In the AI + Crypto space, there are many projects focused on aggregating edge or idle computing power and using crypto incentives to provide compute for AI. The AI DePIN project @Acurast, recently listed on Bithumb, is a typical example. Here’s a quick overview:
1) Demand for AI inference keeps changing. The hardware arms race for model compute, which initially required H100-class hardware, is gradually giving way to more fragmented, lightweight demand driven by Agent applications. When Agents frequently call small models to make decisions, use tools, and perform real-time inference, centralized data centers are no longer necessarily the best solution. Edge computing—where individual compute needs are modest, but latency, privacy, verifiability, and cost matter more—becomes the primary need.
Acurast chose the world’s billions of smartphones as its entry point. While phones may not offer the most powerful compute, they are the most widely distributed and accessible edge nodes, and come with built-in hardware security capabilities. For users, this makes them the lowest-barrier, lowest-cost way to participate.
2) How does it work? Smartphones generally come with a TEE (Trusted Execution Environment) and hardware security modules. These provide basic inference capabilities and can also address key challenges around privacy and verifiability. Once users register a Processor app and enroll their idle phones as nodes, developers can deploy programs such as lightweight LLM inference, API calls, scheduled tasks, and data processing. The phones can then run node tasks and earn rewards.
3) The network currently has over 300,000 devices running across more than 170 countries. It has processed over 1 billion on-chain transactions and supported hundreds of thousands of deployments, providing an initial validation of demand and the token model. However, many DePIN projects in the past have failed because their incentive models weren’t sustainable. Acurast has taken steps to address this:
1. It links rewards to real commitments. Nodes must stake base:0xc5fed7c8ccc75d8a72b601a66dffd7a489073f0b and commit to providing a certain amount of compute. They can be slashed if they fail to deliver, rather than simply receiving rewards based on how long they stay online.
2. Around 70% of the bulk of protocol inflation goes to the staked compute pool, and the real cash flows paid by compute buyers go to the stakers who actually run workloads. The allocation to early investors is kept relatively low, with the aim of tying the token model to real workloads and reducing sell pressure from those who make no contribution.
4) From the user’s perspective, participation means installing an app and putting an idle phone to work. This allows the DePIN service to reach everyday users rather than professional miners. For developers on the demand side, Acurast also offers deployment workflows for everything from simple APIs to LLM inference, using various Node.js or Cargo environments, without the need to manage additional servers.
Clearly, the low-barrier design is intended to help the business scale in the real world. That said, its biggest challenge remains on the demand side: how many Agent-related inference, tool-use, and data-processing use cases will there be?
That’s the gist.
Only strong demand, combined with growth on the supply side and a sustainable token model, can make this business model work. If you’re asking for a precise answer right now, there definitely isn’t one.
But following the direction in which the narrative is evolving, Acurast is positioned in the ecosystem at the point where AI is shifting from an arms race for large-model compute to the deployment of Agent applications. As long as the trend is there, demand cannot be disproven outright—and scaling the business is then just a matter of time.
Original post (X): https://x.com/tmel0211/status/2114617571693764749
1) Demand for AI inference keeps changing. The hardware arms race for model compute, which initially required H100-class hardware, is gradually giving way to more fragmented, lightweight demand driven by Agent applications. When Agents frequently call small models to make decisions, use tools, and perform real-time inference, centralized data centers are no longer necessarily the best solution. Edge computing—where individual compute needs are modest, but latency, privacy, verifiability, and cost matter more—becomes the primary need.
Acurast chose the world’s billions of smartphones as its entry point. While phones may not offer the most powerful compute, they are the most widely distributed and accessible edge nodes, and come with built-in hardware security capabilities. For users, this makes them the lowest-barrier, lowest-cost way to participate.
2) How does it work? Smartphones generally come with a TEE (Trusted Execution Environment) and hardware security modules. These provide basic inference capabilities and can also address key challenges around privacy and verifiability. Once users register a Processor app and enroll their idle phones as nodes, developers can deploy programs such as lightweight LLM inference, API calls, scheduled tasks, and data processing. The phones can then run node tasks and earn rewards.
3) The network currently has over 300,000 devices running across more than 170 countries. It has processed over 1 billion on-chain transactions and supported hundreds of thousands of deployments, providing an initial validation of demand and the token model. However, many DePIN projects in the past have failed because their incentive models weren’t sustainable. Acurast has taken steps to address this:
1. It links rewards to real commitments. Nodes must stake base:0xc5fed7c8ccc75d8a72b601a66dffd7a489073f0b and commit to providing a certain amount of compute. They can be slashed if they fail to deliver, rather than simply receiving rewards based on how long they stay online.
2. Around 70% of the bulk of protocol inflation goes to the staked compute pool, and the real cash flows paid by compute buyers go to the stakers who actually run workloads. The allocation to early investors is kept relatively low, with the aim of tying the token model to real workloads and reducing sell pressure from those who make no contribution.
4) From the user’s perspective, participation means installing an app and putting an idle phone to work. This allows the DePIN service to reach everyday users rather than professional miners. For developers on the demand side, Acurast also offers deployment workflows for everything from simple APIs to LLM inference, using various Node.js or Cargo environments, without the need to manage additional servers.
Clearly, the low-barrier design is intended to help the business scale in the real world. That said, its biggest challenge remains on the demand side: how many Agent-related inference, tool-use, and data-processing use cases will there be?
That’s the gist.
Only strong demand, combined with growth on the supply side and a sustainable token model, can make this business model work. If you’re asking for a precise answer right now, there definitely isn’t one.
But following the direction in which the narrative is evolving, Acurast is positioned in the ecosystem at the point where AI is shifting from an arms race for large-model compute to the deployment of Agent applications. As long as the trend is there, demand cannot be disproven outright—and scaling the business is then just a matter of time.
Original post (X): https://x.com/tmel0211/status/2114617571693764749