Euclyd Unveils CRAFTWERK: The World’s Most Power-Efficient Exascale Token Factory for Agentic AI
SANTA CLARA, Calif., September 22, 2025 — Euclyd, a leader in ultra-efficient AI computing, today announced CRAFTWERK, a breakthrough inference architecture currently in advanced design, engineered to deliver the lowest power and cost per token in the industry. The architecture has been formally introduced at the KISACO Infrastructure Summit 2025 in Santa Clara, showcasing Euclyd’s radical approach to agentic AI infrastructure.
At its core is the CRAFTWERK SiP (system-in-package), a palm-sized powerhouse featuring 16,384 custom SIMD processors delivering up to 8 PFLOPS (FP16) or 32 PFLOPS (FP4). Paired with 1 TB of custom ultra-bandwidth memory (UBM) offering a staggering 8,000 TB/s bandwidth, the SiP sets a new benchmark for datacenter inference efficiency.
CRAFTWERK powers Euclyd’s flagship rack-scale system, CRAFTWERK STATION CWS 32, which integrates 32 SiPs to deliver 1.024 exaflops of FP4 compute and 32 TB of UBM. In multi-user mode, CWS 32 is projected to achieve 7.68 million tokens per second at just 125 kW, representing a 100x improvement in power efficiency and cost per token over leading alternatives—based on modeled performance for Llama 4 Maverick.
“Our Crafted Compute philosophy reimagines inference from the ground up—custom processors, custom memory, and advanced 2.5D/3D packaging,” said Bernardo Kastrup, Euclyd Founder and CEO. “We’ve engineered every gate for maximum efficiency and minimal power draw—by far the lowest in the industry.”
“I believe AI inference will dominate datacenter silicon,” added Peter Wennink, Euclyd investor and former ASML CEO. “CRAFTWERK’s breakthrough economics will accelerate agentic AI adoption and usher in an era of abundant inference.”
About Euclyd
Euclyd is an European technology startup pioneering ultra-efficient silicon systems for foundation AI models, including large language models. By rethinking every layer of the stack—from custom neural processors to memory architecture and system-level design—Euclyd dramatically reduces the energy, cost, and footprint of AI datacenter infrastructure.
Rooted in European engineering values, Euclyd builds environmentally conscious, socially responsible, and meticulously crafted AI solutions. The company is headquartered in Eindhoven, The Netherlands, with offices in San Jose, California.
Euclyd is led by a team of visionary engineers, mentored and backed by industry legends including Peter Wennink (former CEO of ASML), Federico Faggin (inventor of the microprocessor and founder of Zilog and Synaptics), and Steven Schuurman (founder of Elastic).
Learn more at www.euclyd.ai
Related Semiconductor IP
- Simulation VIP for Ethernet UEC
- Bluetooth® Low Energy 6.2 PHY IP with Channel Sounding
- Simulation VIP for UALink
- General use, integer-N 4GHz Hybrid Phase Locked Loop on TSMC 28HPC
- JPEG XL Encoder
Related News
- Arteris Accelerates AI-Driven Silicon Innovation with Expanded Multi-Die Solution
- Marvell Develops Industry’s First 2nm Custom SRAM for Next-Generation AI Infrastructure Silicon
- Astera Labs and Alchip Announce Strategic Partnership to Advance Silicon Ecosystem for AI Rack-Scale Connectivity
- Siemens turbocharges semiconductor and PCB design portfolio with generative and agentic AI
Latest News
- Mixel MIPI IP Integrated into Automotive Radar Processors Supporting Safety-critical Applications
- GlobalFoundries and Navitas Semiconductor Partner to Accelerate U.S. GaN Technology and Manufacturing for AI Datacenters and Critical Power Applications
- VLSI EXPERT selects Innatera Spiking Neural Processors to build industry-led neuromorphic talent pool
- SkyWater Technology and Silicon Quantum Computing Team to Advance Hybrid Quantum-Classical Computing
- Dnotitia Revolutionizes AI Storage at SC25: New VDPU Accelerator Delivers Up to 9x Performance Boost