Boundless, a distributed compute startup that began with a focus on zero-knowledge proof settlement for Bitcoin, has expanded its 4,000-GPU network to support artificial intelligence inference, marking a shift from a purely crypto-focused compute model toward a broader market for on-demand processing power.
The company said the same infrastructure used to handle demanding cryptographic workloads is now being applied to AI tasks, where demand for graphics processing units remains high and access to affordable compute is a major challenge for developers, startups and enterprises. Boundless said its existing network has been reconfigured to run AI inference jobs while continuing to support its original verification and proof-related services.
The expansion places Boundless at the intersection of two high-demand compute markets: zero-knowledge proofs, which require intensive processing to verify data without exposing the underlying information, and AI inference, the stage where trained AI models are used to generate responses, predictions, images, code or other outputs. Both workloads depend on large amounts of GPU power, efficient scheduling and reliable coordination across machines.
According to the company’s announcement, Boundless has optimized its GPU cluster for inference through changes in hardware configuration, workload distribution, automated routing and scheduling. The network previously helped manage computational connections between Ethereum, Base and Bitcoin. Boundless said it will continue handling verification tasks while also opening capacity for AI applications.
The company reported early data indicating that inference costs on its network can be as much as 50% lower than those charged by major cloud providers. Boundless attributed the cost reduction to the use of lower-cost GPUs, including consumer-grade units and repurposed mining hardware, as well as improved routing of workloads across its distributed network.
The move reflects a broader trend in which infrastructure originally built for cryptocurrency mining or blockchain computation is being adapted for high-performance computing and AI. Bitcoin mining companies have increasingly explored AI and data-center services as a way to diversify revenue and improve utilization of their power and hardware assets. Boundless, however, represents a different category: a protocol-based compute network attempting to apply decentralized coordination to AI workloads.
The company said it plans to operate its proof network alongside the new AI platform. It also intends to introduce a staking model tied to its ZKC token, with participation and potential earnings for operators linked to the size of their token commitment.
Why the expansion matters
Boundless’s announcement comes as AI inference is becoming one of the largest and fastest-growing parts of cloud computing demand. While much of the public focus in artificial intelligence has centered on training large models, inference is the recurring cost that comes after a model is built. Every chatbot response, image generation request, automated code suggestion, voice transcription or enterprise AI query requires inference.
That makes inference a continuous expense rather than a one-time development cost. For companies running AI products at scale, the price and availability of GPUs can directly affect margins, product performance and growth plans. If compute is too expensive or unreliable, teams may limit usage, slow feature development or rely on smaller models that produce weaker results.
Research firm Gartner expects inference workloads to account for 55% of all cloud AI spending this year, rising above 65% by 2029. That forecast points to a market in which the ability to run models efficiently may become more important than the ability to train them from scratch.
Boundless is positioning its distributed GPU model as a lower-cost alternative to centralized cloud platforms. Instead of relying only on large data centers with premium hardware, the company is using a broader pool of machines that can include less expensive GPUs and repurposed equipment. The goal is to match incoming jobs with available compute capacity wherever it exists in the network.
Chief Executive Shankar said open models give teams more freedom, but running those models every day is still constrained by high costs and a shortage of reliable servers. Boundless said its platform now routes heavy data queries to available machines in real time, using software that was originally designed to coordinate proof generation and verification.
From zero-knowledge proofs to AI inference
Boundless was initially built around zero-knowledge proof computation, a technical area closely tied to blockchain scaling, privacy and verification. Zero-knowledge proofs allow one party to prove that a statement is true without revealing the underlying data. In blockchain systems, they can be used to verify transactions, compress data, connect networks or improve privacy.
These proofs can be computationally expensive. Generating them often requires specialized software and significant processing power. That made distributed GPU coordination a natural fit for Boundless’s original business model.
AI inference has different end uses, but similar infrastructure demands. It requires fast access to GPUs, good scheduling, strong reliability and the ability to process many requests without excessive delays. A model may be hosted in one location, but requests can arrive from many users and applications at once. The system must decide where to send each job, how to balance loads and how to avoid bottlenecks.
Boundless said its network architecture was already designed to coordinate distributed GPU capacity. That same design can be extended from cryptographic proving to AI workloads. In practice, that means the platform can continue to serve blockchain verification tasks while also processing model queries from AI applications.
The company’s shift does not appear to be a full departure from its original protocol strategy. Instead, Boundless is seeking to broaden the market for its compute layer. The proof network remains part of the platform, while AI inference adds a new demand source for the same underlying hardware.
Cost claims and hardware strategy
The company’s claim of inference costs up to 50% below major cloud providers is likely to draw attention from AI teams watching infrastructure expenses. Cloud platforms remain the dominant providers of AI compute, particularly for companies that need scale, reliability and managed services. But their pricing can be a hurdle for smaller teams and high-volume applications.
