French cloud infrastructure company Antimatter has launched an ambitious plan to build a distributed network of AI-focused micro data centres, positioning itself as a new “neocloud” provider for the inference era. The company says the platform brings together power sourcing, modular data centre infrastructure, and distributed cloud software to address the rising demand for AI workloads that must run closer to users and available energy. The announcement comes as AI deployment increasingly shifts from model training in large centralised campuses toward high-volume inference for copilots, agents, and real-time applications.
Strategic Combination Creates a Full-Stack Platform
Antimatter has been formed through the combination of three businesses: Datafactory, a US-based energy and power infrastructure company; Policloud, a modular micro data centre network; and Hivenet, a distributed cloud provider. By integrating these capabilities, the company aims to control the full value chain from megawatts to cloud APIs rather than relying on separate partners for power, hardware, and orchestration. This structure is intended to shorten deployment timelines, reduce cost, and give customers more control over where AI compute and data are processed.
Energy-First Infrastructure Model
The central idea behind Antimatter’s strategy is to move compute to energy rather than waiting for energy to reach traditional data centre locations. The company says it has secured more than 1GW of power capacity across the United States, Europe, and the Gulf Cooperation Council region, with operational capacity already in Texas and Oregon. Its Policloud units can be installed at or near existing wind, solar, hydro, and biogas assets, turning underused or constrained power into AI infrastructure in months instead of years.
Roadmap and Capacity Targets
Antimatter is seeking €300 million to support the deployment of its first 100 Policloud units by 2027, although published figures differ on whether that phase represents more than 30,000 or 40,000 GPUs. Each modular unit can house up to 400 GPUs and is designed to be deployed in roughly five months, compared with more than 24 months often associated with hyperscale data centre builds. By 2030, the company is targeting a global footprint of 1,000 Policlouds, with sources citing more than 300,000 to 400,000 GPUs and over 36 exaFLOPS of distributed AI inference capacity.
Market Relevance and Industry Context
The launch reflects a broader infrastructure challenge facing the AI industry: demand for compute is rising faster than grid connections and large campus construction can accommodate. Antimatter argues that inference workloads require lower latency, more geographic reach, and stronger data sovereignty than the first generation of centralised AI infrastructure. Its model also speaks to governments and regulated sectors that want AI capacity within local jurisdictions while reducing dependence on a small number of hyperscale regions.
Antimatter’s proposal is significant because it reframes AI infrastructure as an energy-led, distributed system rather than a purely data-centre-led expansion. The company still has to prove that its financing, deployment pace, commercial pipeline, and performance claims can scale across regions and regulatory environments. If it executes on its roadmap, Antimatter could become an important European challenger in AI infrastructure and a test case for how inference capacity is built in the next phase of cloud computing.