On-Chain GPU Rental Platform by @SumiroStudio
Singapore
Joined May 2026
- Tweets157
- Following6
- Followers1.5K
- Likes49
ICPX Protocol retweeted
We are happy to announce the he official partnership with Nosana.
Nosana powers ICPX!
Read down below on V1 release.
V1.1 to V2 releasing over the coming days.
github.com/orithellama/ICPX/…
Nosana 🤝 ICPX
@ICPXProtocol is a pay-per-job GPU compute platform for AI agents, model deploys, inference, fine-tuning and evals, giving builders a simple flow to pick a workload, pay and launch.
Nosana powers ICPX end to end, running the GPUs behind every job 🙌
🔗 icpx.cloud/
Host open source AI models on decentralised infrastructure, pay only for the jobs you run.
That is what ICPX is built for, and @nosana_ai is the compute network powering every single job behind the scenes. 🤝
@ICPXProtocol 🤝 @nosana_ai
Nosana 🤝 ICPX
@ICPXProtocol is a pay-per-job GPU compute platform for AI agents, model deploys, inference, fine-tuning and evals, giving builders a simple flow to pick a workload, pay and launch.
Nosana powers ICPX end to end, running the GPUs behind every job 🙌
🔗 icpx.cloud/
Need serious GPU power for AI but don’t want to buy hardware or fight cloud dashboards?
ICPX lets you rent it by the job.
icpx.cloud/models
Without ICPX, your apps are juggling 12 different integrations across laptops, home servers, enterprise clusters, cloud providers, GPU marketplaces and decentralised networks. Different APIs, different auth methods, different billing systems.
ICPX sits in the middle and coordinates all of it through one protocol layer.
ICPX Protocol retweeted
i came across @ICPXProtocol, now i can run heavy ai models on my calculator
🤝 Paid partnership
80,000,000 $ICPX tokens have been locked for 1 month via @streamflow_fi 🔒
app.streamflow.finance/token…
The ICPX Github Organisation is being set up.
1. Code that has reached a state of readiness beyond ICPX beta, will be open-sourced.
2. Github App integration into ICPX cloud, for one-click deploys similar to Vercel.
Currently we support GO, Nextjs, Python, Rust.
icpx.cloud
Our designer @SumiroStudio is working 25 hours a day to create all the compute and GPU cards.
Small preview, how do we like it?
ICPX Protocol retweeted
This is what $ICPX with @nosana_ai actually solves.
Most GPU providers rent you a machine and charge you whether it is running or not. You book the instance, it sits there, and the clock keeps ticking regardless of whether your workload is active. For small teams and startups that only need compute in bursts, that model does not make sense. You end up paying for a lot of nothing.
@ICPXProtocol and @nosana_ai change this completely. Nosana brings the decentralised GPU supply, a distributed network of available machines that ICPX can route jobs to on demand. ICPX sits on top and handles the orchestration, smart routing your workloads to active capacity and only billing you for the time your instance is actually running. When the job is done, the compute is released. You are not holding onto a machine you are not using.
The result is that you get enterprise grade hardware, H100s, H200s and RTX 4090s, accessible at a fraction of what centralised providers charge, and you only pay for useful work. Training runs, inference, model evaluation, fine tuning, AI agent execution. All of it on demand, all of it settled in $ICPX tokens, with every job burned into the protocol supply.
Decentralised compute that actually makes financial sense. Run everything, pay only for what you use.
icpx.cloud
Sharded inference can theoretically create better economics because 4090s are cheaper to run than H100s.
But the problem is sharded inference setups need to solve slow comms bandwidth for inference of large neural networks, which is not trivial.
So without major innovations, they have a shot at slightly cheaper smaller models, ergo slightly cheaper video/audio models.
But running medium to large LLMs is going to be very tough, and even if they eek it out, it will be slow and unreliable versus consumer products.
The path for founders here is to advance the state of the art of communication bandwidth bottlenecks and show you can inference large models cheaply and quickly on consumer hardware.
If you own the foundation models as well, that helps.
ICPX Protocol retweeted
Payment and runtime
Choose the payment asset and runtime behavior. Bounded jobs auto-stop. Services like terminal, hosted apps, model APIs, and agents can run until stopped.
Can be paid in ICPX, SOL or USDC.
End of thread @ICPXProtocol
ICPX now has a better aligned visual identity.
