Idle

The first universal compute layer. Turn any idle device - GPU, laptop, PC - into a node on the global AI compute network. |

[Idle @IdleProtocol]

$IDLE is now live on Solana.

AjLhrxN2yrCe45Y2KGPMZCkBm6NpN43jWqPkdZq6pump

Quick rundown on what it does and why it exists.

IDLE Protocol takes 15% of every gateway payment. That fee is the entire token economy. 10% auto-swaps to $IDLE on Jupiter every hour and sends it to the burn address. 5% covers infrastructure and development. The other 85% goes straight to the resource owner's wallet, same block.

Staking gives you a bigger cut. Base split is 85/15. Stake 10k $IDLE and you keep 87%. Stake 50k and it's 90%. 250k gets you 95%. The protocol adjusts your split ratio on the next payout.

Burning $IDLE unlocks features: priority routing (your jobs match first), advanced task types, analytics on your gateway traffic, and custom endpoint domains. Priority routing means lower latency and higher uptime for your resources.

The thesis: $IDLE supply can only go down. Every gateway call generates a burn. More resources on the network means more API calls, which means more fees, which means more buying pressure and more tokens removed from circulation permanently. The flywheel is mechanical, not speculative.

https://t.co/0n8skIkzCO

[Idle @IdleProtocol]

Introducing IDLE.

Your PC earns from AI agent tasks. Your wallet compounds yield. Your agent sells its downtime.

Everything idle. Now earning.

[Elon Musk @elonmusk]

@levie People have no idea

[Idle @IdleProtocol]

Agents that work 24/7 need compute that runs 24/7.

Most of the world's phones, laptops and gaming GPUs sit unused for most of the day. IDLE Protocol puts that hardware to work on inference and pays its owners per request in USDC on Solana.

The buildout doesn't have to start from zero.

[Aaron Levie @levie]

The compute needed for the stage of AI we’re about to enter is going to be insane.

Personal agents, agent swarms defending enterprises, agents that review all of our code for security issues, agents that process nearly all enterprise data in workflows, background agents working on tasks 24/7, and on and on.

This will take a mix of inference volume that is going to keep growing by orders of magnitude, as well as infrastructure that agents need alongside the tokens. They need computers, networking, file systems, and so on.

We’re only in the early stages of what this buildout is going to look like.

[Idle @IdleProtocol]

AMD says it will substantially increase GPU supply in 2027. Until then, analysts expect the AI compute crunch to stay acute.

Premium data center GPUs now take 36 to 52 weeks to arrive. They sell for 25 to 35% above list price, and cloud GPU instances cost 15 to 25% more. The hyperscalers get new capacity first. Everyone else waits.

One adviser's guidance to CIOs: "The key measure should be cost per successful AI workload, not simply GPU price."

IDLE Protocol doesn't wait for a shipment. The network runs on phones, laptops and gaming GPUs that are already in people's homes. Requests for open models go to devices that are online now, and operators are paid per request in USDC on Solana.

Data center GPUs take up to a year to arrive. IDLE's supply is already plugged in.

https://t.co/n426K6ndQm

[PayAI Network | x402 Facilitator @PayAINetwork]

BlackRock just wrote x402 into its AI thesis.

"AI represents machine-native intelligence, while digital assets represent machine-native money."

We read the whole whitepaper: the demand thesis, the data that says it's early & the layer it doesn't cover 👇 https://t.co/6Fxxhsu7hR

[Idle @IdleProtocol]

Bain just put a number on the AI buildout. The industry needs $6 trillion a year in revenue by 2031 to justify the data centers it's building.

Bain thinks today's AI products get to about $1.8 trillion of that. The other $4.2 trillion has to come from things that don't exist yet.

The spending is already committed. $5 to $6.5 trillion on 150 gigawatts of new capacity through 2030, with single campuses now costing $150 to $200 billion. And $68 billion of U.S. projects were blocked or delayed in the June quarter alone.

Bain's lead author says the industry needs "a wave of innovation that will dwarf what mobile and cloud unlocked" to make the math work.

IDLE Protocol doesn't have that math problem. The hardware on the network was bought years ago for other reasons: phones, laptops, gaming GPUs. There's no campus to finance and nothing to pay back. Operators get paid per request in USDC on Solana when their device does the work.

Data centers have to earn back what they cost. IDLE's supply was already paid for.

https://t.co/kwWdMVjp5F

[Idle @IdleProtocol]

IDLE Protocol now supports Xiaomi MiMo.

Xiaomi released MiMo-V2.6 last week - Flash and Pro. The company that ships tens of millions of phones every quarter is now shipping open frontier-class models built for the agent era.

Both are live on the IDLE network. MiMo-V2.6-Flash runs across consumer GPU nodes - fast, cheap, built for exactly the hardware our operators already run. V2.6-Pro routes through NVIDIA NIM and high-VRAM operator rigs. Same OpenAI-compatible API, same per-request USDC settlement on Solana.

