Artificial Intelligence Latest news

When bits become light: Q.ANT ignites Europe’s optical AI flagship project – and doesn’t just play into TSMC’s gears

Somewhere between the romance of mechanical engineering and the fantasy of saving the world with AI, a company from the German state of Baden-Württemberg is working on the future of information processing. Q.ANT, a spin-off of the Swabian Trumpf Group, has pulled out a card that one would hardly have expected on the global semiconductor market: photonic chips – not on PowerPoint, but as real PCIe cards with computing power that packs a punch. Not in watts, but in wavelengths.

 

Light instead of electricity: from vision to hardware

What sounds like a laser show and research demo is actually a potential affront to what Nvidia, AMD and Intel sell as modern computing power. While the big players are blowing up their silicon monsters with even more layers and fans, the Swabians are taking a different approach: light does the math – low-loss, passively cooled and so efficient that you wonder why nobody took it seriously before. The star of the show? A native processing unit, i.e. an optical co-processor that works in tandem with classic CPUs/GPU systems. The processing unit itself is based on lithium niobate waveguides – a material previously known mainly from telecommunications. Now as a computing unit for Fourier transformations, convolution kernels and even AI inference. Not emulated, not simulated – calculated for real. With light. And the whole thing runs on converted 200 mm lines in Stuttgart. No TSMC, no ASML gold dust, no 3 nm circus. Just good old equipment, cleverly upgraded.

Pragmatism beats billions: The factory as a statement

You can grumble about subsidies, or you can do it like Q.ANT. 14 million euros of investment, a few smart retrofits – and a pilot line that produces 1000 wafers per year with photonic logic is up and running. A kind of anti-Intel strategy: no billions for desert factories, no diplomatic kowtowing to Asia – but local expertise, rebuilt with technical elegance. In times when “strategic autonomy” is propagated at EU coffee parties, Q.ANT shows that it is possible to keep high-tech in Europe with a manageable budget. No hype, no buzzword-speak – just good work.

Computing with photons: Fairy tale or math?

According to Q.ANT, the technology saves up to 90% energy compared to classic AI accelerators – with 40-50% fewer operations per cycle, as the light logic performs certain transformations natively in one step. FFT? No problem. Activation? Optically. Convolutions? Directly in the waveguide structure. It all sounds too good to be true – and yes, you can remain skeptical. Because the real art is not in the chip, but in the environment: compilers, toolchains, drivers, debuggers. Everything has to be new, everything has to be compatible. Nvidia has built up an ecosystem here that functions almost monopolistically – Q.ANT wants to break through the wall. Elegant, but also a bit kamikaze.

Source: Q.ANT

HPC as an entry drug

But: The market for High Performance Computing is changing. Energy consumption and cooling capacity are growing faster than the benefits. Data centers are desperately looking for solutions that do not collide with the power grid. This is precisely where Q.ANT is positioning itself as a silent revolution: photonically accelerated AI inference combined with classic infrastructure. A prototype of the Native Processing Server ran at ISC 2025 – with impressive demos. Analysts from Hyperion Research are already talking about the “technological realignment of the HPC economy”. If that’s not a declaration of war.

Geopolitics meets physics

Europe likes to moan about tech dependency – Q.ANT does. No billions from the state, no big announcements – simply a functioning proof of concept in the middle of Baden-Württemberg. And all this while Intel is discussing cement price increases in Magdeburg and TSMC is treating Europe like an annoying secondary market. Q.ANT’s strategy is not a start-up roulette, but a targeted demonstration of power: “We are specifically replacing individual tools in the fab flow – minimum conversion costs, maximum effect,” says CEO Michael Förtsch. That’s not a disruptive slogan – it’s Baden-Württemberg realism with a laser.

Between vision and caution

The market is brutal. Nvidia dominates not only technologically, but also psychologically. Developers, infrastructure, software – everything is geared towards CUDA. Even AMD is struggling to gain a foothold in this enclave. And then Q.ANT comes along with a system that reads like quantum magic. But it should not be underestimated. The technology is there, the use case is there, the political momentum is favorable. And the ability to achieve a lot with little has always been a German virtue – if you let them.

Source: Q.ANT

Kommentar

Lade neue Kommentare

big-maec

Urgestein

1,051 Kommentare 624 Likes

Mal einige Fragen, wie schnell muss denn die CPU Board und RAM sein damit die KI Karte vernünftig läuft?

Kann man auch mehrere Karten kombinieren?

Antwort Gefällt mir

komatös

Veteran

153 Kommentare 128 Likes
e
eastcoast_pete

Urgestein

2,609 Kommentare 1,730 Likes

Schon eine tolle Idee, bzw eine ganze Reihe toller Ideen. Nicht ganz unähnlich dem, was auch zB Google versucht, allerdings scheint Quant sogar schon weiter gekommen zu sein. Die Befürchtung die ich hier habe ist, was passiert wenn die Branchenriesen - v.a. Nvidia, Meta, Google oder auch Hyperscalers wie Microsoft - denken bzw merken, daß da in der Tat was dran ist. Denn dann werden die sich die Technik und die Firma krallen, d.h. das dicke Checkbuch kommt raus, und Quant wird einfach gekauft. Und weg ist sie, die Technik aus Deutschland. Für Firmen, die zur Zeit mehrere Billionen Dollar (im deutschen Gebrauch, also 1000 Milliarden) wert sind, sind selbst Investitionen im Milliarden Dollar Bereich einfach "cost of doing business".

Antwort 3 Likes

H
Headyman

Veteran

172 Kommentare 92 Likes

Wundert mich jedenfalls nicht, Trumpf macht vieles anders in ihrer Branche als die Mitbewerber.

Interessant wird sein, wo die Grenzen/Einschränkungen des Systems liegen. Bisherige optische Rechner sind meist stark eingeschränkt in dem Einsatzgebiet, und wenig flexibel. Und das machte die teuer und verhinderte breitere Einsatzgebiete.

Schönes Beispiel, wo die Staatsgelder hin müssten. Stattdessen werden bald mit 300 Mrd. Subventionen und Steuergeschenken unproduktive und innovationsträge Industrien am Leben erhalten.

Ich fürchte, wenn das was taugt, dann werden die massiv hochskalieren müssen in der Fertigung. Das ist teuer, also muss da jemand mit viel Geld investieren. Und wie das typ. in Deutschland ausgeht...

Antwort 4 Likes

Danke für die Spende



Du fandest, der Beitrag war interessant und möchtest uns unterstützen? Klasse!

Hier erfährst Du, wie: Hier spenden.

Hier kannst Du per PayPal spenden.

About the author

Samir Bashir

As a trained electrician, he's also the man behind the electrifying news. Learning by doing and curiosity personified.

Werbung

Werbung