Lilac

What Lilac does

Lilac Research Inc. is a San Francisco-based two-sided marketplace for idle enterprise GPU capacity, founded in 2025 and backed by Y Combinator. On the supply side, GPU cluster operators install Lilac's Kubernetes-based GPU Operator (a controller with a 30-second reconciliation loop and LIFO preemption that completes in under 60 seconds) to monetize idle NVIDIA capacity that would otherwise sit unused; Lilac routes inference traffic into that capacity and shares 70% of gross token revenue with suppliers, retaining 30% as a marketplace fee. On the demand side, AI developers and engineering teams access frontier open-weight models — Kimi K2.6 (Moonshot AI), GLM 5.1 and 5.2 (Z.ai), Gemma 4 (Google), and MiniMax M2.7 and M3 — through an OpenAI-compatible REST API at api.getlilac.com, with pay-per-token usage-based pricing, no contracts or minimums, cache read discounts for repeated context, and subscription plans (Basic $10/mo, Pro $30/mo, Max $100/mo) that layer in live supply-linked discounts of up to 75%.

Beyond chat inference, Lilac sells per-second batch container compute on idle H100 ($1.00/hr) and H200 ($1.50/hr) capacity and brokers committed cluster reservations (H100 to B300) from neo-cloud partners, with enterprise customers reached via founder-led direct sales and monthly invoicing. The go-to-market is product-led and self-serve, accelerated by YC distribution, a Discord community, technical blog content on idle-GPU economics, and drop-in integration with OpenAI-compatible developer tools including OpenCode, Cursor, and Continue. As of early 2026 the company has roughly $3M in disclosed funding ($500K YC in May 2025 and a $2.5M round in October 2025) and operates with 11-50 employees, serving a long tail of AI startups (Osmosis, Understudy Labs, NanoGPT, Eden AI, Linum, Infron, Coral Bricks, Floot, Channel3) alongside model-provider partners and select enterprise teams.

Name

Lilac

Legal name

Lilac Research Inc.

Website

https://getlilac.com

Company type

Private

Founded year

2025

Operating status

Operating

Headcount range

11–50 employees

Short description

Lilac is a two-sided marketplace that routes AI inference onto idle enterprise GPUs via a Kubernetes operator, offering pay-per-token access to frontier open-weight models (Kimi K2.6, GLM 5.1/5.2, Gemma 4, MiniMax M2.7/M3) through an OpenAI-compatible API, with subscriptions, batch compute, and cluster reservations for AI developers and small teams.

Ownership category

akta.pro rank

Product category

AI Inference Cloud

akta.pro primary industry

Decentralized GPU/AI Compute Networks (training/inference, GPU renting, model hosting) (FSAPALAC)

akta.pro secondary industry

AI Compute Cloud & GPU-as-a-Service (HDAAAAAK)

Keywords

• GPU inference API
• Idle GPU network
• AI model serving
• GPU cluster reservation
• Batch GPU compute

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Operations, Marketing or Sales

Revenue model

1. Per-token Inference Revenue: Lilac charges customers per token for inference on open frontier models (Kimi K2.6, GLM 5.1, GLM 5.2, Gemma 4, MiniMax M2.7, MiniMax M3) routed through idle enterprise GPUs. Pricing is per million tokens with no contracts or minimums. This is the primary revenue stream from AI developers and teams using the API.
2. Subscription Plans (Personal): Monthly subscription plans (Basic $10/mo, Pro $30/mo, Max $100/mo) bundle dollar-denominated included model usage with live per-model discounts. Included usage does not roll over. Subscriptions are for individual, non-commercial use only. Commercial workloads must use prepaid credits or enterprise billing.
3. Batch / Container Jobs: Batch GPU compute jobs billed per second at $1.00/hr for H100s and $1.50/hr for H200s. Customers submit a container image and command; Lilac runs on spare GPU capacity. Used for training, fine-tuning, and batch processing workloads.
4. Supplier Revenue Share (30%): Lilac takes 30% of gross token revenue from inference processed on supplier GPU infrastructure. Suppliers keep 70%. Payouts are monthly via wire transfer or ACH. This is Lilac's take rate from the two-sided marketplace.
5. Cluster Reservations: For enterprise customers needing dedicated GPU capacity (H100, H200, B200, B300), Lilac brokers competitive quotes from neo-cloud partners and helps offset cost with idle-GPU software. Estimated pricing ~$2.00/hr for H100 on 1-month reservation.

