Euclyd Unveils CRAFTWERK: Exascale Inference Architecture Targets Agentic AI at Record Efficiency
Cette publication existe aussi en Français
European AI chip startup Euclyd has announced CRAFTWERK, a new inference architecture designed to dramatically cut the cost and power consumption of large-scale AI deployments. Positioned as the world’s most power-efficient exascale token factory, the system made its formal debut at the KISACO Infrastructure Summit 2025 in Santa Clara.
Radical Approach to Inference
CRAFTWERK. Source: euclyd.ai
Unlike traditional accelerators, CRAFTWERK is built from the ground up for agentic AI workloads. At its core lies the CRAFTWERK SiP (system-in-package) – a palm-sized module integrating:
-
16,384 custom SIMD processors
-
Up to 8 PFLOPS (FP16) or 32 PFLOPS (FP4) compute
-
1 TB of custom ultra-bandwidth memory (UBM) delivering 8,000 TB/s bandwidth
The SiP combines custom processors, custom memory, and advanced 2.5D/3D packaging to push inference efficiency far beyond conventional designs.
Scaling to Exascale
Multiple SiPs power the company’s flagship rack-scale system, the CRAFTWERK STATION CWS 32. Configured with 32 SiPs, the system delivers:
-
1.024 exaflops of FP4 compute
-
32 TB of UBM
-
7.68 million tokens per second in multi-user mode
-
Power consumption of just 125 kW
According to Euclyd, this represents a 100× improvement in tokens per joule and cost per token compared with today’s leading inference alternatives, based on modeled performance with Llama 4 Maverick.
Industry Backing
Euclyd CEO Bernardo Kastrup described the philosophy behind the design:
“Our Crafted Compute philosophy reimagines inference from the ground up—custom processors, custom memory, and advanced packaging. We’ve engineered every gate for maximum efficiency and minimal power draw—by far the lowest in the industry.”
Investor Peter Wennink, former CEO of ASML, added:
“I believe AI inference will dominate datacenter silicon. CRAFTWERK’s breakthrough economics will accelerate agentic AI adoption and usher in an era of abundant inference.”
Founded in Eindhoven, The Netherlands, Euclyd positions itself as a European challenger in AI infrastructure. The company focuses on environmentally conscious and socially responsible engineering, while pursuing radical efficiency gains in datacenter compute.
Its advisory and investor network includes industry figures such as Peter Wennink (former CEO of ASML), Federico Faggin (microprocessor inventor and founder of Zilog and Synaptics), and Steven Schuurman (founder of Elastic).
Euclyd maintains offices in Eindhoven and San Jose, California.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