The logo is built from six independent nodes arranged in a continuous circle. Each node represents a different compute source, device, provider, cluster or network.
Together, they form one coordinated system.
icpx.cloud/branding
Traditional GPU -> Productive time.
With traditional GPU rental you are paying for a dedicated instance whether you are actually using it or not. Idle time costs the same as productive time. settle payments in $ICPX
ICPX Protocol retweeted
The problem with most crypto projects is simple.
They build a product, launch a token, and let people pay in dollars or SOL anyway. The token becomes a badge with a price attached. No one actually needs it.
$ICPX flips this completely.
GPU compute jobs on ICPX are priced and settled in $ICPX tokens. If you want to run a model, fine tune it, or execute an AI agent, you need the token to do it. Not hype. Actual demand.
This has worked before.
A project priced their services directly in their own token a couple years ago. Nothing massive. But because people had to own the token to access the service, the coin went to a 10 million market cap anyway. No one even really used it that much.
$ICPX has real infrastructure behind it.
Enterprise grade H100s, H200s and RTX 4090s. Powered by Nvidia and Google GPU cloud partner programs. No downtime. No waitlists. Accessible to anyone at a fraction of what centralised providers charge.
If the userbase is real, the market cap follows.
Not from hype. From builders who need GPU compute having to acquire and spend $ICPX to get it.
That is the difference between a token and a protocol.
icpx.cloud
We are adding support for GLM 5.2 and @qvac one-click deployments.
Will be rolling out in the coming days.
Stay tuned, multiple updates on progress every day!
Paid in ICPX tokens, SOL or USDC.
icpx.cloud
We have released Draft Version 1 of our ICPX paper.
As we mature and reach a milestone of launching the full web application, it is time to start releasing the first draft of our paper and roadmap.
Read below 👇
github.com/orithellama/ICPX/…
Our website will have a small overhaul too, with updated information and interactive roadmap.
icpx.cloud
A Universal Compute Orchestration Protocol
We are adding support for GLM 5.2 and @qvac one-click deployments.
Will be rolling out in the coming days.
Stay tuned, multiple updates on progress every day!
Paid in ICPX tokens, SOL or USDC.
icpx.cloud
ICPX protocol paper releases tomorrow.
The full foundational layer for ICPX and future.
Every search, document, design, software application, autonomous agent and business workflow increasingly depends on AI inference. Yet nearly all AI today relies on centralised cloud providers where users surrender control of their data, developers rent infrastructure indefinitely and compute remains concentrated inside a handful of hyperscale datacenters.
This model introduces structural weaknesses.
Data must leave the environment where it originates. Privacy depends on trusting third parties. Infrastructure failures affect millions of users simultaneously. Costs continue to increase as GPU demand grows faster than supply. Developers become dependent on centralised pricing, availability and policy decisions.
ICPX proposes a different architecture.
Rather than building another AI provider, ICPX functions as a universal orchestration layer capable of executing AI workloads across multiple classes of compute while maintaining a single developer interface.
Compute may originate from:
- Private home devices
- Office infrastructure
- Enterprise clusters
- Edge hardware
- Community-owned GPU operators
- Decentralised compute providers
- Future confidential compute networks
The protocol determines where workloads should execute according to privacy requirements, latency, available hardware, cost, trust assumptions and user-defined policy.
icpx.cloud
The full ICPX product roadmap will be released in the coming days.
Including:
- Staking ICPX
- Full circle protocol rewards to stakers
- One-click set-up "Local home networks" to run models
- Buyback mechanism
- More models, more compute
- RBAC for startups
- Personal use & business account
icpx.cloud
Big update releasing for ICPX tonight.
Everyone is able to spend ICPX to pay for compute. Prices are USDC related, so you will not overpay for a computing run or hosting.
Terminal has been updated for one-click deployments, and real-time log streams. There is also an option to run custom terminal commands.
ICPX web app is now fully responsive. In next big update, we'll wrap the web app into an Android app, and deploy ICPX to @solanamobile.
The UI/UX is slightly polished with a more minimal font, easier click-through pages and easier to use buttons [less noise more one-click actions].
You can also connect ICPX terminal to your local terminal and manage the GPU locally to connect and deploy your apps.
Enjoy! Update at 11 PM CET.
Check out the partial update @ icpx.cloud.
Powered by @nosana_ai.
