GPU Network

AI without the datacenter.

A mesh of consumer GPUs serving open-source models.

0x243508fb1c4ff24351b595385d0bbb7cb158fa7e

[GPU Network @GPUNetwrk]

GPU Network $GPU is live.

CA: 0x243508fb1c4ff24351b595385d0bbb7cb158fa7e

Run a node on any machine - Mac, Linux, or even a browser tab - and earn GPU Points for every inference task the network routes to you. 

The hardware is already built. It was just waiting for a network.

https://t.co/DN4FNMtZJN

[GPU Network @GPUNetwrk]

After shipping the network with native and WebGPU providers, we're now moving fast toward split-model inference.

Soon, several community GPUs will cooperatively serve models too large for any single card to hold - Llama 70B on consumer hardware.

Idle GPUs shouldn't sit unused.

[GPU Network @GPUNetwrk]

The next phase: points become $GPU.

Everything you earn in Season 0 leads here - rewards paid in $GPU, settled on Robinhood Chain.

Keep your GPU serving the mesh.
https://t.co/U04GXCJqLz

[GPU Network @GPUNetwrk]

DEX paid

https://t.co/HoEzTCaKwn

[GPU Network @GPUNetwrk]

https://t.co/4IyvP0tO6s

[GPU Network @GPUNetwrk]

Running a large open-source model -say, Llama 3.3 70B - takes a graphics card with at least 40 gigabytes of VRAM. A single consumer card can't hold it. Most workstation cards can't either. Which is why the largest open models today mostly live in racks inside corporate datacenters, and why the ones you actually get to talk to are hosted on hardware that costs six figures per node.



We're now working on the hardest problem on our roadmap: split-model inference across consumer GPUs.

[GPU Network @GPUNetwrk]

If we ship this to production, GPU Network becomes the first place where a normal person can call a 70B-class open model without either owning a $30,000 workstation or paying a corporate API by the token. Community-hosted frontier open source, in the literal sense of the word. It's the hardest thing on the roadmap. It's also the one we're most excited to build.

[GPU Network @GPUNetwrk]

Almost every AI prompt sent on the internet today ends up inside one of five corporate datacenters. That's an unusual concentration of compute, controlled by a small group of companies that effectively set the price of AI for everyone else. We're building the alternative. GPU Network is an open mesh where any consumer GPU can serve inference, and where every request is routed across hardware that belongs to the community rather than to a corporation.

[GPU Network @GPUNetwrk]

Every model on the network is open-source: Llama, Gemma, Qwen. Every response is generated by a real, community-operated node. None of it touches a third-party API. Open infrastructure for AI.

Anyone joins. Anyone serves.