The massive data centers that support the artificial intelligence movement have grown roughly as popular as infant toys manufactured from asbestos.
Protesters globally point out the enormous amounts of water and power that the facilities demand, whereas financial markets express concern regarding the revenue needed to reimburse the $38 billion expense required to construct a single one-gigawatt AI data center.
Nevertheless, might there exist another method to supply energy to the artificial intelligence boom resting inside your pocket right now?
As of October 2026, there are an estimated 7.68 billion cellular connections globally, while billions of mobile phones currently sit idle inside pockets, on tabletops, and within drawers.
Every single one features a processor, memory, and a web connection, which signifies an enormous reserve of processing capability just waiting to be utilized.
Consequently, could a network of ordinary smartphones genuinely drive artificial intelligence at a massive scale?
Putting smartphones to work
Decentralized AI networks are developing with the objective of converting everyday mobile phones into a distributed infrastructure layer, thereby lessening dependence upon expensive, centralized data centers.
Acurast reports that it has integrated greater than 335,000 mobile phones onto its system. Co-founder Alessandro De Carli mentions that roughly 40,000 of those remain entirely active on a typical day, supplying computational power for operations spanning from web data extraction and network security to artificial intelligence inference and agent hosting.
Data center locations across the US. Source: World Resources Institute
Related: DePIN grows into a $10B sector despite token slump, Messari says
He states that mobile phones prove particularly efficient concerning specific workloads. Their integrated hardware security capabilities “excel” in domains such as confidential computing or a “segregated space” capable of deploying and executing programs. They also align well with tasks that demand high CPU usage, or those necessitating broad distribution alongside “unfiltered entry” to the internet.
“Servers perform better at multi-threading, but smartphones deliver unparalleled compute efficiency,” De Carli informs Magazine.
While standard mobile phones possess restricted memory, and major servers maintain an advantage when managing multiple operations simultaneously, he asserts that numerous workloads can still fit within the accessible memory limits.
Considered collectively, the active mobile phones across the platform “equal the compute capacity of approximately 2,340 high-end data center servers,” he maintains, evaluated against the ASUS RS720A-E11, a high-density server designed for artificial intelligence and high-performance computing.
Mobile connectivity in 2026. Source: Datareportal
An alternative initiative, Phase, debuted Phase Proxy during October, permitting mobile phone proprietors to supply web connectivity and blockchain validation for artificial intelligence agents. Founder Kirill Eves clarifies that, “websites frequently block or throttle automated web traffic originating from data centers,” meaning the framework utilizes users’ mobile and residential connections as a proxy instead. Phase intends to bill 0.5 USDT per 1,000 requests alongside 1GB of traffic.
The subsequent phase involves Phase Router, which will direct simpler inquiries toward models operating on mobile phones within the network, or alternatively route them to major external suppliers such as OpenAI and Anthropic, relying upon “task complexity, model capabilities, performance and pricing.”
The network could eventually grow beyond mobile phones to encompass desktop computers, servers, and graphics processing units, he points out.
Nonetheless, Markus Levin, co-founder of decentralized infrastructure provider XYO, suggests that decentralized platforms might prove better fitted for gathering and confirming real-world information instead of executing artificial intelligence models.
Mobile phones and sensors can supply localized data that centralized datasets overlook, he remarks, though exclusively if the information remains accurate and resilient against tampering.
“A functional decentralized AI network ought to be evaluated on whether it resolves a dilemma that consumers will pay to fix, rather than by its token valuation or node count,” he notes.
Coordination is the tricky part
Among the primary obstacles concerning mobile phone networks involves synchronizing the devices to deliver predictable hardware, electricity, and connectivity, matching the manner of data centers.
A 2025 research study evaluated small collections of mobile phones executing image recognition processes and discovered that clusters of phones could furnish competitive performance under specific circumstances, yet those outcomes emerged from a regulated testing environment.
Related: Web3 companies turn to hardware with crypto-powered phones, consoles
A Google-supported undertaking situated at the University of California, San Diego determined that a group of 20 retired mobile phones could manage the peak computing load for a classroom exceeding 75 students, achieving grading speeds surpassing the university’s standard cloud configuration. Regardless, those instruments were also organized inside a controlled setting, rather than dispersed globally.
Within the physical world, a distributed platform must cope with mobile phones that disconnect, experience overheating, deplete their charge, switch networks, and possess differing levels of processing strength and memory.
Inside a blockchain network such as Ethereum for instance, capacity is presently constrained by the duration required to communicate across the system, alongside the processing capabilities of the lowest-specification units.
20 retired smartphones could handle a class of more than 75 students. Source: Google Research
Equally, when tasks are divided amongst multiple phones, sluggish links or disconnected hardware can disrupt operations. Allocating separate, autonomous assignments across accessible mobile phones is generally more straightforward.
Can the economics work?
Even if technical hurdles can be conquered, networks still need to prove financially viable for both consumers and device proprietors alike.
Utilizing pre-existing mobile hardware avoids the substantial capital expenses associated with constructing a data center, though expenses encompass factors including connectivity, synchronization, and incentives for participants within the network.
Regarding compensation, De Carli mentions that there exists “no fixed figure” since payouts for Acurast contributors fluctuate based on “device uptime, hardware specifications (benchmark performance), total ACU staked and active compute demand” originating from developers. He states that operators vie for a pool of inflation-backed rewards, instead of obtaining direct disbursements from users.
Phase network incentives rely upon uptime, participation, and user demand.
“Phase will utilize a portion of its service revenue to acquire PHASE on the open market and distribute it toward node operators in accordance with their contributions to the network,” Eves states.
Executing an artificial intelligence model to operate correctly on a mobile phone is no longer the primary hurdle — achieving functional economics while coordinating tens of thousands of phones to function as a dependable cloud utility remains the true challenge.
Magazine: Will gold’s ‘infinite mint bug’ make Bitcoin more valuable?
Originally published at https://cointelegraph.com/magazine/can-decentralized-phone-networks-power-the-ai-revolution?utm_source=rss_feed&utm_medium=rss&utm_campaign=rss_partner_inbound.
