Sparxity’s cumulative funding reaches $90 million amid chip push

By Olivia Hart

Sparxity has closed a Series A financing that brings its cumulative funding to $90 million, although the company did not disclose the round’s size. The Dresden-based semiconductor and AI infrastructure company, formerly known as SpiNNcloud, will use the capital to develop specialized inference chips that concentrate computing resources on operations AI models require.

What does the financing buy?

ICM HPQC Fund led the round. Lockheed Martin Ventures, GlobalFoundries, GS Futures, Cloudberry Ventures, TGFS and the European Innovation Council Fund also participated.

Individual investors included Richard Socher of Recursive Superintelligence and Nancy Xu, whose previous company Moonhub was acquired by Salesforce. Existing investors Atlantic and Silicon Catalyst also participated.

The announcement gave no check sizes, valuation or investor commitments. Sparxity says the money will accelerate next-generation inference chips and support additional customer deployments.

Hector Gonzalez, co-founder and CEO of Sparxity, said:

“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. 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 datacenters with dedicated silicon for ultra-sparse workloads to unlock greater AI performance and lower costs.”

Bottom line: the financing buys development time, not proof that Sparxity can convert technical gains into scaled revenue.

How does sparse computing work?

Sparxity builds full-stack computing systems around sparse processing. Its approach reduces unnecessary mathematical operations by activating only the portions of a model or system required for a particular task. Sparxity contrasts this with dense architectures, which can perform many operations even when capacity is unnecessary for a given workload.

The company combines custom semiconductor designs with systems capable of operating at rack and supercomputer scale. Sparxity believes specialized hardware can improve the efficiency of AI models designed around sparse activation. It is developing hardware and software for these architectures across a range of AI workloads.

Xu Owen He, strategic advisor to Sparxity and former Google DeepMind researcher, said:

“Our work with PEER demonstrated the potential of dynamic sparsity, activating only a fraction of the compute across millions of fine-grained experts. Yet today’s hardware limits how far we can take these approaches. Co-designing AI models, distillation pipelines and hardware around sparsity unlock a new level of computing efficiency.”

The announcement does not explain PEER or provide independent benchmark data. Xu’s statement therefore identifies a hardware constraint, not production-scale proof of performance gains.

The practical point is that Sparxity needs purpose-built silicon and software, not merely faster general-purpose processors.

How strong are the performance claims?

Sparxity says its next-generation chip technology targets fine-grained dynamic sparsity in AI inference. It projects performance improvements of up to 44 times faster inference for certain ultra-sparse models. The company says those projections depend on model characteristics and workload.

Sparxity reports three workload results:

• Tenfold acceleration for certain sparse solver workloads.
• 18 times lower energy consumption for selected event-based machine learning applications.
• 20 times shorter execution times for certain optimization problems.

Sparxity says these figures reflect specific workloads rather than universal performance improvements across all computing tasks. They do not establish how often customers will encounter the best case.

The financing follows several commercial and technical milestones, including three semiconductor tapeouts and more than 70,000 chips manufactured. Sparxity’s existing systems support high-performance computing and AI applications. The company has deployed 12 computing racks and serves more than 30 customers worldwide. Sparxity says its customers include major supercomputing facilities.

Sparxity reports that revenue quadrupled between 2024 and 2025, a 300% increase. The announcement discloses neither absolute revenue nor customer concentration, so the sales gain remains difficult to size.

The evidence is stronger than a laboratory claim, but the headline gains still apply only to selected workloads.

Where do the next deployments lead?

Sparxity announced a multimillion-dollar deployment of its computing systems at the Jülich Supercomputing Centre in Germany. According to Sparxity, the center is associated with Forschungszentrum Jülich and operates advanced high-performance computing infrastructure, including Europe’s first exascale supercomputer.

Sparxity also established a partnership with Deutsche Telekom to integrate sparse-native computing into the data center infrastructure of its T-Systems subsidiary. The collaboration is intended to support energy-efficient AI computing at industrial scale.

Ferri Abolhassan, CEO of T-Systems, said:

“We are looking forward to having Sparxity on board in our AI factory soon. The German company scores particularly well with its energy-efficient use of AI computing power.”

Matthew Gould, ICM HPQC Fund, said:

“Sparxity has a unique complementary positioning at the infrastructure stack. Its architecture is grounded in first principles, addressing the inevitable shift toward sparse activation and the structural limitations of conventional hardware.

Sparxity has demonstrated the ability to manufacture its technology and deploy systems with major customers. We are pleased to lead this financing as the team builds on that track record to address next-generation AI inference.”

Sparxity sees demand for specialized architectures increasing as AI infrastructure operators seek lower power consumption and better utilization. It argues training and inference systems frequently require substantial resources, creating pressure to improve performance per unit of electricity. Sparxity says sparse architectures offer one approach by reducing redundant operations and data movement.

The new funding will accelerate development of Sparxity’s next-generation inference chips while supporting additional customer deployments. The company aims to provide an alternative to relying exclusively on conventional dense processors for increasingly complex AI workloads.

The next test is whether Jülich and T-Systems turn workload-specific gains into repeat deployments and disclosed revenue, because the $90 million currently shows investor appetite rather than a proven market lead.