Sparxity, the Dresden chipmaker formerly SpiNNcloud, closes Series A led by ICM HPQC Fund at $90M raised

Sparxity, a Dresden company that designs its own AI processors and the rack systems around them, has closed a Series A led by ICM HPQC Fund, lifting its total funding to $90M. The company did not say how much of that sum the new round accounts for. The investor list stands out for a European chip start-up: it mixes a defence group's venture arm, a chip foundry and the EU's own fund.

The EIC Fund, Lockheed Martin Ventures, GlobalFoundries, Cloudberry Ventures, GS Futures and TGFS all joined the round. Two angels also put in money: Richard Socher, who founded Recursive Superintelligence, and Nancy Xu, whose recruiting start-up Moonhub was acquired by Salesforce.

A bet on chips that skip work

The pitch rests on a simple observation. Large AI models often use only a fraction of their network to answer any given request, yet the GPUs that run them are built for dense maths and process everything anyway. Sparxity wants hardware that only fires where the model actually needs it, an approach the company calls dynamic sparsity.

The new money goes into developing that next generation of chips, aimed squarely at inference, the stage where a trained model serves users. Sparxity says the target is inference up to 44 times faster on very sparse models. That figure is a design goal, not a measured result on shipping silicon.

Co-founder and chief executive Hector Gonzalez said the funding will help data centres shift away from power-hungry dense designs towards set-ups built around sparse-native chips. In an earlier statement tied to a partnership deal, he put the idea in one line:

“Our computing technology is designed to concentrate computing power only where it is needed. This makes AI more energy-efficient and scalable.”

For ICM HPQC Fund, which backs high-performance computing, quantum and data-centre companies, the case was about execution rather than theory. “Sparxity has demonstrated the ability to manufacture its technology and deploy systems with major customers,” said Matthew Gould of the fund.

From brain-inspired research to a new name

The company began life in 2021 as SpiNNcloud, spun out of TU Dresden to commercialise SpiNNaker2, a many-core chip modelled loosely on how neurons in the brain communicate. Christian Mayr, the TU Dresden professor who led the SpiNNaker2 research, is a co-founder alongside Gonzalez.

On 28 October it renamed itself Sparxity. The change signals a narrower focus: the brain-inspired roots stay, but the commercial story is now about sparse computation for AI rather than neuromorphic research.

By its own count, the business is past the prototype stage. Sparxity says it has completed three tape-outs, produced more than 70,000 chips and installed 12 racks of systems. It lists more than 30 customers, among them Sandia National Labs in the US and the Jülich Supercomputing Centre in Germany, and says revenue grew fourfold between 2024 and 2025. It has not published absolute revenue figures.

Its current machines serve scientific computing as well as AI. On selected workloads the company claims sparse solvers run ten times faster, event-based machine learning uses 18 times less energy, and optimisation problems reach a usable answer 20 times sooner.

Jülich, T-Systems and the 2028 milestone

The round caps a busy few weeks. Alongside the rebrand, Sparxity announced a multi-million-dollar installation at Jülich, the German centre that houses JUPITER, Europe's first exascale supercomputer. The system there, called JUBIS, will be used for sparse machine learning, large HPC simulations and emulating quantum computers. European regional development money and the state of North Rhine-Westphalia helped pay for it.

Then on 3 November, at the German government's Digital Summit, Sparxity and T-Systems, Deutsche Telekom's corporate IT unit, put their names to a letter of intent. Digital minister Karsten Wildberger attended the signing. The two sides plan to use Sparxity's current hardware to work out how its next-generation chips can be integrated into Telekom data centres, with a first deployment targeted for 2028.

T-Systems chief executive Ferri Abolhassan said the company was looking forward to having Sparxity in its Munich AI factory soon. The letter is not a contract, though. The partners say the detailed terms still have to be negotiated over the coming months.

For European investors, the deal shows two things. Sovereign AI infrastructure is pulling local chip start-ups into the plans of incumbent telecoms groups, and US strategic money from Lockheed Martin and GlobalFoundries is willing to back a German chip designer at Series A. The EU has been an early supporter too: in 2025 Sparxity was selected for the EIC Accelerator, which brought €10m in blended grant and equity funding.

The harder test sits a couple of years out. Sparxity now has to turn its next-generation chip from a target into silicon that hits its own inference claims, and to convert the T-Systems letter into a signed deployment before the 2028 date arrives.