Boundless Expands ZK GPU Network Into AI Inference, Targets Up to 50% Lower Costs

Boundless, which has built a computing network for zero-knowledge proof, or ZK, workloads, is expanding its decentralized graphics processing unit network into the artificial intelligence inference market.

The company said in a July 14 statement that it will repurpose the GPU network it originally built for ZK computations into infrastructure aimed at meeting AI inference demand. AI inference is the execution stage in which a trained model responds to user requests or performs tasks.

“We started four years ago to solve one hard computing problem,” Chief Executive Officer Shiv Shankar said. “In the process, we built something bigger: a network that coordinates distributed GPU capacity.”

AI now requires the same foundation at a much larger scale, he said. Open models give teams more control over what they build, but inference in production environments remains constrained by cost, capacity and reliability.

“Boundless was built to change that,” he added.

Boundless initially built its zero-knowledge proof network to handle transaction demand from the crypto industry. In doing so, it developed technology to coordinate GPU supply, assign high-performance computing jobs and operate distributed hardware as a single stable system.

The company sees that orchestration layer as applicable to AI inference infrastructure as well. Boundless currently operates a network of about 4,000 GPUs and plans to optimize it for AI workloads before offering it as managed inference infrastructure.

Many AI inference tasks do not require expensive data-center GPUs, according to Boundless. Some can be handled efficiently on consumer-grade GPUs or lower-cost hardware originally purchased for crypto mining and proof computations.

Early benchmarks showed Boundless's inference costs were as much as 50% lower than those of large cloud providers, the company said. It expects the cost advantage to be especially strong for asynchronous workloads that do not require an immediate response.

Boundless also plans to give its native token, ZKC, a role in the AI network. As in its existing proof network, AI operators will need to stake ZKC to participate, with the size of the stake tied to revenue opportunities.

Boundless plans to continue operating its zero-knowledge proof network alongside its AI inference infrastructure business. A formal product launch is scheduled for late this summer.