Euclyd Unveils Revolutionary AI Memory Tech with Unmatched Speed and Efficiency
Euclyd, a Swedish tech company, introduces a new AI memory chip boasting unprecedented speed and efficiency, promising to revolutionize AI data centers.
- • Euclyd's UBM memory offers 8,000 terabytes per second bandwidth, surpassing Nvidia's HBM.
- • CRAFTWERK SiP achieves up to 8 petaflops in FP16 and 32 petaflops in FP4 precision.
- • CRAFTWERK STATION CWS 32 attains 1,024 exaflops FP4 with 32 TB UBM, consuming 125 kW power.
- • Claims of 100x energy and cost efficiency gains over current AI inference solutions.
- • Investor Peter Wennink supports Euclyd's potential to transform AI inference economics.
Key details
Swedish technology company Euclyd has announced a groundbreaking advancement in artificial intelligence hardware with its UBM memory technology. The company claims its new UBM memory delivers an extraordinary bandwidth of 8,000 terabytes per second, significantly outperforming established memory solutions such as Nvidia's HBM technology.
Euclyd has introduced the CRAFTWERK SiP, a system-in-package featuring 16,384 processors that achieves up to 8 petaflops in FP16 precision and an impressive 32 petaflops in FP4 precision. Taking the technology further, the CRAFTWERK STATION CWS 32—a rack-mounted platform composed of 32 SiPs—reaches an astonishing 1,024 exaflops in FP4 computations with a total of 32 TB UBM memory. Operating at an energy consumption of 125 kilowatts, it enables a multi-user mode output of 7.68 million tokens per second.
Crucially, Euclyd's technology promises to deliver a hundredfold improvement in energy efficiency and cost-effectiveness compared to current AI inference systems, based on performance modeling with the Llama 4 Maverick AI model. This could herald a new era of highly efficient and affordable AI inference in data centers.
Investor confidence is high, with Peter Wennink, former CEO of ASML, expressing optimism about the potential of Euclyd's approach to accelerate agentic AI deployment and reshape data center economics. Euclyd underscores its commitment to European engineering standards, prioritizing environmentally friendly technology and efficient infrastructure.
However, independent verification of these performance claims has yet to be completed. Industry observers caution that Euclyd will face significant challenges scaling production and establishing robust software ecosystems to support the hardware.
As it stands, Euclyd's innovations highlight a potentially transformative leap in AI hardware technology, with the next phase focused on real-world testing, verification, and adoption at scale.