Pomegra
Dresden's Sparxity closed a Series A that lifts its cumulative funding to €80M, backing sparse-native inference chips with 30-plus customers.
Key Takeaways
- Sparxity's Series A, led by ICM HPQC Fund, brings total funding to €80M ($90M).
- The Series A amount and the valuation were not disclosed.
- Revenue quadrupled from 2024 to 2025; the company counts 30-plus customers.
Lead
Sparxity, a Dresden-based chip and systems company formerly known as SpiNNcloud, announced on November 10, 2026 that it closed a Series A bringing cumulative funding to €80M, about $90M. ICM HPQC Fund led the round. The company will put the money toward specialized chips for fine-grained dynamic sparsity, which it says can run ultra-sparse AI models up to 44 times faster in inference.
The headline figure needs careful reading. €80M is the running total raised since the company's 2021 founding, not the size of the new round. Sparxity has not said how much of it came in the Series A, and it has not disclosed a valuation.
Who Backed the Round?
ICM HPQC Fund led, and a mix of strategic and financial backers joined. Lockheed Martin Ventures, GlobalFoundries, GS Futures, Cloudberry Ventures, TGFS, the EIC Fund, Atlantic and Silicon Catalyst took part. Angel investors Richard Socher, founder of Recursive Superintelligence, and Nancy Xu, founder of Moonhub, which Salesforce acquired, also joined.
The syndicate leans toward strategic money. A foundry (GlobalFoundries) and a defense prime's venture arm (Lockheed Martin Ventures) have commercial reasons to back a chip company beyond financial returns. The public EIC Fund was already a backer: the European Innovation Council provided €10M in grants and equity in 2025.
Matthew Gould of ICM HPQC Fund said the company has shown it can manufacture its technology and deploy systems with major customers. That is a narrower claim than market leadership, and it matches the evidence on offer.
What Does Sparxity Actually Build?
Sparxity builds AI inference chips and full-stack systems that skip zero-value operations and reduce data movement. The architecture descends from SpiNNaker2, a brain-inspired design, and the company began as a TU Dresden spinoff. Dense accelerators from the large GPU vendors compute every value in a matrix, including the many that are zero. Sparse-native hardware aims to avoid that waste.
The company reports three tapeouts, more than 70,000 chips produced and 12 racks installed. It also cites 10 times faster sparse solvers, 18 times lower energy use for event-based machine learning and 20 times shorter runtimes for optimization tasks. These are company-reported benchmarks, and the 44 times figure is a target for ultra-sparse models, not a measured result on mainstream workloads.
How Much Traction Does the Company Have?
Sparxity reports more than 30 customers worldwide, including major supercomputing centers, and revenue that quadrupled between 2024 and 2025. Revenue in absolute terms is undisclosed, so the growth rate carries little weight on its own. A small base can quadruple quickly.
Named deployments are more informative. The Jülich Supercomputing Centre in Germany runs a multi-million-dollar system, and the company has a partnership agreement with Deutsche Telekom's T-Systems. CEO and co-founder Hector Gonzalez framed the pitch as intelligent, efficient use of compute rather than raw compute power. Xu Owen He, a former Google DeepMind researcher, serves as a strategic advisor.
Why Does the Round Matter for European AI Hardware?
The round matters because European startups rarely raise sums of this size for chip development, and the company is betting that efficiency, not scale, is where it can compete. Nvidia dominates dense AI training and inference, and a startup cannot beat it at its own game. Sparsity is a narrower claim: that some workloads, such as pruned models, event-based networks and optimization problems, leave most dense hardware idle.
The counterargument is real. Software stacks tuned for dense GPUs have years of tooling behind them, and models must be sparse enough for the hardware advantage to appear. If mainstream models stay dense, Sparxity's market stays specialized, centered on research computing and efficiency-sensitive deployments.
The rebrand from SpiNNcloud, completed the month before the announcement, signals a shift from academic lineage toward a product identity built around sparsity itself.
What Comes Next?
Sparxity has to prove that its tapeouts translate into repeat orders outside supercomputing centers. Watch for disclosure of revenue, the size of the Series A tranche and any commercial deployments with telecom or enterprise customers beyond the T-Systems agreement. Chip startups typically need additional capital before volume production, so a larger round could follow if the customer base converts.
Outlook
Sparxity enters its next phase with €80M in cumulative funding, a strategic investor group and a product already in racks. Missing figures, including the Series A amount, valuation and absolute revenue, leave the scale of the bet unclear. The next test is whether sparse-native hardware can move from research centers into production inference.
