The sprawling data centers that underpin the AI revolution have become a growing target for public criticism. Campaigners around the world highlight the massive amounts of energy and water the infrastructure requires, while markets are increasingly worried about the revenue needed to pay back the roughly $38 billion cost of building a single, one-gigawatt AI data center. But is there another way to power the AI revolution — one that is already sitting in people's pockets?

As of October 2026, there are an estimated 7.68 billion mobile connections worldwide, and billions of smartphones are currently lying unused in pockets, on desks and in drawers. Each device has a processor, memory and an internet connection, and together they represent a vast pool of computing power waiting to be put to work. The question is whether a network of ordinary smartphones can realistically support AI at scale.

Putting smartphones to work

Decentralized AI networks are emerging that aim to turn everyday smartphones into a distributed infrastructure layer, reducing reliance on costly, centralized data centers.

Acurast says it has onboarded more than 335,000 smartphones to its network. Co-founder Alessandro De Carli says that around 40,000 of those are fully active on an average day, providing computing capacity for tasks ranging from web scraping and cybersecurity to AI inference and agent hosting.

According to De Carli, smartphones are particularly efficient for certain workloads. Their built-in hardware security features "excel" in areas such as confidential computing or an "isolated environment" that can install and run executables, and the devices are also well suited to workloads that are "CPU intensive," or which require wide distribution and "unfiltered access" to the internet.

"Servers are better at multi-threading, but smartphones offer unmatched compute efficiency," De Carli told Cointelegraph Magazine.

While ordinary smartphones have limited memory, and big servers retain an advantage when handling multiple tasks in parallel, he said many workloads can still fit comfortably within the memory available on individual devices.

Taken as a whole, the active phones on the network "match the compute capacity of about 2,340 high-end data-center servers," he claimed, benchmarked against the ASUS RS720A-E11, a high-density server built for artificial intelligence and high-performance computing (HPC).

Another project, Phase, launched Phase Proxy in October, which allows smartphone owners to provide internet connectivity and blockchain validation for AI agents. Founder Kirill Eves explained that "websites routinely block or rate-limit automated traffic originating from data centers," so the system employs users' mobile and residential connections as a proxy instead. Phase plans to charge 0.5 USDT per 1,000 requests and 1GB of traffic.

The next step is Phase Router, which will direct simpler requests to models running on phones in the network, or route them instead to major external providers like OpenAI and Anthropic, based on "task complexity, model capabilities, performance and cost." Eves said the network could eventually expand beyond smartphones to desktops, servers and GPUs.

Markus Levin, co-founder of decentralized infrastructure provider XYO, said decentralized networks may be better suited to collecting and verifying real-world data than running AI models. Phones and sensors can provide localized information that centralized datasets miss, he said, but only if the data is accurate and resistant to manipulation.

"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," Levin said.

Coordination is the tricky part

One of the biggest challenges facing mobile phone networks is coordinating devices to deliver predictable hardware, power and connectivity in the same way data centers do.

A 2025 study tested small groups of smartphones carrying out image recognition tasks and found that clusters of phones could deliver competitive performance under certain conditions, though the results came from a controlled testing environment rather than a live deployment.

A Google-backed project at the University of California, San Diego found that a cluster of 20 retired smartphones could handle the peak computing workload of a class of more than 75 students, with grading times faster than the university's standard cloud setup. Those devices, however, were also organized in a controlled environment rather than scattered around the world.

In real-world conditions, a distributed network has to contend with phones that disconnect, overheat, run out of battery, switch networks, and vary widely in processing power and memory. In a blockchain network like Ethereum, for example, capacity is currently limited by how long it takes to communicate across the network and by the processing abilities of the lowest-spec devices.

Similarly, when tasks are split between several phones, slow connections or disconnected devices can throw a spanner in the works. Distributing separate, independent tasks across available phones is generally more straightforward.

Can the economics work?

Even if the technical challenges can be overcome, these networks still need to make financial sense for customers and device owners alike.

Using existing smartphone hardware does not incur the high capital costs of building a data center, but costs remain in the form of connectivity, coordination and incentives for the people who keep devices on the network.

When it comes to compensation, De Carli said there is "no fixed figure," because rewards for Acurast participants vary by "device uptime, hardware specifications (benchmark performance), total ACU staked and active compute demand" from developers. Operators compete for a pool of inflation-funded rewards rather than receiving direct payments from customers, he said.

Phase's network rewards depend on uptime, participation and customer demand. "Phase will use a portion of its service revenue to purchase PHASE on the open market and distribute it to node operators based on their contributions to the network," Eves said.

Running an AI model properly on a phone is no longer the main challenge — getting the economics to stack up while coordinating tens of thousands of phones to behave like a reliable cloud service is the real trick.

This article is based on reporting by Cointelegraph Magazine.