have domestic E2E capacity in all 3.
If I was Queen of Germany, I would put everything into a platform like Arduino or RaspPi which can be widely used across industries and education. Some amount of taxes will be allocated to buy these boards and guarantee demand for manufacturing. Then every citizen gets such a device in return. Basically mandatory purchase. The education sector can completely lean in on the platform, abundance will have it dominate DIY projects and may create additional demand as reference platform. Whoever doesn't need it, can sell it or give it away to charity. This will stabilize a critical industry and aid digital education, engineering and so on. I feel like people would be more easy about a tax, if they get a physical product in exchange.
Over time this program would be extended to include and bring back other critical industries and manufacturing capabilities, ultimately leading to citizens being able to choose their mandatory product to some extent and preference. For example it would be really cool to have a basic, but very robust and repairable sewing machine, 3D printer, ... which all aid survivability/adaptability of the collective in crisis, if widely distributed. These products would also set the baseline for quality and accessibility expectations.
Of course this goes hand in hand with a 4 day work week, so people can actually learn to appreciate their mandated crisis hobbies and indulge their family and friends doing so. And if all of this doesn't pan out economically, I would simply plunder and enslave a neighboring country <3
Infinity neon! A nice company name.
As I mentioned about this before [0], this is a compound semiconductor fab - a very critical bottleneck for European industry and a much more worrisome NatSec issue than sub-14nm logical chip fabrication or arguably even AI.
This is not directly related to AI or logical compute, so kvetching about GPUs, SoCs, TSMC, AI, and other buzzwords is dumb.
[0] - https://news.ycombinator.com/item?id=48557914
I would also like to see local PCB manufacturers - pcbway etc like. Modern production facilities might even be locally competitive given the amount of automation that can be had.
Does AISLER[1] do the same thing as pcbway? They seem to be based in Germany/EU
[1] https://aisler.net/en
Looks good - will have to give them a go!
Netherlands actually
Company might be registered in NL but phone number + hosting happens in Germany => maybe only tax reducing?
They should, but sadly it's extremely difficult for PCB board manufacturing to return to Europe.
EU has FTAs with Japan and SK, and others that dominate the segment like Taiwan, China, Thailand, Malaysia, Vietnam, and India have already unlocked public-private subsidized for the sector.
Additionally, the big players in the industry like ZDT, Unimicron, Nippon Mekatron, Foxconn, Compeq, TTM, and Flex have much stronger financial and political linkages in Asia or the Americas.
This fab itself is important, but was extremely difficult to stand-up and was largely a result of the supply chain issues that the automotive industry faced during zero covid, so it basically took 6 years to execute on this project. That lag-time is the biggest issue unless individual European states decide to take industrial policy their own hands, which becomes expensive very quickly.
Concentrating on building a niche in compound semiconductors as well as 2.5/3D packaging would probably be the best bet for the EU today, but I expect to see French-German industrial rivalry to undermine coordination.
Funny that the article didn't mention it.
Infineon got €1bn of tax payer money to open the plant (~$1.1bn).
The article says:
> The facility was backed by the EU's Chips Act with one billion euros in subsidies
There are worse ways to spend tax payers money.
Let's see the state of this project in 5 years. Experience tends to show European projects dragging forever and then suddenly closing whilst funds mysteriously move into semi public companies with boards full to the brim with retired political figures
Still better odds of success than if that money was used to pay welfare for unskilled migrants instead.
Yeah, most europeans are not aware how much do we spend on agriculture, pensions and healthcare.