Acurast Deployed AI Decision Model on Mobile Network
The Laya model now runs on decentralized smartphone hardware, avoiding the need for centralized cloud data centers.
Updated on Sept. 28, 2026 in Artificial Intelligence
Live Poll
Do you trust decentralized smartphone networks to handle complex AI tasks as well as cloud centers?
Acurast has deployed the open-source Laya AI decision model across its decentralized compute network of 280,000 smartphones. This deployment demonstrates that AI workloads can be processed locally on mobile CPUs rather than in centralized data centers.
Why it matters
By leveraging a massive distributed smartphone network, Acurast offers an alternative to centralized cloud computing for AI workloads. This approach seeks to bypass the infrastructure lock-in typical of traditional data center providers.
The Laya model uses a non-autoregressive architecture, with the English checkpoint containing 421 million parameters and a multilingual version holding 322 million parameters. Decisions are processed on smartphone CPUs in 0.2 to 1 second.
The players
Acurast
A decentralized compute provider that aggregates processing capacity from over 280,000 global smartphones.
Convai Innovations
The developer of the Laya open-source AI model licensed under Apache 2.0.
Gaia
A hardware developer that introduced an AI-focused smartphone in September 2025.
The details
The network utilizes Trusted Execution Environments—secure, isolated hardware areas within a processor—to protect data and prevent workload tampering. Work is dynamically distributed across participating Android devices, which are compensated with ACU tokens. The Laya model, developed by Convai Innovations under an Apache 2.0 license, is designed to provide structured decisions without the repetitive computation required by standard autoregressive models.
Timeline
May 2025: Acurast raised $5.4 million for its network.
September 2025: Gaia introduced an AI smartphone.
September 28, 2026: The Laya model deployment was documented.
The Tech Race
Acurast is competing against centralized hyperscalers by demonstrating that AI inference can be distributed across commodity mobile hardware. This approach contrasts with the proprietary cloud-based stacks dominated by major platform providers.
Android smartphone users participating in the Acurast network can earn ACU tokens by contributing processing capacity to decentralized AI tasks. The performance of these models depends on the local hardware specs, with current decision latency ranging from 0.2 to 1 second per request.
The takeaway
This deployment highlights an emerging shift toward edge-based inference to reduce reliance on centralized compute. Watch for future performance benchmarks comparing these decentralized mobile networks against established cloud-based AI inference services.
Further reading
For more on the current state of decentralized computing, explore our coverage of Artificial Intelligence.
Live Poll
Do you trust decentralized smartphone networks to handle complex AI tasks as well as cloud centers?