Pricing tiers

Go-to-market motion2 records

Distribution channels4 records

Marketing channels6 records

Core offering

Lilac runs a two-sided idle GPU network: customers access pay-per-token inference on frontier open-weight models (Kimi K2.6, GLM 5.1, GLM 5.2, Gemma 4, MiniMax M2.7, MiniMax M3) via an OpenAI-compatible API, plus subscription plans, container batch compute, and brokered GPU cluster reservations. On the supply side, GPU operators install Lilac's Kubernetes operator to monetize idle GPU capacity, keeping 70% of token revenue while Lilac takes a 30% marketplace commission.

Product overview

Lilac is a unified idle GPU network platform offering four primary compute services: Inference (pay-per-token GPU inference on idle enterprise GPUs via OpenAI-compatible API), Subscriptions (monthly plans with up to 12x value through live discounts), Batch (container job processing on spare GPU capacity), and Clusters (dedicated GPU reservation with idle-GPU cost offset). The platform operates a two-sided market: customers access affordable frontier AI model inference while GPU operators monetize idle capacity through the Lilac GPU Operator (a Kubernetes controller) and GPU Pool configuration system. Lilac supports multiple frontier models including Kimi K2.6, GLM 5.1, GLM 5.2, Gemma 4, MiniMax M2.7, and MiniMax M3, accessible via OpenAI-compatible API endpoints.

Differentiator

Problem solved

Functional benefit

Products and services

• Inference (GPU Inference API) Pay-per-token GPU inference on idle enterprise GPUs. Runs frontier open-weight models (Kimi K2.6, GLM 5.1, GLM 5.2, Gemma 4, MiniMax M2.7, MiniMax M3) via an OpenAI-compatible API with no contracts or minimums.
• Subscriptions Monthly subscription plans (Basic $10/mo, Pro $30/mo, Max $100/mo) that bundle dollar-denominated model usage with live per-model discounts up to 75% off, delivering up to 12x value vs prepaid credits. Individual, non-commercial use.
• Batch Container job processing on spare GPU capacity. Customers submit a container image and command; Lilac runs it on spot-style idle GPU compute at $1/hr H100s and $1.50/hr H200s, priced per second. Used for training, fine-tuning, and batch processing.
• Clusters GPU cluster reservation service. Lilac sources competitive quotes from neo-cloud partners for dedicated GPU clusters (H100, H200, B200, B300) and helps offset cost with idle-GPU software. ~$2.00/hr for H100 on 1-month reservation.
• GPU Supplier Network Network for GPU operators to monetize idle GPU capacity. Suppliers install the Lilac Kubernetes operator to automatically pick up inference traffic when GPUs are idle, earning 70% of gross token revenue (Lilac keeps 30%).
• Lilac GPU Operator Kubernetes controller deployed on supplier clusters (Helm chart, v0.3.5). Discovers idle GPUs, communicates with Lilac's control plane, and manages inference workload pods using a customized vLLM fork. Runs a 30-second reconciliation sync loop for capacity calculation and workload scheduling, with LIFO preemption and sub-60-second graceful shutdown.

Quantifiable outcome

• 30% cheaper tokens vs typical on-demand GPU inference rates
• +5 more outcomes

Named customers13 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration3 records

AI capability7 records

Feature6 records

Partnerships

Six partnerships are on record, tiered flagship and core.

