Rebellions (Spain-based AI chip startup) and ai& (compute operator) announced a partnership to deploy energy-efficient AI inference infrastructure in Japan.

Rebellions, a global leader in AI inference infrastructure, and ai&, a vertically integrated global AI technology company, are partnering to deploy Rebellions RebelRack within ai&'s heterogeneous infrastructure as part of the ai& inference service in Japan. The collaboration will give enterprises, government institutions, and developers access to Rebellions-powered inference deployed within the country, expanding locally available compute options while supporting Japan's sovereign AI priorities.

ai& will begin deploying Rebellions-powered infrastructure at its Tokyo data center with an initial purchase, planning to scale the deployment rapidly to 100 or more RebelRack units. This initial deployment can expand alongside ai&'s broader data center buildout, which is backed by more than $2 billion in committed infrastructure capital and includes five sites planned to be operational by year-end 2026 and 40 MW of capacity targeted by year-end 2027.

The economics of AI inference directly influence how widely AI can be deployed and its utilization rates. Rebellions' hardware is engineered to lower the unit cost of serving tokens, giving providers room to create new pricing tiers, support more applications, and serve customers with different compute requirements and budgets. ai& has built a heterogeneous infrastructure platform around this flexibility.

"The economics of inference directly influence how widely AI can be deployed and how much it can be used," says Sunghyun Park, Co-founder and CEO of Rebellions. "Lowering the unit cost of serving tokens gives providers room to create new pricing tiers, support more applications, and serve customers with different compute requirements and budgets. ai& has built a heterogeneous infrastructure platform around that flexibility. Establishing Rebellions inside Japan gives us a direct path to enterprise, government, and developer demand while adding efficient inference capacity to the country's AI infrastructure."

ai& operates a heterogeneous infrastructure model incorporating multiple compute architectures across AI workloads. Rebellions adds an inference-optimized architecture engineered for high power efficiency and lower operating costs, giving ai& greater flexibility in how it provisions inference capacity and structures its services. These economics can support new, lower-cost inference tiers designed to make AI services accessible across a wider range of applications, organizations, and budgets.

"As AI moves from experimentation to production, inference economics—particularly performance, power efficiency, and cost—will increasingly shape infrastructure decisions," said Matt Eastwood, SVP, WW Research at IDC. "ai&'s deployment of Rebellions at commercial scale is an important example of purpose-built AI silicon emerging as a viable infrastructure option for inference. Greater choice at the accelerator layer will be increasingly important as service providers, enterprises, and governments look to scale AI capacity efficiently."

Rebellions hardware integrates directly into ai&'s existing engineering workflows. ai&'s technical teams already work with open-source software packages common across GPU environments, and Rebellions supports widely adopted open-source frameworks. This compatibility reduces integration effort, with the first systems expected to be operational the same day they are delivered.

"Heterogeneous infrastructure means using the right hardware for the right workload," says David Bennett, Co-Founder and CEO of ai&. "Rebellions fits that model—its efficiency gives us more flexibility in how we structure the economics of our services, and it builds on the same open-source frameworks our engineers already use. This partnership expands the inference options we can offer customers in Japan, with a foundation to take it further."

Rebellions recently raised $400 million in a pre-IPO round, bringing its total funding to $850 million, and is now shipping its RebelRack and RebelPOD AI infrastructure systems. ai& is backed by more than $2 billion in committed infrastructure capital, with five sites planned to be operational by year-end 2026 and 40 MW of capacity targeted by year-end 2027.