Sparxity, a company that develops sparse-native computing technology, has announced the close of its Series A round. The latest deal takes the total capital secured by its team to $90 million. The company plans to use this capital to develop advanced chips that can make artificial intelligence (AI) faster, more affordable and more energy-efficient.
The Series A round was led by ICM HPQC Fund. Several major companies and investment groups also took part in the deal. These include Lockheed Martin Ventures, GlobalFoundries, GS Futures, Cloudberry Ventures, TGFS and the EIC Fund.
The round also received support from well-known angel investors. These include Richard Socher, a TIME100 AI leader and part of Recursive Superintelligence, and Nancy Xu, another TIME100 AI leader whose company, Moonhub, was acquired by Salesforce.
Atlantic and Silicon Catalyst also continue to support the company. This mix of investors brings experience from the fields of technology, defence, semiconductors, AI and business.
The new capital marks an important step for Sparxity as it seeks to improve how computers handle complex AI tasks. Its main goal is simple: make computer systems perform only the work that a task needs instead of wasteful extra operations.
What Makes Sparxity Different?
Most traditional computer chips follow a dense computing model. This means they process large amounts of data and perform many operations, even when some of those operations do not add much value to the final result.
Sparxity takes a different path through sparse-native computing. This method allows a computer to focus on the parts of a task that matter most. It can skip unnecessary calculations and reduce the amount of data that must move across the system.
This approach is especially useful for modern AI models. Many advanced AI systems do not need every part of a model to work at the same time. Only a small group of experts, connections or other model parts may be necessary for a particular task.
Sparxity wants to build chips that can handle this type of work at a much deeper level. Its next-generation technology will focus on fine-grained dynamic sparsity. This means that the system can select which parts of a model need to work at a very detailed level, based on the task at hand.
The company says this approach could deliver up to 44 times faster AI inference for ultra-sparse models. AI inference refers to the process through which a trained AI model responds to a request or produces an answer.
Such a change could have a major effect on AI data centres. These facilities require large amounts of electricity and expensive hardware. If chips can complete more useful work with fewer operations, companies may be able to reduce costs while also improve performance.
PEER Research Shows the Potential of Dynamic Sparsity
Sparxity’s work has a strong link to research on dynamic sparsity. One example is PEER, a research project that showed how AI models can use millions of fine-grained experts while activate only a small share of their total compute capacity.
Xu Owen He, a former Google DeepMind researcher and strategic advisor to Sparxity, spoke about this work.
He explained that the PEER project showed the value of dynamic sparsity. It proved that AI models could handle large expert systems without the need to use all available compute power for every task.
However, current hardware places limits on how far this method can go. Most existing chips were not built for such detailed and flexible patterns of work.
According to He, the next step is to develop AI models, distillation pipelines and hardware as parts of one complete system. Model distillation is a process that helps a smaller model learn from a larger one. This can help reduce model size and compute needs while retain useful performance.
Sparxity believes that a close link between software and hardware can unlock much better results. Instead of force sparse AI models to work on chips built for dense tasks, the company wants to create hardware that suits sparse models from the start.
Strong Business Growth and Customer Base
The new Series A deal comes after a period of strong business growth for Sparxity. The company quadrupled its revenue between 2024 and 2025. It has also reported continued momentum in 2026.
Its progress is not limited to research or product design. Sparxity has already built a track record in chip production and real-world deployments.
The company has completed three tapeouts. A tapeout is a major step in chip development, where the final chip design is sent for manufacture. Sparxity has also produced more than 70,000 chips and deployed 12 racks.
These figures show that the company has moved beyond the early research stage. It has built products, produced hardware and placed its systems at customer sites.
Sparxity now has more than 30 customers across the world. Its customer base includes some of the largest supercomputing facilities.
This business progress gives the company a useful base as it moves toward its next generation of AI chips. It also helps show that there is real demand for systems that can reduce the cost and power needs of large computing tasks.
Better Speed and Lower Energy Use
Sparse computing aims to reduce waste in computer systems. It does this by cutting unnecessary calculations and reducing data movement. This can lead to faster results and lower power use.
Sparxity already offers systems for high-performance computing (HPC) and AI. The company has shared several results from selected workloads to show the value of its technology.
Its systems have delivered 10 times faster sparse solvers. Sparse solvers are tools that help computers solve mathematical problems that contain many zero values or unused data points.
The company has also reported 18 times lower energy use for event-based machine learning. This type of machine learning works with changes or events rather than process all data in the same way.
For selected optimization problems, Sparxity has reported runtimes that are 20 times shorter. Optimization tasks help computers find better solutions under a set of rules or limits.
These results show how sparse-native systems may offer benefits across different areas of computing. However, the results apply to selected workloads and should not be treated as a promise of the same speed or energy gains for every task.