• MiniMaxflagshipStrategic or Co-development Partner · 7 May 2026Lilac partnered with MiniMax to bring commercially licensed MiniMax M2.7 to Lilac's network. MiniMax provides open-weight models with commercial licensing; Lilac handles the hosted inference endpoint and routing onto GPU capacity. The partnership gives customers a clear commercial path to use the model without managing deployments.
• Moonshot AIcoreTechnology or IntegrationMoonshot AI's Kimi K2.6 model (1T total parameters, 32B activated, MoE architecture) is served on Lilac's idle GPU network. Kimi K2.6 supports 262K context, multimodal input, and reasoning. Lilac routes inference traffic to capacity providers running the model.
• Z.ai (Zhipu AI)coreTechnology or IntegrationZ.ai's GLM 5.1 (754B parameters, 202.8K context) and GLM 5.2 (524K context) are served on Lilac's idle GPU network. GLM 5.1 is MIT licensed. GLM 5.2 supports configurable reasoning effort and tool calling for agentic workflows.
• AWS (Amazon Web Services)coreTechnology or IntegrationAWS is used as cloud infrastructure for Lilac's operator Helm chart (hosted on AWS ECR public registry). Capacity Providers may run on AWS infrastructure. Subprocessor for cloud hosting.
• vLLMcoreTechnology or IntegrationLilac's inference pods run a customized fork of vLLM, a high-performance inference engine. vLLM handles model serving, batching, and GPU memory management. Lilac's fork is tuned for idle-GPU scheduling and shared warm endpoints.
• StripecoreTechnology or IntegrationStripe handles all payment processing for Lilac's credit-based billing system. Subscriptions, auto top-up, and enterprise invoicing all run through Stripe. Listed as subprocessors for payment processing.

Scale indicators8 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• Together AI: Together AI provides a GPU cloud and inference API for open-source and open-weight AI models, serving frontier LLMs with per-token pricing. Direct competitor in the AI inference cloud market with overlapping customer base of AI developers and engineering teams.
• Fireworks AI: Fireworks AI offers a low-latency inference cloud for open-weight LLMs and custom models, with per-token pricing and an OpenAI-compatible API. Closely comparable business model to Lilac's inference product targeting similar developer and enterprise customers.
• Vast.ai: Vast.ai operates a decentralized marketplace where GPU operators rent idle capacity to AI customers. Most structurally similar peer to Lilac—same two-sided idle-GPU marketplace model, though Vast.ai is more consumer/developer oriented and less focused on frontier LLM inference.
• RunPod: RunPod provides GPU cloud services including on-demand and spot GPU rentals for AI training and inference. Overlaps with Lilac's batch and inference products, serving similar AI developer and researcher customers.
• Anyscale: Anyscale offers a compute platform for AI workloads built on Ray, including managed inference and training on GPU clusters. Competes with Lilac for enterprise AI team workloads and provides similar OpenAI-compatible inference endpoints.
• Modal Labs: Modal provides serverless compute for AI workloads, including GPU-backed batch jobs and inference endpoints. Direct overlap with Lilac's batch and inference products with developer-first PLG motion.
• Replicate: Replicate runs an API platform for hosting and serving open-source AI models on GPU infrastructure, with pay-per-prediction billing. Closely comparable developer-first inference cloud targeting similar AI application builders.

Broad incumbents

• CoreWeave: CoreWeave is a large-scale GPU cloud provider specializing in AI workloads with dedicated NVIDIA GPU capacity. Competes with Lilac's cluster reservation offering and broader inference market as a well-capitalized incumbent with completed enterprise certifications.
• Lambda Labs: Lambda provides GPU cloud services and AI inference APIs, with on-demand and reserved capacity for training and inference. Larger, more established peer competing in the same GPU inference and cluster markets as Lilac.

Emerging players

• io.net: io.net operates a decentralized GPU network for AI compute, aggregating idle GPUs from data centers and crypto miners to serve AI training and inference workloads. Comparable to Lilac's idle GPU aggregation model with a blockchain/crypto-economic layer.

Market position

Strengths5 records

Weaknesses4 records

Competitive moat5 records

Key risks5 records

Key highlights6 records

Customer concentration

• Xx.com/LilacML
• LinkedInlinkedin.com/company/get-lilac

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Profiles2 records

Funding overview

Funding rounds3 records

Investors3 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Lilac