Boundless Enters AI Inference Market, Aiming for Up to 50% Cost Reduction

Repurposing GPU Network for AI Inference

Boundless, a company specializing in zero-knowledge (ZK) proof networks, announced on the 14th that it will enter the AI inference market. The company plans to repurpose its distributed GPU network, built for ZK proofs, for AI inference, with a formal launch scheduled for summer 2026.

The company operates a network composed of approximately 4,000 GPUs, which will be optimized for AI inference and provided as a managed inference infrastructure. They will handle everything from hardware optimization, workload distribution, batch processing, scheduling, monitoring, to operational support.

ZK (zero-knowledge proof) is a cryptographic technique that allows the correctness of certain information to be proven without revealing the content itself. It is used in blockchain transaction verification, and due to the high computational load of proof processing, a dedicated distributed GPU network is required.

As the practical application of AI inference expands, costs are becoming a bottleneck for companies. According to a Gartner survey, the proportion of inference spending within AI-optimized IaaS is expected to reach 55% by 2026 and exceed 65% by 2029.

Boundless points out that while many AI market prices are set based on high-performance data center GPUs, in reality, many inference workloads can be adequately processed using consumer-grade GPUs or hardware procured for cryptocurrency mining and proofing. The company has adjusted its surplus GPU capacity for inference and confirmed up to a 50% cost reduction compared to hyperscalers in initial benchmarks. They believe this effect is particularly significant for asynchronous processing workloads.

Role of ZKC Token

Boundless has indicated its intention to incorporate its own token, "ZKC," into the AI network. Similar to the traditional ZK proof network, AI operators will be required to stake ZKC when participating in the network, with revenue opportunities varying according to the amount staked.

CEO Shiv Shankar explained that the network was initially built to solve a single computational challenge related to ZK proofs, but as a result, a foundation for decentralized adjustment of GPU capacity has been established, and AI also requires a similar foundation on a larger scale.

The company has revealed that it has begun transitioning to the AI inference business alongside its announcement on the 14th, with a formal product launch planned for this summer. They will continue to operate the ZK proof network in parallel.