Vendors & EquipmentSparse Native AIAI SiliconAI InfrastructureData Centers

Sparxity banks $90m Series A to push sparse-native AI silicon

German sparse-native compute firm Sparxity, formerly SpiNNcloud, has raised $90m in a Series A led by ICM HPQC Fund, with Lockheed Martin and GlobalFoundries backing chips promising up to 44x faster AI inference.

4 min read

Sparxity raises $90m following Series A funding roundVendors & Equipment
Sparxity raises $90m following Series A funding roundSlimeyFox / pixabay

Why it matters

  • Sparxity raised $90 million in a Series A led by ICM HPQC Fund, with Lockheed Martin Ventures and GlobalFoundries participating.
  • The company targets up to 44x faster AI inference in ultra-sparse models using custom chips for fine-grained dynamic sparsity.
  • Sparxity reports three tapeouts, more than 70,000 chips produced and 12 racks deployed, with revenue quadrupling between 2024 and 2025.
  • In October 2026 it deployed a sparse-native rack at Jülich Supercomputing Centre; last week it partnered with Deutsche Telekom to bring the technology into T-Systems' data centers.

The story

German compute specialist Sparxity has closed a $90 million Series A round, money it will spend on custom silicon for what it calls fine-grained dynamic sparsity in AI workloads.

The round was led by the ICM HPQC Fund. Lockheed Martin Ventures, GlobalFoundries, GS Futures, Cloudberry Ventures, TGFS and the European Innovation Council (EIC) Fund all participated. The mix of a defense contractor, a foundry and an EU-backed vehicle signals that Sparxity's pitch — cheaper, less power-hungry AI infrastructure — has landed with strategic buyers as well as financial investors.

Sparxity is the new name of SpiNNcloud, a Dresden-rooted company founded in 2021. It builds full-stack systems that, in its own words, "compute only what matters": hardware and software designed to skip the parts of an AI model that contribute little, cutting both the arithmetic performed and the data moved between chips. Fewer data movements mean fewer bottlenecks and lower energy draw, the company argues.

What does the money buy?

The headline claim is throughput. Sparxity says its custom chips target up to 44 times faster AI inference in ultra-sparse models, against the dense architectures that dominate today's GPU fleets. That figure is a vendor target for fine-grained sparsity rather than a measured benchmark in independent hands, and the company's own commercial record, while real, remains at rack scale.

The operational numbers the company discloses:

  • Revenue quadrupled between 2024 and 2025, according to Sparxity's own account.
  • Three tapeouts completed.
  • More than 70,000 chips produced.
  • Twelve racks deployed.
  • More than 30 customers globally, including some of the largest supercomputing centers.

Two deployments anchor the customer story. In October 2026, Sparxity installed a sparse-native computing rack system at the Jülich Supercomputing Centre (JSC) in Germany. Last week it announced a partnership with Deutsche Telekom to bring the same sparse-native technology into the data center infrastructure of T-Systems, the operator's enterprise IT arm.

The Deutsche Telekom deal is the one that matters for the telecom sector. Operators across Europe face rising energy bills and capacity constraints as AI inference demand grows in their data centers, and a technology that promises to cut both compute and data movement sits squarely against that problem. Whether 44x holds up at T-Systems scale, rather than in a single rack at Jülich, is the question the next twelve months will answer.

The efficiency argument

Sparxity frames its pitch as a shift in how AI infrastructure gets built. "The next era of AI will be defined not just by how much computing power we can deliver, but by how intelligently and efficiently we use it," co-founder and CEO Hector Gonzalez said in a statement.

"This financing is a powerful validation of our vision for sparse-native computing and the complementary positioning of our systems. With the backing of leading, strategic investors, we're accelerating the transition from dense, energy-intensive architectures to heterogeneous data centers with dedicated silicon for ultra-sparse workloads to unlock greater AI performance and lower costs."

Note the framing: complementary, not replacement. Sparxity is not proposing to displace GPUs wholesale but to sit alongside them as dedicated silicon for workloads that tolerate extreme sparsity. Heterogeneous data centers, in this reading, mix dense and sparse hardware rather than betting on one architecture for every task.

That positioning is realistic. Sparsity techniques trade precision of computation for speed and energy, and they suit some inference workloads far better than others. Buyers will deploy them where the math fits, not everywhere.

Why now

The funding lands as AI infrastructure costs and energy consumption have become board-level concerns for operators, cloud providers and supercomputing centers alike. European institutions in particular face pressure to deliver AI compute without matching US-scale power budgets, which explains both the EIC Fund's presence in the round and the early traction at Jülich.

GlobalFoundries' participation adds a supply-side dimension: a major foundry with a stake in the company has an obvious interest in fabricating its silicon at volume. Sparxity has already moved past the prototype stage with three tapeouts and 70,000-plus chips out the door, so the Series A funds scaling, not first silicon.

What does this change for operators and data center operators watching from the sidelines? A third architecture option is now funded well enough to reach production scale, and Deutsche Telekom's T-Systems will be the reference case. If the sparse-native racks cut inference cost and power measurably there, expect wider operator interest in 2027; if not, Sparxity stays a supercomputing niche.

Also reported

Share this article:

« Previous

More from James Calloway

James Calloway

Show full bio

Staff writer covering companies, deals and strategy in Carrier business, 5G and 6G, fiber and broadband at Telecom Gazette.

498 articles