- EUCYD CRAFTWERK SIP requirements on bandwidth levels far beyond Nvidia’s design
- It is said that the rack scale reaches over an Exaflop performance
- Claim efficiency requirements suggest massive gains, though independent tests are missing
European startup EUCYD has announced new hardware aimed at large-scale AI inferences.
The system, called Craftwerk, was introduced on Kisaco Infrastructure Summit 2025 in Santa Clara.
The company describes it as designed specifically for agent AI workloads with specifications that separate it from current accelerators.
Inside the Craftwerk architecture
The core of the release is Craftwerk SIP, a system-in-package that fits into the palm.
It integrates 16,384 custom SIMD processors along with 1 TB of custom ultra-band width memory or UBM.
EUCYD claims that this memory can deliver 8,000 terabytes per day. Second of bandwidth.
The Compute performance is listed for up to 8 Petaflops in FP16 and 32 Petaflops in FP4 precision.
These numbers place the module of what established companies like Nvidia is currently announcing.
“Our designed calculation philosophy is reimaginated inferens from scratch, custom processors, custom memory and advanced packaging,” said Bernardo Kastrup, CEO of EUCYD.
“We have designed each port for maximum efficiency and minimal power, by far the lowest in the industry.”
The company also revealed Craftwerk Station CWS 32, a rack-scale platform built of 32 Slurk.
In this configuration, EUCYD indicates that the system reaches 1,024 Exaflops of FP4 calculation, supported by 32 TB UBM.
It is said to generate 7.68 million tokens per day. Second in Multi-User mode and its power consumption is reported at 125 kilowatts.
According to the company, this is a hundred times gain in both energy consumption and cost -effectiveness compared to current alternatives.
Benchmark used to establish these improvements was modeled performance with Llama 4 Maverick.
The headquarters of Eindhoven, the Netherlands, the company also maintains offices in San Jose, California.
It promotes environmental technology and efficiency gains in data center infrastructure.
“I think AI -Inferens will dominate Data Center Silicon. Craftwerk’s economy will speed up agent AI -adoption and initiate an era with ample inference,” said investor Peter Wennink, former CEO of ASML.
While the specifications of Craftwerk are ambitious, the allegations remain untested outside the company’s own framework.
Startups in the semiconductor room are often facing challenges in the production of scale, building reliable software support and ensuring integration with existing data center infrastructure.
EUCYD’s message suggests a design that can surpass leading accelerators on paper, but whether in practice it can deliver depends on results seen in real implementations.
Until these results emerge, hardware remains a set of impressive numbers with an uncertain path to widespread adoption.
