{"id":"c0f955ba-d19a-4d93-9a31-ba9cc37e7668","name":"Vast AI","slug":"vast-ai","description":"Vast AI runs a two-sided GPU rental marketplace that aggregates compute from independent hosts and small colocation operators, then rents it to researchers, startups, and inference workloads by the second. Unlike hyperscale neoclouds such as CoreWeave or Lambda that own or master-lease their datacenters, Vast is asset-light: the company is the exchange, the hosts own the hardware. That posture puts consumer GPUs (RTX 4090, RTX 3090) alongside datacenter parts (H100, H200, A100) on the same order book, with prices set dynamically by supply and demand across roughly 40-plus datacenters worldwide.\n\nThe company was founded in 2016 in Los Angeles by Jake Cannell and Christian Horne on the thesis that compute would become the bottleneck for independent AI research, and that a spot-style market would clear far below hyperscaler list prices. Vast layers three tiers on top of the raw marketplace: an interruptible bid market for batch training, a reserved tier with 1, 3, or 6 month terms, and a \"Secure Cloud\" that pins jobs to audited datacenters for customers who need compliance guarantees. A serverless inference product sits on top of the same pool.\n\nVast has stayed small and largely bootstrapped relative to the neocloud pack: roughly 40 employees, no disclosed venture round, and an inventory that is federated across hosts rather than centrally purchased. The tradeoff is that Vast does not compete for hyperscaler anchor contracts or multi-hundred-megawatt buildouts; the customer base is the long tail of researchers, indie developers, and price-sensitive inference workloads that hyperscalers underserve.","status":"active","founded":2016,"hq":"Los Angeles, California, USA","hqLat":null,"hqLng":null,"hqGeocodedAt":null,"hqAddress":null,"hqPrecision":null,"website":"https://vast.ai","fundingTotal":null,"type":"infrastructure","roles":[],"reviewStatus":"reviewed","heatIndex":0,"lifecycleState":"active","supersededByCompanyId":null,"sources":[{"url":"https://vast.ai/about","title":"About Vast.ai","publisher":"Vast.ai"},{"url":"https://vast.ai/","title":"Vast.ai homepage","publisher":"Vast.ai"},{"url":"https://vast.ai/pricing","title":"Vast.ai Pricing","publisher":"Vast.ai"},{"url":"https://docs.vast.ai","title":"Vast.ai Documentation","publisher":"Vast.ai"}],"keyFacts":[{"label":"Business model","value":"Two-sided GPU rental marketplace: independent hosts and small datacenters list capacity, renters bid or pay on-demand","sourceUrl":"https://vast.ai/about"},{"label":"Founded","value":"June 2016 by Jake Cannell and Christian Horne","sourceUrl":"https://vast.ai/about"},{"label":"Headquarters","value":"Los Angeles, California (relocated 2024); prior San Francisco office","sourceUrl":"https://vast.ai/about"},{"label":"GPU inventory","value":"~17,000+ GPUs across 350+ independent hosts (company disclosure, 2024); exact H100/H200/B200 mix not published","sourceUrl":"https://vast.ai/about"},{"label":"Datacenter footprint","value":"~40+ datacenters worldwide, all owned by third-party hosts; Vast owns no facilities","sourceUrl":"https://vast.ai/pricing"},{"label":"Pricing tiers","value":"On-demand (per-second billing), Reserved (up to 50% off, 1/3/6 month terms), Interruptible bid market (50%+ cheaper); exact per-GPU rates dynamic, not published statically","sourceUrl":"https://vast.ai/pricing"},{"label":"GPU generation mix","value":"Mixed datacenter and consumer parts: H100, H200, A100 alongside RTX 4090 and RTX 3090; consumer GPU coverage is a Vast differentiator","sourceUrl":"https://vast.ai/pricing"},{"label":"Product tiers","value":"GPU Cloud (marketplace), Clusters (multi-node), Secure Cloud (audited datacenters), Serverless inference","sourceUrl":"https://vast.ai/about"},{"label":"Positioning","value":"Long-tail researcher and inference customers; explicitly not competing for hyperscaler anchor tenants or hundred-megawatt buildouts","sourceUrl":"https://vast.ai/about"},{"label":"Employees","value":"~40 as of 2025 (company About page)","sourceUrl":"https://vast.ai/about"},{"label":"Funding","value":"No venture round publicly disclosed; appears bootstrapped","sourceUrl":"https://vast.ai/about"},{"label":"Public/private","value":"Private, US-incorporated","sourceUrl":"https://vast.ai/about"}],"aliases":[],"collisionRisk":"low","reviewNote":null,"atsProvider":null,"atsSlug":null,"factoryLocations":null,"manufacturingCapacity":null,"orgType":null,"burnSignal":null,"litigationEvents":null,"insurancePostureDisclosed":null,"unitEconomicsDisclosed":null,"tickerSymbol":null,"stockExchange":null,"secCik":null,"revenueUsd":null,"revenueBasis":null,"revenueAsOf":null,"employeeCount":40,"employeeCountBasis":"Company About page (self-reported, approximate)","employeeCountAsOf":"2025-01-01T00:00:00.000Z","marketCapUsd":null,"marketCapAsOf":null,"marketCapSource":null,"createdAt":"2026-08-28T01:48:21.938Z","updatedAt":"2026-08-30T03:38:44.882Z","jsonLd":{"@context":"https://schema.org","@type":"Organization","@id":"https://registry.deploy.report/companies/vast-ai","url":"https://registry.deploy.report/companies/vast-ai","name":"Vast AI","description":"Vast AI runs a two-sided GPU rental marketplace that aggregates compute from independent hosts and small colocation operators, then rents it to researchers, startups, and inference workloads by the second. Unlike hyperscale neoclouds such as CoreWeave or Lambda that own or master-lease their datacenters, Vast is asset-light: the company is the exchange, the hosts own the hardware. That posture puts consumer GPUs (RTX 4090, RTX 3090) alongside datacenter parts (H100, H200, A100) on the same order book, with prices set dynamically by supply and demand across roughly 40-plus datacenters worldwide.\n\nThe company was founded in 2016 in Los Angeles by Jake Cannell and Christian Horne on the thesis that compute would become the bottleneck for independent AI research, and that a spot-style market would clear far below hyperscaler list prices. Vast layers three tiers on top of the raw marketplace: an interruptible bid market for batch training, a reserved tier with 1, 3, or 6 month terms, and a \"Secure Cloud\" that pins jobs to audited datacenters for customers who need compliance guarantees. A serverless inference product sits on top of the same pool.\n\nVast has stayed small and largely bootstrapped relative to the neocloud pack: roughly 40 employees, no disclosed venture round, and an inventory that is federated across hosts rather than centrally purchased. The tradeoff is that Vast does not compete for hyperscaler anchor contracts or multi-hundred-megawatt buildouts; the customer base is the long tail of researchers, indie developers, and price-sensitive inference workloads that hyperscalers underserve.","identifier":"c0f955ba-d19a-4d93-9a31-ba9cc37e7668","foundingDate":"2016","address":"Los Angeles, California, USA","publisher":{"@id":"https://deploy.report/#organization"}},"framework_metadata":{"framework_schema_version":"0.1.0","verification_status":"verified","maturity_stage":null,"lifecycle_state":null,"architectural_position":{"cohort":null,"sub_cohorts":[]},"within_cohort_verified_vs_claimed_pair":null,"cap_flags":[],"verification_depth":{"sources_count":4,"primary_source_types":[]}}}