Backed by Y Combinator

    The compute already exists. It's just idle.

    Lilac is a network of idle GPUs. Run inference, reserve clusters, fine-tune models, and process batch jobs on unused capacity, or join the network to automatically monetize downtime in your cluster.

    Use idle GPUsSupply idle GPUs

    Trusted by

    Z.aiY CombinatorOsmosisUnderstudy LabsSaturn CloudEden AINanoGPTLinumInfronCoral BricksFloot

    Why Lilac

    Using wasted compute lowers costs.

    30%Cheaper tokens

    Idle supply lets Lilac price open frontier models below typical on-demand rates, with no contracts or minimums.

    60%Cheaper batch jobs

    Run container jobs on idle, spot-style GPU compute instead of reserving dedicated hardware.

    40%Idle capacity, reclaimed

    Most clusters run far below capacity. Lilac turns that wasted GPU time into compute you can actually buy.

    Products

    One GPU Network.
    Four ways to use Compute.

    Inference

    Frontier models, idle-GPU prices

    Pay per token for open frontier models on warm idle GPUs. Point your OpenAI SDK at Lilac and go, with no contracts or minimums.

    6+ modelsGet started

    MiniMax M2.7

    Speed
    -tok/s
    TTFT
    -
    Input /M
    $0.30
    Cache /M
    $0.055
    Output /M
    $1.20
    200KFP8

    MiniMax M3

    Speed
    -tok/s
    TTFT
    -
    Input /M
    $0.28
    Cache /M
    $0.05
    Output /M
    $1.10
    1MFP8

    Kimi K2.6

    Speed
    -tok/s
    TTFT
    -
    Input /M
    $0.70
    Cache /M
    $0.20
    Output /M
    $3.50
    262KINT4

    GLM 5.1

    Speed
    -tok/s
    TTFT
    -
    Input /M
    $0.90
    Cache /M
    $0.27
    Output /M
    $3.00
    203KFP8/NVFP4

    GLM 5.2

    Speed
    -tok/s
    TTFT
    -
    Input /M
    $0.90
    Cache /M
    $0.27
    Output /M
    $3.00
    524KFP8/NVFP4

    Gemma 4 (31B)

    Speed
    -tok/s
    TTFT
    -
    Input /M
    $0.11
    Cache /M
    --
    Output /M
    $0.35
    262KFP8

    Subscriptions

    Credits that stretch on idle supply

    A flat monthly price for a pool of inference credits. When GPUs sit idle, idle-supply discounts let those credits buy up to 12x more.

    Up to 12x valueGet started
    Basic$10/moUp to $80 valuePro$30/moUp to $300 valueMax$100/moUp to $1200 value

    Batch

    Run containers to completion

    Submit a container image and a command. Lilac runs it on spare GPU capacity with pricing at $1/hr H100s and $1.50/hr H200s.

    Priced per secondContact us
    Batch jobsPrivate beta
    eval-suite-nightlyrunning

    H100 $1.00/hr · 2.82 hrs · ~$2.82

    train-baselinerunning

    H200 $1.50/hr · 1.14 hrs · ~$1.71

    data-processdone

    H100 $1.00/hr · 0.46 hrs · ~$0.46

    Clusters

    Cluster quotes, brokered for free

    Tell us the GPU type and scale. We source competitive quotes from our neo-cloud partners and help you offset the cost with idle-GPU software.

    H100 to B300Contact us

    Reserve H100s

    Pricing is estimated. We gather quotes in real time.

    *Estimate pricing

    Setup

    Idle GPUs in. Working GPUs out.

    Most clusters sit 30-50% idle. Run the Lilac operator on your Kubernetes and earn from that spare capacity. Your own jobs always come first. Learn more.

    Get onboarded

    Get onboarded as a supplier

    Provider onboarding is by demo and waitlist. Book a call to get set up.

    Install our operator

    Apply the Lilac GPU Agent to your cluster and declare a GPUPool. Your GPUs never leave your infrastructure.

    Earn on your idle GPUs

    Keep 70% of revenue from workloads your cluster runs. Payouts are usage-based with per-cluster reporting. Settlement is handled directly with our team.

    Blog

    Notes from the team

    Updates, thinking, and technical deep-dives from the Lilac team.

    Lucas Ewing7 May 2026

    We're partnering with MiniMax to bring M2.7 to Lilac

    We are partnering with MiniMax to bring commercially licensed MiniMax M2.7 access to Lilac.

    Read details...
    Lucas Ewing7 May 2026

    How to keep frontier open weights viable

    Why Lilac supports open-weight licensing, and why commercial rights can help more frontier models stay open.

    Read details...
    Lucas Ewing1 May 2026

    Kimi K2.6 is live on Lilac

    Kimi K2.6 is now available on Lilac with OpenAI-compatible chat completions, 262K context, cache-read pricing, and no commitments.

    Read details...
    Lucas Ewing27 Apr 2026

    Cache read pricing is now live on Lilac

    Supported Lilac models now show lower cache read rates for repeated context, making long-context and agent workloads cheaper to run.

    Read details...
    View all posts

    Idle GPU Network

    The compute is already on.
    Put it to work.

    Get an API keySupply your GPUs