Xiaomi builds its models for the devices sitting in a billion pockets. IDLE is the network that turns those devices into infrastructure.

The hardware maker went open. The compute layer was already here.

[Idle @IdleProtocol]

IDLE Protocol now supports Qwen3.8.

Alibaba just shipped the second-largest open-weights model ever released. Only Kimi K3 is bigger. A year ago a release this size would have been unthinkable outside a closed lab. Now it's a download.

Both ends of the family route through IDLE. The flagship runs through high-VRAM infrastructure for frontier tasks. The distilled Qwen3.8 27B is Apache 2.0 licensed and runs directly on IDLE's 24GB consumer nodes, RTX 3090s and 4090s serving a model distilled from one of the largest ever trained.

Same OpenAI-compatible endpoint. Paid per request in USDC on Solana.

The frontier keeps going open. The network keeps serving it.

[Idle @IdleProtocol]

The world's largest asset manager just described IDLE's production system as the future.

Page 8 of this paper: AI agents will discover compute through marketplace APIs, compare on price and latency, and pay per request over x402 in stablecoins.

That exact flow has been live on IDLE all year. Agents hit our endpoints, pay in USDC on Solana over x402, and consume compute with no human in the loop. 550,000+ on-chain transactions settled this way.

BlackRock calls it the machine-native economy. We've been running it in production.

[BlackRock @BlackRock]

Our latest research paper explores the growing connection between AI and digital assets and explains why broad AI adoption may drive new demand, utility and applications across the digital asset economy. https://t.co/z5T88Orble

[Idle @IdleProtocol]

We're hiring at IDLE.

The network keeps growing and we need more hands:

Backend / Infra Engineer - Rust or Node, Solana experience a plus
DevRel - help developers and agents integrate the API
Growth / Community Lead - X, Telegram, partnerships
Node Operations - onboarding and supporting operators

DMs open.

[Idle @IdleProtocol]

IDLE Protocol has submitted to Stocklana!

[Solana @solana]

STOCKLANA

The stock market is open for building.

One week to build something innovative with stocks on Solana. $100K in prizes.

Sept 11 - 18, opening bell to closing bell:
https://t.co/JU1uP3xWh7

[PayAI Network | x402 Facilitator @PayAINetwork]

Solana vs Base: same USDC, same x402 flow, very different fee economics. Here's what actually happens to your costs when the network gets busy 🧵

[Idle @IdleProtocol]

IDLE Protocol now supports million-token context inference.

Million-token windows became table stakes at the closed frontier this year - OpenAI, Anthropic, Google all ship them. What changed is that the open-weight side caught up: DeepSeek V4 Flash runs a full 1M-token context, MIT-licensed, at 79.0% on SWE-bench Verified - within 1.6 points of the best open-weights score ever recorded, at a fraction of frontier pricing.

That capability now routes through IDLE. Flag a job as long-context and the network handles it: entire codebases, full contract stacks, complete transcript archives - read in one pass instead of chunked into fragments. Same OpenAI-compatible endpoint, paid per request in USDC on Solana.

The network now routes across three dimensions. Cost-optimized jobs run on 78,000+ consumer nodes. Latency-critical jobs hit the LPU speed tier. Long-context jobs get the million-token window.

Speed became a tier last month. Context just became one too.

[Idle @IdleProtocol]

There's a false floor under AI pricing, and enterprises are starting to plan around it.

The reported numbers behind the cheap API calls: OpenAI's inference costs grew 4x last year to roughly $8.4 billion, and margins fell to 33%. Anthropic's inference ran 23% over their own projections. Frontier pricing is subsidized by venture capital to win market share - and subsidies end. The cheap fallback tiers are already disappearing, enterprise billing is already being restructured.

Companies see it. CNBC reported the CEO of Lindy moved 100% of his AI traffic to open-weight models: "you could see that cost curve crash to the ground... it's a matter of survival for the business." Salesforce is on track to spend $300 million on tokens this year. Menlo Ventures found 74% of AI builders now run majority-inference workloads, up from 48% a year ago - always-on production, not experiments. When pricing normalizes upward, that bill has nowhere to hide.

The hedge is structural, not negotiated: open-weight models on infrastructure whose costs fall as it grows. IDLE routes inference across 78,000+ distributed nodes - pay per request in USDC on Solana, no subsidy to expire, no contract to restructure, no single vendor deciding what your fallback tier costs.

Subsidized pricing is borrowed time. Distributed pricing is just the cost.

https://t.co/mHkWzEV5gv

[Idle @IdleProtocol]

IDLE Protocol now supports Intel AI PCs.

Intel Core Ultra machines are the most widely shipped AI-capable laptops on earth - millions of them on desks everywhere, idle most of the day. Starting today they connect to IDLE like any other node: install the client, hardware gets detected on registration, jobs route to what the machine can handle, USDC settles on Solana per completed job.

