Green Compute is preparing another $4 million UK data centre deployment as it expands Bittensor Subnet 110, its marketplace for renewable-powered AI compute. According to the project’s official platform, SN110 has operated on Bittensor mainnet since April and Green Compute says enterprise inference traffic is already flowing through participating miners. The project separately brought a roughly $3 million GPU facility online in August, making the newly announced site an additional expansion rather than the initial production deployment.
Founder Josh Riddett disclosed the next phase during Exploit Summit 2026 in Montreal, where Green Compute said it plans to deploy another $4 million data centre by the end of October. The summit’s own account subsequently highlighted the target, while reporting from the event identified the previously deployed facility as a farm-based site. The $4 million figure remains planned capital deployment rather than currently operating capacity, making the distinction between live infrastructure and announced expansion central to evaluating the update.
SN110 Connects Renewable GPUs With Inference Demand
Green Compute describes SN110 as an inference marketplace where miners provide RTX 4090 and RTX 5090 capacity and validators assess performance, reliability and compliance with its renewable-energy requirements. Its current provider documentation says eligible power sources include biogas, solar, hydro, wind and geothermal, with certification records and hardware-location proofs used in the verification process. The subnet is designed to reward compute that meets both execution and energy-source requirements rather than treating available GPU capacity as interchangeable.
The model is intended to connect that supply with paying customers through an OpenAI-compatible inference API and direct GPU rentals. Enterprise users can pay in conventional currency, while Green Compute says part of inference revenue is converted into TAO and used to purchase and lock SN110 alpha tokens. The project reported more than $82,000 in revenue during its first month of Bittensor operations and disclosed an $8,200 buyback using 10% of contract revenue. Those figures are company-reported operating metrics and do not independently establish the scale or persistence of external customer demand.
That distinction is increasingly relevant across Bittensor. Other subnets are similarly moving from incentive-driven model competition toward services that outside applications can consume, including Bittensor deployments extending into inference and physical AI and Subnet 2’s cryptographically verifiable inference architecture. Running miners and validators demonstrates operational infrastructure, but commercial adoption requires separate evidence such as paid workloads, utilization and repeat customers.
$3M Cluster Is Live, $4M Site Is Next
Green Compute’s August deployment provides a more concrete infrastructure benchmark. The project announced that 240 RTX 5090 GPUs had gone live as permanent capacity intended for longer-term commercial workloads, after earlier promotional material had referenced a 250-GPU target. The completed cluster is therefore the approximately $3 million deployment, while the additional $4 million facility remains on the roadmap.
The strategy mirrors a broader effort to turn distributed GPU networks into usable cloud infrastructure. Render recently launched Dispersed_ai for on-demand decentralized inference, while providers such as io.net are focusing on GPU orchestration and workload routing. Green Compute differentiates its model by tying compute eligibility directly to renewable-power verification and by locating part of its permanent capacity at UK agricultural energy sites.
The remaining question is utilization rather than hardware availability. Green Compute now has a production subnet, an operating 5090 cluster and company-reported enterprise activity, while another large facility is targeted for deployment within weeks. Evidence that SN110 is becoming durable commercial infrastructure will ultimately come from recurring paid inference, GPU utilization, customer retention and verifiable revenue rather than installed GPU value alone. The $4 million expansion would materially increase available capacity if completed, but capacity and demand remain separate metrics.