Boundless said its lower costs come partly from using a wider range of GPUs. These include consumer GPUs and repurposed mining units that may not fit the standard model of a large cloud data center but can still perform useful inference work. The company also pointed to workload scheduling as a source of savings, arguing that better routing can improve utilization and reduce idle capacity.
This approach depends heavily on software coordination. Distributed compute networks can be cheaper in theory, but they must handle differences in hardware quality, connection speed, location, uptime and performance. For AI inference, reliability and latency are especially important. A low-cost network that cannot return responses quickly or consistently may not be suitable for many commercial applications.
Boundless said its system has been adjusted to handle these requirements through automated routing and scheduling. The company has not framed the AI expansion as a replacement for major cloud providers, but as an alternative compute layer that can serve workloads where cost efficiency is especially important.
A new use for idle mining hardware
The expansion also highlights a practical issue in crypto infrastructure: large amounts of hardware can become underused when market conditions change, mining profitability falls or older machines are replaced. GPUs and mining-related equipment may sit idle if they are no longer profitable for their original purpose.
AI has created a potential second market for some of that capacity. Not all mining hardware is suitable for AI workloads. Application-specific integrated circuits used for Bitcoin mining, for example, are designed for one narrow task and generally cannot be repurposed for AI inference in the same way GPUs can. But GPU-based systems, including those once used for other blockchain computations, may be adaptable.
Boundless said it is using lower-cost GPUs, including repurposed mining hardware, as part of its network. That approach could allow operators with unused or underused equipment to connect to a market with large compute demand.
The idea is not unique to Boundless. Several companies connected to Bitcoin mining and data-center infrastructure have pursued high-performance computing and AI hosting as a way to improve revenue streams. What distinguishes Boundless is its protocol-oriented structure and its attempt to coordinate distributed GPU supply through a network model rather than only through traditional data-center contracts.
Token model and operator participation
Boundless said it plans to introduce staking through its ZKC token. Under the planned model, operator participation and potential earnings would be tied to the amount of ZKC committed. The company has not presented the token model as separate from the compute network; instead, it appears intended to help govern or secure participation in the system.
Token-based coordination is common in decentralized infrastructure projects, where networks often use staking to encourage reliable behavior and align operators with protocol performance. However, the success of such models depends on whether the network generates real demand for the service it provides.
For Boundless, that demand would need to come from two sides: cryptographic proof and verification workloads, and AI inference customers seeking lower-cost compute. If the network can attract steady usage, the token model may have a clearer connection to economic activity. If demand is inconsistent, token incentives alone may not be enough to sustain long-term participation.
Traders following the sector are likely to focus on whether Boundless can show measurable usage beyond announcements, including active workloads, repeat customers, operator growth and stable performance under real-world conditions. The company’s planned late-summer product launch may become an important milestone for assessing how much demand exists for its AI inference platform.
The broader market shift
The move by Boundless reflects a larger convergence between blockchain infrastructure and AI infrastructure. During the previous decade, crypto encouraged the buildout of specialized computing, mining facilities, energy agreements and distributed coordination systems. AI is now creating demand for many of the same resources, especially access to power, chips and efficient workload management.
The connection is not automatic. Blockchain networks and AI platforms have different users, revenue models and performance needs. AI customers often care less about decentralization as an ideology and more about cost, reliability, speed and data handling. A decentralized compute provider must therefore compete on practical terms, not just on network design.
Boundless’s announcement suggests the company understands this shift. Its pitch is centered on lower inference costs and better use of existing GPU capacity. The crypto component remains important, particularly through the proof network and ZKC staking model, but the AI expansion is aimed at a much larger pool of compute demand.
If inference takes a growing share of cloud AI spending as Gartner expects, alternative compute networks may find more room to compete. Enterprises and developers may be willing to test lower-cost platforms for workloads that do not require the most advanced chips or the tightest latency guarantees. Smaller model providers, open-source AI teams and high-volume application developers may be especially sensitive to pricing.
Still, execution will be critical. Boundless will need to prove that distributed GPUs can deliver dependable inference at scale, not just cheaper theoretical pricing. The company will also need to manage quality control across varied hardware, maintain uptime and provide tools that make it easy for AI developers to use the platform.
For now, Boundless’s expansion shows how infrastructure built for one wave of digital technology is being redirected toward another. A network designed to verify cryptographic proofs across Bitcoin, Ethereum and Base is now being used to process AI workloads. The outcome will depend on whether the company can convert idle or low-cost compute into a reliable service for a market that is growing quickly and becoming increasingly cost-conscious.
Explore how blockchain and AI intersect in real markets with our guide to web3, AI, and crypto evolution.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
