Acurast’s AI deployment now runs on 280,000 smartphones worldwide

A new Acurast AI deployment is turning ordinary Android phones into the processing layer for a decision-making artificial intelligence system, bypassing the data centers that normally handle this kind of workload. The company has pushed its open source model, Laya, onto its decentralized smartphone network, letting real smartphones around the world make structured AI decisions in roughly a second or less. It’s a small but telling experiment in what happens when compute power comes from people’s pockets instead of server farms.

Key takeaways

• Acurast deployed the open source AI model Laya across its decentralized smartphone compute network, running decisions entirely on mobile edge nodes.
• Laya is a “System 1” decision model built for fast, structured outputs like classification and moderation, not conversational text.
• Decisions typically take between 0.2 and 1 second and run on Android smartphone CPUs without dedicated GPUs.
• The Acurast network now spans over 280,000 smartphones across more than 175 countries, with on-chain transactions passing 918 million.
• Laya’s model weights and code are open under the Apache 2.0 license, and operators earn ACU tokens for contributing processing capacity.

Acurast Deploys Laya AI Model on Decentralized Smartphone Network

Acurast has put Laya, an open source AI decision model, to work on its decentralized smartphone network, allowing structured decisions to be processed directly on Android devices rather than in centralized data centers. The model was built by Convai Innovations and released under an Apache 2.0 license, meaning anyone can inspect, copy or redeploy it. Instead of generating conversational answers the way chatbots do, Laya is designed to evaluate a defined situation and pick from a set of possible actions, using formats that include choosing among fixed options, scoring against a range, or making a yes-or-no call.

Network Scale and Geographic Reach

The scale behind this Acurast AI deployment is what makes it notable. According to the company, more than 280,000 smartphones spread across over 175 countries have joined the network, and the number of on-chain transactions processed has climbed past 918 million. That’s a sharp jump from May 2025, when Acurast raised $5.4 million to build out its smartphone-powered decentralized cloud and reported around 72,000 devices and 256 million processed transactions at the time, as previously reported by crypto.news.

Laya Runs Real Time AI Decisions on Android Smartphones

Every decision inside this deployment happens on a smartphone connected to the network, not on a remote server. Convai Innovations describes Laya as a “System 1” decision model, a term meant to distinguish quick, structured judgment calls from the slower, generative reasoning typical of large language models. Because Laya uses a non-autoregressive architecture, it doesn’t build an answer word by word — it evaluates the input and lands on a decision in one pass.

Workloads Supported by Laya

Acurast has rolled out several live demos to show what this looks like in practice. Laya has been set up to play Snake and Tetris, choosing movements and deciding where falling pieces should land, and a Doom-based test pushes the model through decisions around movement, aiming, shooting and item collection. Beyond gaming, the model sorts emails into inbox, spam or phishing categories, classifies headlines as news, satire, clickbait or manipulation, and has been tested for spotting prompt injection attempts against AI assistants and flagging toxic messages in live chats. Users can feed their own letters, headlines or prompts into the demos and watch the model’s calls in real time.

Performance Metrics and Hardware Utilization

Speed is part of the pitch. Acurast says decisions in this deployment typically take roughly 0.2 to 1 second, handled entirely by smartphone CPUs rather than dedicated GPUs, with work distributed dynamically across whichever devices are available on the network at that moment. Laya’s repository lists a 421 million parameter English checkpoint, a separate 322 million parameter multilingual model supporting more than 100 languages, and another checkpoint trained specifically for typed decision workflows such as customer service, invoice processing and security incidents.

Security, Incentives, and Technical Architecture

Every output from this Acurast AI deployment carries cryptographic proof, and the phones doing the work get paid for it. The network relies on Trusted Execution Environments already built into modern smartphones to isolate each workload and shield data while a decision is being processed. That architecture lets Acurast avoid asking operators to buy dedicated servers or GPUs — the same consumer hardware people already carry becomes the compute layer. In return, smartphone operators earn ACU, Acurast’s native token, for supplying processing capacity to the network.

Purpose and Implications of Acurast’s Deployment

The point of this deployment is narrower than building a general-purpose chatbot rival — it’s about handling the kind of repeated, structured decisions that don’t need a large generative model on the other end. Routing, moderation and classification tasks can run on smaller models like Laya instead of pinging a centralized provider’s API for every request, which is the core argument behind this particular decentralized smartphone network setup.

Leadership Perspective on Decentralized AI

Acurast founder Alessandro De Carli framed the deployment as proof that lightweight AI doesn’t need a data center behind it. “By running Laya on Acurast, we’ve changed that,” De Carli said. “We are proving that System 1 AI can run securely and at scale on hardware everyone already owns, providing a truly open, decentralized alternative to centralized cloud lock-in.” He added that the goal is to show developers what’s possible with this kind of compute: “Our goal is to open developers’ eyes to the quite incredible opportunities that come from using Acurast compute. This deployment proves that decision-oriented AI can run cheaply, verifiably, and without a data center in sight on a decentralized smartphone network today.”

Model Limitations and Developer Accessibility

Laya isn’t marketed as flawless. Its own repository notes that the general model performs poorly on some zero-shot typed decision tests and recommends specializing it for particular tasks, along with keeping choice questions under roughly 20 options and recalibrating probabilities using application-specific data. Still, because the weights and code sit publicly under Apache 2.0, developers can deploy Laya themselves without waiting on API access or paying a centralized cloud provider for server capacity. That openness fits into a broader pattern of pushing AI compute onto consumer devices — Gaia, for instance, introduced an AI smartphone in September 2025 built to run AI locally while letting users contribute compute to its own decentralized network.

FAQ

What is the Laya AI model deployed by Acurast?

Laya is an open source System 1 AI decision model designed for quick structured decisions, running on decentralized smartphone networks.

How does Acurast run AI decisions without centralized cloud servers?

Acurast runs Laya on smartphones in its decentralized network, using Android CPUs and trusted execution environments, avoiding the need for centralized data centers.

What types of tasks can Laya handle on the Acurast network?

Laya processes tasks including classification, gaming, navigation, security, and content moderation.

How are smartphone operators rewarded for participating in the Acurast network?

Operators earn ACU tokens when their smartphones provide processing capacity to run AI decisions.

{"@context":"","@type":"FAQPage","mainEntity":[{"@type":"Question","name":"What is the Laya AI model deployed by Acurast?","acceptedAnswer":{"@type":"Answer","text":"Laya is an open source System 1 AI decision model designed for quick structured decisions, running on decentralized smartphone networks."}},{"@type":"Question","name":"How does Acurast run AI decisions without centralized cloud servers?","acceptedAnswer":{"@type":"Answer","text":"Acurast runs Laya on smartphones in its decentralized network, using Android CPUs and trusted execution environments, avoiding the need for centralized data centers."}},{"@type":"Question","name":"What types of tasks can Laya handle on the Acurast network?","acceptedAnswer":{"@type":"Answer","text":"Laya processes tasks including classification, gaming, navigation, security, and content moderation."}},{"@type":"Question","name":"How are smartphone operators rewarded for participating in the Acurast network?","acceptedAnswer":{"@type":"Answer","text":"Operators earn ACU tokens when their smartphones provide processing capacity to run AI decisions."}}]}

Article produced with the assistance of artificial intelligence and reviewed by the editorial team.