The company’s next goal is to bring these benefits to AI inference through a new chip architecture built for dynamic sparsity.
For AI model developers, this could offer more detailed and flexible sparse architectures. For data centre operators, the main benefits could be higher throughput, lower power use and reduced infrastructure costs.
Major Deployment at Jülich Supercomputing Centre
Before its Series A announcement, Sparxity shared news of a major deal with the Jülich Supercomputing Centre (JSC) in Germany.
The company confirmed a multi-million-dollar deployment of its sparse-native computing systems at JSC. The centre is part of Forschungszentrum Jülich, one of Europe’s major research institutions.
JSC is also home to Europe’s first exascale supercomputer. An exascale system can perform at least one quintillion calculations per second. Such systems support complex scientific work across several fields.
The deal marks an important step for Sparxity because it places its technology in a major supercomputing environment. It also gives the company a chance to show how sparse-native systems can support demanding workloads at a large scale.
Supercomputing centres need high levels of performance, but they also face major power and cost limits. Any technology that can deliver more useful compute with less energy may offer clear value to these facilities.
The Jülich deal also adds to the company’s record of real-world deployments and gives further weight to its claims about the value of sparse-native hardware.
Deutsche Telekom Partnership Supports European AI Infrastructure
Sparxity has also announced a major partnership with Deutsche Telekom. The deal will bring its sparse-native technology into the data centre infrastructure of T-Systems, the group’s IT services business.
This partnership marks another important step for the company. It shows that its technology has a role beyond research centres and specialist computing sites.
The goal is to bring more energy-efficient AI compute to industrial use. Large businesses and public institutions need access to AI systems, but the cost of hardware and electricity can make such services expensive.
Sparxity’s technology could help address these issues through more efficient use of available compute power.
Ferri Abolhassan, CEO of T-Systems, said the company looks forward to having Sparxity as part of its AI factory. He also praised the German company’s energy-efficient use of AI compute power.
The partnership is important for Europe’s wider AI plans. Countries and businesses across the region want strong local AI infrastructure that can support their needs and reduce their reliance on outside providers.
With T-Systems as a partner, Sparxity can take its technology closer to large commercial data centre operations.
Investors Back Sparxity’s Long-Term Vision
ICM HPQC Fund led the latest Series A round. Matthew Gould from the fund said Sparxity has a unique position within the computing infrastructure market.
He noted that the company’s architecture addresses a major shift toward sparse activation. He also pointed to the limits of conventional hardware, which was not designed for this new type of AI workload.
Gould also praised Sparxity for its ability to manufacture its technology and deploy systems with major customers.
His comments reflect an important part of the company’s appeal. Sparxity is not just a chip design company with a new idea. It has already produced hardware and built customer relationships.
The support of Lockheed Martin Ventures, GlobalFoundries and other investors also gives the company access to a wide network of industry knowledge and business experience.
The involvement of Richard Socher and Nancy Xu adds further support from experienced AI leaders.
For Sparxity, the new capital offers the chance to build on its current products, improve its chip technology and reach more customers.
From SpiNNcloud to Sparxity
The company has also adopted a new name. It was formerly known as SpiNNcloud and has now introduced Sparxity as its new identity.
The change reflects its wider focus on sparse computing. Rather than limit its identity to a specific technology or product area, the new name points to its broader goal of making computer systems more efficient.
Its core idea remains the same: compute only what matters.
This principle is at the centre of its current HPC products as well as its next-generation AI chips. It also connects the company’s work across hardware, software and AI model design.
What Comes Next for Sparxity?
The AI industry faces a major challenge. As models grow larger and more complex, the need for powerful chips, electricity and data centre space also rises.
Traditional hardware has helped drive AI growth, but its limits are becoming more clear. Companies now need better ways to use their available compute resources.
Sparxity believes sparse-native computing can help address this problem. Its new chips will focus on fine-grained dynamic sparsity and could offer up to 44 times faster inference for ultra-sparse AI models.
The company’s current record offers a useful base for this next phase. It has completed three tapeouts, produced more than 70,000 chips, deployed 12 racks and built a customer base of more than 30 organisations worldwide.
Its work with Jülich Supercomputing Centre and Deutsche Telekom also shows that major institutions have an interest in its technology.
The $90 million in total capital secured by the team gives Sparxity more room to develop its products and expand its reach.
The next test will be how well its new chips perform under real-world AI workloads and how quickly customers adopt them.
If the company can turn its technical goals into reliable products at commercial scale, it could play an important role in the next phase of AI infrastructure. Its vision is clear: faster AI, lower energy use and less waste across the computing stack.
Also Read – InMobi Wins 30-Year Naming Rights to Bengaluru Metro Station