Do we really need expensive GPUs for every AI task?
Think about an app that needs to filter spam, sort emails, or check content thousands of times a day. Using a large AI model for every little task can get pretty expensive.
@Acurast is exploring a different approach with Laya, a lightweight AI model running on Android phone CPUs. According to the team, each decision takes around 0.2–1 second.
Developers don't even need to own the phones themselves. They can deploy workloads on Acurast's decentralized compute network instead.
Of course, getting AI to run on a phone is just the first step. The bigger question is whether it can stay reliable and cost-effective at scale.
For me, the real potential isn't about how many phones join the network, but how much useful work those phones can actually get done.
Think about an app that needs to filter spam, sort emails, or check content thousands of times a day. Using a large AI model for every little task can get pretty expensive.
@Acurast is exploring a different approach with Laya, a lightweight AI model running on Android phone CPUs. According to the team, each decision takes around 0.2–1 second.
Developers don't even need to own the phones themselves. They can deploy workloads on Acurast's decentralized compute network instead.
Of course, getting AI to run on a phone is just the first step. The bigger question is whether it can stay reliable and cost-effective at scale.
For me, the real potential isn't about how many phones join the network, but how much useful work those phones can actually get done.