Antimatter launched today out of a three-way combination of US power-infrastructure company Datafactory, modular-data-centre operator Policloud and distributed-cloud provider Hivenet, and is raising €300 million to deploy the first 100 units of its distributed AI-inference network in 2027. The plan scales to 1,000 container-sized data centres by the end of 2030, representing more than 400,000 GPUs and upward of 36 exaFLOPS of inference capacity.
The thesis is that the era of training massive models inside centralised hyperscale campuses is giving way to an inference era with different infrastructure demands. Running models billions of times a day for copilots, agents and real-time decision systems rewards latency, geographic reach and energy efficiency. The company's pitch to investors is that it can deploy those sites in roughly five months against the 24-plus months a typical hyperscale build requires, and do so at a capital cost of around $7 million per fully-loaded MW compared with figures it puts closer to $35 million at traditional hyperscalers.
The power-first logic is central. Antimatter says it has secured more than 1GW of capacity through grid-connection agreements and site reservations, with more than 160MW already operational across Texas and Oregon, and positions its Policloud units directly at or near existing wind, solar, hydro and biogas sites rather than waiting for new transmission build-out. In Europe alone, the company points out, more than 12 TWh of renewable electricity was curtailed in 2023, worth an estimated €4.2 billion in lost value, while more than 1,000GW of renewable capacity is stuck in permitting or connection queues across Europe and the Gulf.
In the age of AI, intelligence is not the bottleneck — energy is. The infrastructure built for the first era of cloud and AI was designed around centralized scale. But the inference era requires a different model: more distributed, faster to deploy, and sovereign by design. That is the infrastructure Antimatter is building.
Gurlé previously founded Microsoft's Real-Time Communications business (the lineage that became Microsoft Teams), ran Skype's enterprise division through its sale to Microsoft and founded Symphony Communication Services.
The operational baseline today is 10 Policloud units across 8 sites, 2,600 deployed GPUs and a pipeline the company describes as demand for more than 10,000 GPUs and over 500 additional units. Each Policloud houses up to 400 GPUs, runs with Tier 3 availability by default, and is orchestrated through a proprietary distributed-computing and storage fabric the company says is already serving billions of inference requests a day with sub-10 millisecond edge latency for regulated workloads.
AI infrastructure is now a strategic asset class, and the winners will be those who can combine hard assets with software at scale. Antimatter's vertically integrated model — from megawatts to APIs — is exactly the kind of infrastructure we believe can define the next decade of digital growth.
Further investor support has come from Stéphanie Hospital, Founder and CEO of OneRagtime, who said Europe needs sovereign, energy-efficient infrastructure to compete in AI; Noor Sweid, Founder and Managing Partner at Global Ventures, who framed the opportunity in emerging-markets terms; and Bruno Sportisse, Chairman and CEO of Inria, who cast the launch as a strong illustration of the deeptech industrial story he wants to see emerge in Europe.
Antimatter is targeting $250 million-plus in revenue inside 18 months and more than $3 billion by 2030, with a customer base it describes as 35 per cent energy, 30 per cent public sector, 20 per cent corporates and 15 per cent agriculture. The global data centre capacity market is projected to grow from 55GW in 2023 to 220GW by 2030, a compound growth rate of 22 per cent, and grid-connection queues are the defining supply-side constraint on how much of that can be met in the timeframes customers want. A distributed model that brings the compute to the power, rather than the power to the compute, is a direct response to that constraint, and Antimatter is now one of several bets testing whether the inference era really does need a different shape of data centre.
