Your Idle Phone Is the New Data Center, Decentralized AI Networks Say

Your phone spends most of its life bored. Decentralized AI networks want to put it to work. Acurast says more than 335,000 smartphones have joined its network, with roughly 40,000 fully active on a typical day handling web scraping, cybersecurity, AI inference and agent hosting. Co-founder Alessandro De Carli claims the active fleet matches the compute capacity of about 2,340 high-end data-center servers, benchmarked against the ASUS RS720A-E11.

Phase is chasing the same idea from a different angle. It launched Phase Proxy in October, letting phone owners sell their internet connections and blockchain validation to AI agents — a workaround for websites that block or throttle data-center traffic. The service charges 0.5 USDT for every 1,000 requests and each gigabyte of traffic. Phase Router comes next, sorting requests between models running on phones and big external providers like OpenAI and Anthropic based on task complexity, capability, performance and cost, founder Kirill Eves says. Desktops, servers and GPUs could join later.

De Carli argues phones have advantages servers can't match. Their built-in hardware security suits confidential computing and isolated environments that can install and run executables, and they shine on CPU-heavy jobs or work needing wide distribution and unfiltered internet access. "Servers are better at multi-threading, but smartphones offer unmatched compute efficiency," he says. Memory is the tradeoff — phones stay limited, while big servers still win at running many tasks in parallel.

Not everyone buys the pitch. Markus Levin, co-founder of decentralized infrastructure provider XYO, thinks these networks fit better as collectors and verifiers of real-world data than as AI model runners, since phones and sensors surface local detail that centralized datasets miss — but only if that data holds up. "A useful decentralized AI network should be judged by whether it solves a problem customers will pay to fix, not by its token price or node count," he says.

Then there's the coordination headache. A 2025 study found clusters of phones could handle image recognition competitively, though only in controlled conditions. A Google-backed project at the University of California, San Diego showed a cluster of 20 retired smartphones could absorb the peak computing load of a class of more than 75 students, grading faster than the university's standard cloud — again, in an organized setting. Real networks deal with phones that disconnect, overheat, drain, switch networks and vary wildly in power. Ethereum faces a similar ceiling, capped by network communication speed and its weakest devices.

The money has to work too. De Carli says there's "no fixed figure" for Acurast rewards, which swing on device uptime, benchmarked hardware, total ACU staked and live compute demand, with operators competing for an inflation-funded pool instead of direct customer payments. Phase will spend part of its service revenue buying PHASE on the open market and distributing it to node operators by contribution. Running a model on a phone isn't the hard part anymore — making the economics stack up while herding tens of thousands of devices into something that behaves like reliable cloud is.