€33m · Series A · Semiconductors · Dresden, Germany
Sparxity, a Dresden company building sparse-native computing systems for AI and high-performance workloads, has raised €33m in Series A funding. ICM HPQC Fund led the round, with Lockheed Martin Ventures, GlobalFoundries, GS Futures, Cloudberry Ventures, TGFS and the EIC Fund participating. Richard Socher and Nancy Xu joined as angel investors, while Atlantic and Silicon Catalyst continued their support. The capital will accelerate specialised chips for fine-grained dynamic sparsity in AI inference.
Skipping idle computation is the product thesis
Sparxity builds full-stack systems around chips designed to avoid operations and data movement when parts of a model are inactive. That is particularly relevant to sparse and mixture-of-experts models, where only a fraction of the network may be needed for a given request. The company’s proposition is that hardware and software designed together can turn that selective activation into higher throughput and lower energy use.
The company reports three tapeouts, more than 70,000 chips produced, 12 racks deployed and over 30 customers. Those figures are company-reported, and its next-generation chip target of up to 44 times faster inference applies to ultra-sparse models rather than every AI workload. The commercial test is whether Sparxity can reproduce those gains across customer software while manufacturing and supporting complete systems at scale.
The round figure is separate from cumulative funding
Sparxity’s company announcement says its team has secured $90m in cumulative funding, but does not present that total as the size of the new round. Participating investor Atlantic identifies the Series A itself as €33m. Keeping those figures separate avoids treating grants and earlier financing as fresh Series A capital.
The strategic investor mix gives Sparxity relationships across chip manufacturing, defence, cloud infrastructure and European public capital. That can help with fabrication and customer access, but specialised AI hardware still faces long development cycles, capital-intensive production and the need for software tools that make a new architecture usable.
Explore the AI hardware market
AI hardware: Where the next opportunities are emerging — Synaptic, hosted on EUVC · 16 October 2026. The market map explains why inference efficiency, software compatibility, manufacturing and customer demand have to advance together for specialised chips to compete.
