Boundless, a company originally created for zero-knowledge proof (ZKP) computations on the Bitcoin network, is making a strategic pivot. Its distributed GPU network, comprising approximately 4,000 graphics processors, will now serve inference tasks — performing computations for artificial intelligence models. This decision, announced by CEO Shiva Shankar, marks a shift from purely cryptographic applications to the broader AI market.
Initially, the Boundless network acted as a coordinator and verifier, linking Ethereum, Base, and Bitcoin. However, developers have now optimized it for AI needs. The initial benchmarks are impressive: inference costs are nearly 50% lower than those of major cloud providers, especially for asynchronous tasks. This gap is achieved by leveraging cheap resources — consumer-grade graphics cards and equipment previously purchased for cryptocurrency mining and ZKP proof generation. This reminds me of how miners are gradually retraining as providers of computing resources for AI, and Boundless is following the same path.
The company is not completely abandoning ZKP, but details of further work in this direction are not disclosed. Instead, Boundless is introducing the native ZKC token into the new ecosystem: AI network operators will need to stake it to join, with the stake size tied to potential revenue. This is a smart move that encourages long-term participation and creates an economic link between the token and real computations.
Although the announcement lacks specifics on business strategy, the very fact of the shift in focus highlights a global trend: the crypto industry is increasingly pivoting toward AI computations. Consider TeraWulf, which raised $3.5 billion in July for a data center for Anthropic. Miners and ZKP startups realize that their GPU assets can be more profitable in the AI space than in pure cryptocurrency operations.
My expert commentary: This move by Boundless is not just a change of direction, but a pragmatic response to market realities. AI inference requires enormous computing resources, and decentralized GPU networks could become serious competitors to centralized cloud providers. However, success will depend on the network's scalability and reliability — for now, 4,000 GPUs is a drop in the ocean compared to Amazon Web Services or Google Cloud. Nevertheless, for niche asynchronous tasks, this could be a breakthrough.