Over a billion queries a day are now answered by LLMs.

A media rich ChatGPT prompt can consume up to 10x the energy of a Google search - a small act of magic powered by sprawling thermoelectric furnaces.

And this future is only going to get hotter.

AI data centre power consumption is expected to double by 2030. Cooling alone accounts for 40% of a data center's energy bill.

We're hitting a wall. As more transistors are crammed into ever-smaller spaces, waste heat becomes the inevitable bottleneck.

While the industry chases larger GPUs and more power hungry silicon, a team in Stuttgart has been building something fundamentally different.

Q.ANT computes with light instead of electrons, sidestepping the thermal and physical constraints that plague conventional chips.

The numbers tell the story. Q.ANT's photonic chips can achieve:

• 30x greater energy efficiency on chip level
• 90x lower power consumption per application, and
• 99.7% accuracy across complex computational tasks.

Where traditional systems need millions of transistors to perform, Q.ANT accomplishes it with a single optical element.

At Cherry, we're backing them as they bring their photonic AI accelerators to market - technology that could reshape the fabric of AI computing - cooler, faster, and built not on heat, but light.

The Light Solution

Q.ANT is working on photonic computer chips, which calculate using light (photons) instead of electricity (electrons).

Typical chips use semiconductors like silicon, filled with millions - even billions - of transistors acting as tiny switches. They work digitally, using just two states - 0 and 1.

But photonic chips work analog. They manipulate light waves using lenses, splitters, and other optical components to encode and process information.

Light waves have more “degrees of freedom” than electrons:

• Amplitude (wave height)
• Phase (position)
• Polarisation (oscillation direction)

By layering and interfering with these light waves (think: stacking patterns), you can compute. Each resulting interference pattern represents the result of a mathematical operation.

This approach from Q.ANT sidesteps the fundamental constraints of traditional computing.

Their Native Processing Unit leverages thin-film lithium niobate to execute complex mathematical operations in the optical domain, delivering results that seem almost impossible by conventional standards.

The breakthrough extends beyond efficiency. Q.ANT's chips excel at native non-linear computing, something traditional CMOS (Complementary Metal-Oxide-Semiconductor) architectures struggle with.

While digital processors approximate non-linear functions using millions of parameters, photonic systems can execute sine waves, exponentials, and other complex functions directly in hardware. Michael Förtsch (Q.ANT CEO) calls it “native computing” - the math is baked into the physics of light and on the chip itself.

This capability becomes crucial for the next-generation of AI.

Q.ANT can train image recognition systems with 0.1 million parameters and 0.2 million operations, while conventional approaches require 5.1 million parameters and 10 million operations to achieve similar results.

Perhaps most importantly, photonic chips generate minimal heat.

Q.ANT's systems require no active cooling infrastructure, enabling 100x greater compute density per data centre rack while consuming a fraction of the power.

Building the Impossible in Stuttgart

Dr. Micheal Förtsch founded Q.ANT as a spin-off out of the R&D Labs at TRUMPF, one of Germany’s industrial giants. In his start-up, he developed the core technology with the potential to rewrite computing fundamentals.

The key insight came from understanding materials. Silicon - perfect for electrical operations - absorbs light, making precise optical modulation (controlling how light behaves to carry information) nearly impossible.

Michael’s solution was lithium niobate - a crystalline material that generates almost no heat and allows for extremely precise control of light signals.

Q.ANT has since invested €14 million repurposing an existing semiconductor line into a photonic chip pilot facility (a smaller-scale production line designed to prove new manufacturing methods). Rather than depending on advanced foundries (ultra-modern chip fabrication plants like TSMC), Q.ANT proves that older semiconductor fabs can be upgraded to produce next-generation AI chips - a model that could be replicated in industrial sites around the world.

Cherry Backing Europe’s AI Future

For Cherry, Q.ANT posits a fundamental reimagining of how AI infrastructure can work.

The market opportunity is vast. Data centres upgrade every 2-4 years, creating more replacement windows than obvious from outside the industry and Q.ANT's form factor means adoption can happen during normal refresh cycles without massive infrastructure overhauls.

More importantly, this is about European sovereignty in the most critical infrastructure of the AI age.

Q.ANT's manufacturing approach democratises chip production by reducing dependency on advanced foundries (e.g. TSMC or Samsung) and geopolitically sensitive supply chains. As export controls tighten around advanced semiconductors, this independence becomes increasingly valuable.

The team brings exactly the depth needed for this challenge. Michael's physics foundation from Max Planck (Germany’s leading fundamental research organisation), Tim Stiegler's industrial management from TRUMPF, and Andreas Abt's nine years at AMD create the rare combination of fundamental research, industrial execution, and systems expertise.

Q.ANT is carving out the high-leverage niches where traditional silicon hits physical constraints - energy-constrained data centers, low-power AI inference (when models are run, not trained), physics simulations and specialised training workloads where photonic architectures have genuine advantages.

This is European deep tech attacking a trillion-dollar global market with differentiated technology and the industrial backing to scale.