Most powerful supercomputer in Europe revealed: ‘As fast as a million smartphones’ |
Germany today unveiled Jupiter, now the most powerful supercomputer in Europe. The system can execute many hundreds of quadrillions of calculations per second (10^15), a level of performance the hosting research center likens to the combined power of nearly a million modern smartphones.
Jupiter currently ranks third worldwide. In the United States, several machines surpass a key international performance benchmark; Jupiter is expected to reach that threshold as its configuration is fully dialed in. Built from numerous tightly networked compute nodes that work in parallel, Jupiter accelerates massive workloads that would otherwise take months, shrinking them to far more manageable turnarounds.
Real-world shakedown: Textgain puts Jupiter to work
Belgian AI company Textgain has already been test-driving Jupiter, director Guy De Pauw told NOW. As part of the commissioning phase, the team was allowed to run up to 300,000 hours of computing time to validate performance and reliability. Thanks to Jupiter’s highly parallel design, the actual wall-clock time is far less than the raw hours imply. “We’ve used about half so far,” De Pauw said.
Textgain builds software that detects profanity, threats, and other hate speech across all official European languages. Training systems to recognize such nuance isn’t automatic: models must be taught through examples and clear instructions. The company develops its base models on LUMI, the Finnish supercomputer, then uses its Jupiter allocation to fine-tune specialized variants—targeting narrow tasks like reliably spotting explicit death threats.
Why supercomputers matter for AI and science
Supercomputers can’t be compared one-to-one by a single number: performance depends on the workload. A global climate model, a computational fluid dynamics simulation, and a large-scale AI training run stress very different parts of a system. That’s why the industry uses standardized benchmarks to estimate broad capability.
The Top500 list, maintained by an international organization, offers a widely watched snapshot of peak performance across the world’s largest machines. Its headline metric is measured in petaflops—quadrillions of floating-point operations per second—providing a common yardstick for ranking systems even if real-world speeds vary by task.
“Many researchers in the Netherlands benefit from more computing power,” says Jannis Teunissen, a researcher at the Centrum Wiskunde & Informatica (CWI) in Amsterdam. With more horsepower, “you can run larger simulations or add finer detail, for example to make local predictions.” That doesn’t usually unlock brand-new problem classes overnight, he notes, but moving a step bigger or more precise can turn a rough estimate into a result you can trust. His example: simulate airflow around a Formula 1 car. Higher resolution makes the model more accurate and reliable—and that can translate directly into a faster design.
Speed—and expertise—on tap
Textgain says the benefits extend beyond raw compute. Engineers at Jupiter’s research center know the machine inside and out and can help teams profile code, tune settings, and optimize data movement—critical tweaks that shorten training time and reduce the total compute hours required. The payoff is practical: getting an AI model “ready” sooner for deployment, even though such systems must be continually retrained and refreshed to stay effective.
Why not use even larger U.S. systems? For De Pauw, it’s also a matter of principle. He argues that Europe should build and rely on its own AI infrastructure—both to prove it can and to avoid strategic dependence on American facilities.
The Dutch angle: Snellius
The Netherlands’ national supercomputer, Snellius, is dedicated to non-commercial research. In the most recent Top500 assessment, it delivered nearly 25 petaflops of performance, with a theoretical peak close to 38 petaflops. That capacity underpins a wide range of academic and public projects, from physics and chemistry to climate science and data-intensive humanities research.
A milestone for European computing
Jupiter’s debut marks a significant milestone for Europe’s scientific and AI ecosystem. With hundreds of quadrillions of operations per second on tap—and an upgrade path toward the next international performance tier—the system will help researchers run bigger, more detailed simulations and train more capable language and vision models. For companies like Textgain, it offers a place to build and test advanced AI responsibly on European soil, backed by local expertise and world-class hardware.
