Boundless repurposes 4,000-GPU network for AI inference

On Tuesday Boundless announced it expanded its roughly 4,000-GPU distributed network to support AI inference while continuing to operate its zero-knowledge proving service in parallel.

The company said it optimized the existing GPU fleet for inference through hardware tuning, workload adaptation and managed operations such as routing and scheduling. The network now acts as a coordination layer that matches AI jobs to available capacity and aims to keep utilization high across mixed hardware.

Boundless built the GPU network over four years to handle compute for zero-knowledge proving and cross-chain verification. The infrastructure previously coordinated work between the Ethereum mainnet and a Base Layer 2 connected to Bitcoin. Shiv Shankar, Boundless’s CEO, wrote in the announcement: “Four years ago, we set out to solve one hard compute problem. In the process, we built something bigger: a network for coordinating distributed GPU capacity. AI now needs the same foundation at a much larger scale.”

The company said the network totals about 4,000 GPUs. To serve inference workloads it added model execution tuning, adjusted scheduling for different task patterns, and provided managed routing of jobs. Early benchmarks provided by Boundless show inference costs up to 50% lower than major cloud providers for certain asynchronous workloads.

Boundless attributes the cost gap to using lower-cost capacity, including consumer-grade GPUs and machines originally purchased for crypto mining and proving. The firm plans to keep its proving operations running alongside the AI services.

Boundless also announced a role for its native token, ZKC. Under the model the company described, AI operators must stake ZKC to join the network and the size of an operator’s stake will be tied to their earning potential. The company did not disclose specific minimums or the formula linking stake size to earnings.

Other infrastructure providers have redirected compute capacity for AI, and some teams have shifted from earlier crypto-focused work toward verifiable off-chain compute and AI use cases. Shankar said open models give developers more control but that production inference remains constrained by cost, capacity and reliability. “Boundless is built to change that,” he added.