Startups Build Marketplaces to Rent Idle PCs for AI Inference

A new wave of startups is developing decentralized marketplaces to monetize idle consumer graphics processing units for artificial intelligence inference workloads. Abu Dhabi-based Far Labs and Austin-based Evolving Edge are building platforms that function similarly to peer-to-peer rental services, routing smaller open-source model tasks to personal gaming rigs and home workstations. Both companies are entering an emerging sector already represented by Utah-based Salad, which currently hosts over sixty thousand daily active consumer GPUs across its network. Technically, the platforms employ distinct distribution strategies. Far Labs utilizes a proprietary scheduler that fragments models across multiple machines, with a central orchestrator and load balancer reassembling outputs to achieve claimed latencies of one hundred milliseconds or less. Evolving Edge leverages Ray, the open-source distributed computing framework standard in conventional data centers, and has open-sourced its node software to allow hardware owners to audit the processes executing on their devices. To address security concerns regarding malicious code or unauthorized data access, both operators enforce strict sandboxing protocols. Workloads run in isolated environments with encrypted communications and hard limits on hardware and network access, ensuring customers cannot interact directly with host systems. The economic viability of distributed consumer GPU computing remains unproven. Historical data from Salad, which previously facilitated cryptocurrency mining before pivoting to AI tasks, indicates that hardware owners historically received a modest fraction of platform revenue. With power consumption for high-end cards like the RTX 4090 reaching four hundred fifty watts under sustained load, monthly electricity expenses can easily surpass user payouts at current retail rates. Neither Far Labs nor Evolving Edge has disclosed specific compensation structures, leaving profitability uncertain for individual participants. Advocates of decentralized compute emphasize infrastructure resilience. Company founders argue that distributed networks can absorb individual node failures without disrupting service, contrasting their architecture with centralized cloud providers that remain vulnerable to large-scale outages. Nevertheless, the sector must overcome significant credibility hurdles following past projects that masked non-functional cryptocurrency operations as AI infrastructure. Current platforms route legitimate inference requests rather than speculative token rewards, but independent validation of latency claims and cost efficiency is required before the model can achieve mainstream enterprise adoption. As demand for AI computation outpaces centralized data center capacity, peer-to-peer GPU networks represent a pragmatic, albeit untested, solution. Whether these marketplaces can sustain profitable operations for hardware providers while delivering reliable, low-latency services to developers will determine their long-term viability in the rapidly evolving artificial intelligence infrastructure landscape.