Lilac bets idle enterprise GPUs become a spot market; YC S25, $1.5M seed

Lilac sells flexible GPU contracts and an open-source scheduler that unifies on-prem and cloud GPUs into a single compute fabric for AI teams.

That underused enterprise GPUs can become a cheap spot compute layer: an open-source scheduler opens the door, then a marketplace brokers idle capacity at a discount.

$1.5M seed raised in 2025 (per PitchBook data cited by Startuply)

Brothers Ryan (AWS cloud and networking) and Lucas (Webflow data platforms) Ewing watched data scientists hunt for GPUs across clouds and on-prem fleets; one team they talked to paid $7M a year for guaranteed GPU capacity and still ran out. They founded Lilac in 2025 and joined YC's Summer 2025 batch with an open-source scheduler.

Lilac released its open-source scheduler v0.1.0 at launch and raised a $1.5M seed in 2025. In August 2025, GPU-as-a-service provider BluSky AI signed a letter of intent for Lilac to resell its idle B200/H200/H100/A100 capacity, naming Lilac a preferred marketplace partner. By June 2026 Lilac had a hosted inference API routing requests to idle enterprise clusters, with pay-per-token pricing from $5.

Still active as of September 2026: a four-person San Francisco team, with the open-source scheduler, marketplace, and inference API live, and the business evolved to selling flexible GPU contracts and running operations software for GPU cloud providers.

GPU scarcity is real but mismatched: clusters sit idle at low utilization while teams pay millions for guaranteed capacity, so brokering spare compute attacks a genuine inefficiency.

The self-hosted scheduler is a plausible wedge: it solves internal GPU-hunting pain before asking a company to expose its spare capacity to strangers.

The marketplace faces a chicken-and-egg problem - enterprises must trust Lilac with idle hardware - which is why the BluSky AI letter of intent was the early signal worth watching.

Differentiation is thin against Run:ai, Vast.ai and hyperscaler spot markets, so success hinges on liquidity and execution rather than technology alone.

An open-source wedge can open enterprise doors, but a two-sided market only works if supply actually shows up; a copied scheduler is no moat, so marketplace liquidity is the whole bet.

As of September 2026, Lilac is active as a YC Summer 2025 company (four-person team in San Francisco). Its open-source scheduler remains the entry point, the BluSky AI partnership is the anchor supply-side signal, and since June 2026 its inference API has been available on Eden AI with open-weight models such as Kimi, MiniMax, GLM and Gemma. The company says its next steps are wider model coverage, better routing, and growing the distributed GPU network.