TopoLabs
[Network]

A global GPU network, paid per token.

Staked operators in 40+ regions serve every request. The router picks the best node, attestors check the work, and each token settles on-chain with a receipt.

Run a nodeHow routing works
Tokens served, last 30 days
0.0B
Models on the network
0+
Median time to first token
0ms
Settled on-chain to date
$0.0M
[How it works]

From request to receipt in under a second.

  1. 01

    Your app

    OpenAI-compatible SDKs, REST, or an agent framework. Change one line: the base URL.

    • Chat
    • Embeddings
    • Images
    • Video
  2. 02

    Topo Router

    Authenticates the key or wallet, picks the fastest eligible node, enforces spend limits.

    • Auth
    • Routing
    • Rate limits
    • Verification
    • Settlement
  3. 03

    Node network

    Staked GPU operators in 40+ regions serve the request and attest to the output.

    • H100 · B200 · 4090
    • 40+ regions
    • Staked
  4. 04

    Settlement

    The bill is split between the node and the network and settled on-chain with a receipt.

    • USDT
    • ETH
    • SOL
    • Base · Solana

Works with your stack:

OpenAI SDKVercel AI SDKLangChainLlamaIndexCrewAIMastraPythonTypeScriptOpenAI SDKVercel AI SDKLangChainLlamaIndexCrewAIMastraPythonTypeScript
[The network]

Built for every layer of the inference stack

  • [Developers]

    One API for every model

    Swap your base URL and keep the OpenAI SDK. Frontier and open models, one key, one bill.

    Read the quickstart
  • [Agents]

    Wallets that pay per token

    Give every agent its own spend limit and a USDT balance. No cards, no invoices, no humans in the loop.

    Agent wallets
  • [Node operators]

    Earn for every token you serve

    Plug in H100s or a single 4090. Stake, serve verified inference, get paid daily on-chain.

    Run a node
[Regions and hardware]

40+ regions, from B200 racks to a single 4090.

Requests go to the closest healthy node that already holds the model in memory. Max and Enterprise plans can pin traffic to one region.

RegionGPUsp50 TTFT
us-east-4VirginiaH100, B200164 ms
us-west-2OregonH100, 4090171 ms
eu-west-2LondonB200, H100182 ms
eu-central-1FrankfurtH100176 ms
ap-northeast-1TokyoH200189 ms
ap-southeast-1SingaporeH100, 4090198 ms
sa-east-1São Paulo4090231 ms
Plus 30+ smaller regions. Median time to first token over the last 30 days.

What you can serve

  • RTX 409024 GB
    Typical models
    Est. per month
    Muse Spark 1.3 and small open models
    $180 to $320
  • H10080 GB
    Typical models
    Est. per month
    DeepSeek V4 Pro, Qwen3.8-Max
    $1,400 to $2,100
  • B200192 GB
    Typical models
    Est. per month
    Frontier and large open models
    $2,900 to $4,200

Estimates at 70% utilization and current list prices. Not a guarantee of earnings.

[Node operators]

Run a node. Earn per token.

Keep up to 85% of every token your GPU serves, paid daily in USDT. Setup takes about 30 minutes.

Run a nodeTalk to us