SKN | Can Decentralized Smartphone Networks Provide an Alternative to AI Data Centers?

Key Points:

• Smartphones as Distributed Computing Infrastructure: Decentralized AI projects are attempting to turn unused mobile devices into distributed computing networks. Acurast reports more than 335,000 smartphones onboarded, with approximately 40,000 active on an average day.
• Technical Capabilities and Limitations: Smartphones can support selected AI inference, cybersecurity, and distributed computing workloads, but inconsistent connectivity, limited memory, battery constraints, and hardware differences make large-scale coordination difficult.
• Economic Viability Remains Unproven: Decentralized networks avoid some of the capital costs associated with building data centers, but must still establish reliable customer demand, sustainable incentives, and competitive service economics.

The expansion of artificial intelligence is creating growing demand for computing infrastructure, placing pressure on energy resources, capital investment, and the economics of large-scale data centers. As companies commit billions of dollars to building specialized facilities, decentralized infrastructure projects are exploring a different approach: using the processing capacity of smartphones already owned by billions of people.

With an estimated 7.68 billion mobile connections worldwide as of October 2026, smartphones represent a potentially significant distributed resource. Each device contains processing power, memory, and internet connectivity that could contribute to selected computing tasks.

However, transforming this scattered hardware into a dependable alternative to centralized infrastructure presents substantial technical and commercial challenges. The opportunity is not simply to run AI models on phones, but to coordinate large numbers of independently operated devices into a service that customers can reliably use.

Smartphones Enter the Decentralized AI Infrastructure Market

Several decentralized computing projects are attempting to turn ordinary smartphones into infrastructure for AI-related workloads.

Acurast reports that it has onboarded more than 335,000 smartphones, approximately 40,000 of which are fully active on an average day. The network supports activities including web scraping, cybersecurity, AI inference, and agent hosting.

Co-founder Alessandro De Carli argues that smartphones offer advantages in certain computing environments, particularly where confidential computing, isolated execution environments, and widely distributed internet access are important.

Although conventional servers generally perform better at handling multiple tasks simultaneously, smartphones can provide efficient computing capacity for workloads that do not require extensive parallel processing. De Carli also argues that many applications can operate within the memory limitations of mobile devices.

According to his estimates, Acurast’s active devices collectively provide computing capacity comparable to approximately 2,340 high-end data center servers, using the ASUS RS720A-E11 as a benchmark. This comparison reflects the project’s stated assessment rather than an independently established measure of equivalent performance across all AI workloads.

Another initiative, Phase, is approaching decentralized infrastructure through internet connectivity and blockchain services. Its Phase Proxy product, launched in October, allows smartphone owners to provide mobile and residential internet connections for AI agents.

Founder Kirill Eves says automated traffic originating from conventional data centers is frequently blocked or restricted by websites. By routing requests through residential and mobile connections, Phase aims to provide an alternative for accessing online services.

The company plans to charge 0.5 USDT per 1,000 requests and 1GB of traffic. Its proposed Phase Router would subsequently direct simpler requests to models operating on participating smartphones or route more demanding tasks to external providers such as OpenAI and Anthropic.

This approach could allow decentralized networks to complement established AI infrastructure rather than replace it entirely.

Coordination Remains the Main Technical Challenge

The availability of billions of smartphones does not automatically translate into reliable computing capacity. Unlike data centers, where hardware, electricity, cooling, and connectivity are managed within controlled environments, decentralized mobile networks depend on devices with widely varying capabilities and operating conditions.

Smartphones can disconnect unexpectedly, overheat, lose battery power, switch networks, or become unavailable when their owners need them. Differences in processing power and memory also make it difficult to guarantee consistent performance.

Research offers some evidence that groups of smartphones can perform useful computing tasks under controlled conditions.

A 2025 study examining small smartphone clusters for image recognition found that groups of devices could deliver competitive performance in certain testing environments. However, those results do not establish that comparable performance can be maintained across a geographically distributed network.

A separate Google-backed research project at the University of California, San Diego demonstrated that a cluster of 20 retired smartphones could handle the peak computing workload of a class with more than 75 students. The system also achieved faster grading times than the university’s standard cloud setup.

These experiments demonstrate the potential of coordinated mobile hardware, but they were conducted in organized environments where device availability and network conditions could be managed.

In a global decentralized network, coordinating thousands of independent devices introduces additional communication and reliability constraints. When a task is divided across several smartphones, a slow connection or an unavailable device can delay the entire operation.

Consequently, distributing separate and independent tasks across available devices may prove more practical than splitting complex workloads that require constant communication between multiple phones.

Decentralized Networks May Serve Data Verification Better Than AI Processing

Not every decentralized infrastructure provider believes smartphones will become a major source of AI computing capacity.

Markus Levin, co-founder of decentralized infrastructure provider XYO, argues that distributed networks may offer greater value in collecting and verifying real-world information than in executing AI models.

Smartphones and connected sensors can capture localized information that centralized datasets may not adequately represent. This could support applications requiring geographically distributed observations and independent data verification.

However, the usefulness of this information depends on its accuracy and resistance to manipulation. A decentralized system that collects large volumes of data does not necessarily provide reliable information unless it can establish appropriate verification mechanisms.

Levin argues that the commercial value of decentralized AI infrastructure should be measured by whether it addresses a problem customers are willing to pay to solve, rather than by the number of network nodes or the market price of its associated token.

This distinction is important for projects seeking to establish sustainable businesses. A large device network may demonstrate participation, but customer demand and measurable service quality ultimately determine whether the infrastructure has economic value.

The Economics of Mobile Computing

Decentralized smartphone networks could reduce the need for some of the substantial upfront investment required to construct traditional data centers. However, avoiding large hardware purchases does not eliminate the costs associated with operating a computing service.

Connectivity, device coordination, network security, customer acquisition, and participant compensation all contribute to the overall economics.

Acurast does not offer a fixed reward for device operators. According to De Carli, compensation depends on factors including device uptime, hardware specifications, benchmark performance, ACU staking, and demand from developers. Participants compete for inflation-funded rewards rather than receiving direct payments from customers for each completed task.

Phase follows a different incentive structure. Eves says the network intends to use part of its service revenue to purchase PHASE tokens on the open market and distribute them to node operators according to their contributions.

Both models depend on balancing participant incentives with actual demand for network services. Inflation-funded rewards can encourage participation, while revenue-linked compensation may connect operator earnings more directly to commercial activity. Neither approach, however, automatically guarantees sustainable profitability.

For decentralized networks to compete with established cloud providers, they must demonstrate that their services are dependable and cost-effective while maintaining sufficient incentives for device owners.

Outlook

Decentralized smartphone networks offer a potentially useful additional layer of AI infrastructure, particularly for distributed inference, independent computing tasks, internet connectivity, and real-world data collection. Their strongest opportunity may lie in complementing centralized data centers rather than replacing the specialized hardware required for demanding AI workloads.

The next stage of development will depend on whether projects such as Acurast and Phase can demonstrate consistent performance across larger networks, attract paying customers, and establish compensation models that remain viable as participation grows.