Q.ANT Says 100x. Its Independent Lab Says 50x. Neither Has Published the Benchmark.
Q.ANT, a German photonic computing startup spun off from industrial laser maker Trumpf in 2018, signed its first commercial customer this week: European cloud provider IONOS, which serves 6.8 million customers across 17 markets. The agreement, announced at the re:publica 2026 conference in Berlin, will bring Q.ANT's Native Processing Server into a commercial cloud environment for the first time. The company positions its second-generation photonic co-processor as a complement to conventional GPUs, not a replacement. GlobeNewswire
The news is real. The numbers are not.
Q.ANT's own website says its Gen2 processor delivers up to 100x the performance of the Gen1 chip. An independent evaluation conducted by Germany's Leibniz Supercomputing Centre, the same research facility Q.ANT has repeatedly cited as validation, documented a 50x improvement. That is the same company, the same hardware, the same benchmark — and a two-times gap between what Q.ANT says publicly and what its own chosen verifier found. Q.ANT | GlobeNewswire
Neither figure has a published methodology. Nobody outside Q.ANT knows what workload was run, what baseline was used, or how 'performance' was defined. Q.ANT's internal benchmarks also claim its Gen2 NPU delivers up to 30x the energy efficiency and 50x the per-application performance of conventional processors — numbers no external party has verified. GlobeNewswire
"AI infrastructure cannot scale on its current trajectory — power, cooling, and silicon economics will not keep up," said Dr. Michael Förtsch, Q.ANT's founder and CEO, in the company's announcement. That part is almost certainly true. Data center electricity demand is on pace to reach 3 percent of global consumption by 2030, with individual server racks in advanced facilities drawing as much power as 65 households. The question is whether Q.ANT's chip is actually the answer. GlobeNewswire
The technical foundation is not implausible. Q.ANT's processor runs AI workloads on Thin-Film Lithium Niobate, a material whose non-linear optical properties allow certain mathematical operations — including the non-linear activation functions central to neural networks — to be performed in the photonic domain without converting signals back to the electrical domain. Silicon photonic approaches like those from Lightmatter and Lightelligence require power-hungry digital-to-analog and analog-to-digital conversions that constrain both precision and efficiency. That is a genuine limitation. Whether Q.ANT has solved it at commercial scale is a different question. EE Times | Medium
Q.ANT's photonic chips are manufactured at a pilot line in Stuttgart operated jointly with IMS CHIPS. The company has installed Gen2 hardware at two German supercomputing centers — LRZ in Munich and JSC in Jülich — which are research deployments. The IONOS rollout is planned for later in 2026. That is the first commercial customer. No paying workloads are running yet. GlobeNewswire
The analyst quoted in Q.ANT's press materials — Bob Sorensen of Hyperion Research — covers the HPC and quantum sector. Hyperion's client list includes companies in that exact market. Q.ANT's own benchmarks are cited alongside that endorsement without disclosing the relationship. Q.ANT ISC 2025
Q.ANT was founded in 2018 and secured what it calls Europe's largest Series A for photonic computing in 2025. The company maintains a U.S. office in Austin, Texas. It has real technology, real installations, and now a real commercial agreement. The physics underlying its approach has independent validation from researchers studying Thin-Film Lithium Niobate for photonic computing. None of that means the performance numbers are trustworthy. Q.ANT | Medium
Performance claims in photonic and quantum computing are easy to make and hard to disprove. The gap between what a company claims and what an independent lab measures has closed, widened, or simply vanished in nearly every overhyped hardware story of the past decade. Q.ANT is not a fraud. It is a company with a real product, a real customer, and a technology that might actually work at scale. It is also a company whose most important benchmark numbers cannot be verified, and whose own cited evidence points in two different directions at once.