The timing isn't an accident. Yesterday at Hot Chips, Intel laid out its full agentic AI architecture - from Diamond Rapids in the data center down to Wildcat Lake bringing AI silicon to mainstream laptops. Their CTO's words: "Agentic AI is fundamentally changing how we design and deliver computing." Heterogeneous compute, routed by capability. That's been IDLE's architecture all year.

This completes the set. NVIDIA GPUs. AMD Radeon. Apple Silicon. Ryzen AI Max. Groq LPUs. Now Intel AI PCs. Every major silicon vendor's hardware, earning on one network.

If your laptop has an Intel Core Ultra in it, it's a node now.

[Idle @IdleProtocol]

@PayAINetwork @solana Massive!

[Idle @IdleProtocol]

The compute shortage isn't an American problem. It's planetary.

South China Morning Post, yesterday: Chinese AI firms are stretching scarce Nvidia supply because domestic chips can't handle the hard workloads. An inference startup VP put it bluntly - demand for high-quality tokens "far outstripped supply," and agentic AI is making the squeeze worse by the month.

Both superpowers, effectively unlimited capital, entire national chip industries mobilized - and neither can produce enough compute to run the AI already deployed. The US is bottlenecked on power and permits. China is bottlenecked on silicon. Demand doesn't care about either.

There's exactly one compute pool that no export control, grid queue, or fab shortage touches: the hardware already distributed across the planet. Hundreds of millions of capable devices in homes and offices on every continent, idle most of the day.

IDLE connects that pool. 78,000+ operators serving inference, training, and agent tasks - routed by capability, settled per request in USDC on Solana, running in every jurisdiction at once.

Nations are fighting over chips. The network runs on the ones already shipped.

https://t.co/762uKXxnrE

[Idle @IdleProtocol]

IDLE Protocol x Cloudflare's Agents SDK.

IDLE has been an x402-native compute provider since day one. Which means as of now, any agent built on Cloudflare's SDK can discover and consume IDLE compute out of the box - inference, data reads, web tasks, all 16 endpoints - paying per request in USDC on Solana, settled automatically.

Millions of developers build on Cloudflare. Their agents just gained a distributed compute network they can pay natively, and we didn't have to change a line- the standard did the integration.

That's what betting on open protocols looks like. Build to the standard, and every platform that adopts it becomes distribution.

[Idle @IdleProtocol]

"The demand is there, and supply is massively short on power, data center capacity, compute, memory and network. We're in the very, very early infancy."

That's Cisco's president, to CNN, two days ago - answering the question everyone keeps asking about whether AI is a bubble. His point: the dotcom era built supply ahead of demand. This time it's inverted. Demand is here now, and supply can't keep up on any layer.

The numbers behind him: Nvidia's GPU order book has grown to $1 trillion through 2027, double a year ago, with lead times near a full year. All three HBM memory suppliers are sold out for 2026. And agentic AI rewrote the compute math entirely - a single user running agents can demand 10-100x more compute than one running prompts.

Every layer of new supply is bottlenecked: chips, memory, power, land. But one layer isn't - the hundreds of millions of capable devices already built, already powered, already connected, sitting idle most of the day.

IDLE connects them. 78,000+ operators serving inference, training, and agent tasks, routed by capability, settled per request in USDC on Solana.

Supply is short everywhere except where it's already installed.

https://t.co/J1T5mNKVHo

[Idle @IdleProtocol]

IDLE Protocol has upgraded to x402 V2.

Some context on why this matters. x402 - the payment rail IDLE has settled on since day one -just became an official internet standard. The Linux Foundation launched the x402 Foundation with 40 members and vendor-neutral governance. The protocol has processed 161 million+ transactions, and 98.6% of agent payments settle in USDC. The standard won, and IDLE has been running on it from the start.

V2 changes what agents can do on the network. The headline feature is session tokens: instead of a full payment handshake on every request, an agent opens a paid session once and streams hundreds of jobs through it — inference, data reads, health checks — settled efficiently at session close. Lower latency per job, native support for the multi-step agent workflows that now dominate real usage.

One handshake. Hundreds of jobs. Same per-request pricing, none of the per-request friction.

Pay-per-request built the network. Sessions are how agents actually work. IDLE now speaks both.

[Idle @IdleProtocol]

IDLE Protocol now supports real-time inference.

Some workloads can't wait. Voice agents, trading systems, live monitoring - they need the first token in milliseconds, not seconds. GPUs weren't built for that. LPUs were.

IDLE now routes latency-critical jobs through Groq LPUs - the fastest inference silicon on earth, now part of NVIDIA's Vera Rubin platform. Hundreds of tokens per second, near-instant response, on open-weight models. Same OpenAI-compatible endpoint: flag a job as real-time and it routes to the speed tier automatically.

Cost-optimized jobs run across 78,000+ consumer nodes. Latency-critical jobs hit LPUs. One network, routed by what the job actually needs.

Speed just became a tier on IDLE.