Company profile

Parasail

Pay-per-token inference cloud for AI-native startups, brokering GPU capacity across ~40 data centers plus its own fleet.

Researched 2026-11-05WebsiteLinkedInXDocs

Scope: this is Parasail Inc. of San Mateo, CA (parasail.io), not the older parasail.com or unrelated "Parasail" brands. LinkedIn headcount pages, Glassdoor and SEC EDGAR could not be read from this session (the LinkedIn script needs the logged-in Mac browser; EDGAR and Glassdoor were blocked), so headcount rests on Built In and job ads. No Glassdoor reviews turned up in search.

Snapshot

Parasail sells open-model inference (Kimi, GLM, DeepSeek, MiniMax, gpt-oss and 40+ others in its library) through serverless per-token endpoints, "elastic" priority endpoints, dedicated reserved GPUs and a discounted batch tier. Its pitch is "commit to spend, not GPUs": it aggregates capacity from about 40 third-party data centers in 15 countries and spot "liquidity markets" alongside its own Hopper and Blackwell fleet, then routes and tunes workloads automatically. Customers are mostly seed-to-Series-B AI startups (Elicit, mem0, Venice, Gravity, Kotoba, Reflow). It is a ~2½-year-old private company that has raised about $42M, with roughly two dozen employees as of the last public count, and it competes with Together AI, Fireworks AI and Baseten.

Mission, purpose, vision, values

No formal mission or values list is published. The closest statements, verbatim:

History

Ownership and status

Private, venture-backed. Legal name reported as "Parasail Inc." (SiliconANGLE); jurisdiction not confirmed (EDGAR not reachable from this session, so no Form D check). HQ 4 W 4th Ave, San Mateo, CA 94402 per Built In; press releases are datelined San Francisco. No parent company. A job posting says it is a federal contractor (VEVRAA/Section 503 forms).

Funding

DateRoundAmountLead / investorsValuationSource
2025-04-02 (announced)Seed$10MBasis Set Ventures, Threshold Ventures, Buckley Ventures, Black Opal VenturesNot disclosedBusiness Wire
2026-04-15Series A$32MCo-led by Touring Capital (Samir Kumar) and Kindred Ventures (Steve Jang); Samsung NEXT, Flume Ventures, Banyan Ventures, existing investorsNot disclosedPR Newswire, TechCrunch

Financials

People and headcount

SourceEmployeesDate
Built In company profile23Undated; read 2026-11-05 (likely stale)
Head of DevRel job ad ("our first dedicated developer-facing hire")23 (Built In box)Removed 2026-06-24
LinkedInNot read—

Engineering

Culture and reviews

Recent news (last 12 months)

Signals for a candidate

Green flags

  • Riding the biggest demand curve in AI infra: inference volume claimed to grow from billions to trillions of tokens/day in 18 months.
  • Experienced founder: Henry has built an AI chip company (Mythic) and ran product and cloud at Groq.
  • Capital-light model (brokered capacity, no long-term GPU commitments) limits the stranded-hardware risk that sank some GPU clouds.
  • Early enough (~25–45 people) for broad ownership and real equity upside; direct access to the CEO.
  • Serious systems work: kernels, scheduling, heterogeneous hardware, multi-region reliability.

Red flags and risks

  • Crowded, price-driven market (Together AI, Fireworks, Baseten, DeepInfra, plus hyperscalers and model labs selling their own APIs); token prices keep falling.
  • Small war chest ($42M) for a business that is now also running its own Hopper/Blackwell fleet and hiring data center ops.
  • Positioning has shifted three times in 18 months ("AI Deployment Network" → "AI Supercloud" for agents and training → "Inference Cloud for AI-native startups").
  • Customer concentration risk with startup customers who churn or get acquired; several homepage logos (Z, MiniMax, Kimi) look like model partners rather than paying customers.
  • No public CTO/VP Eng; co-founder Tim Harris also runs Swift Navigation. Metrics (tokens/day, 30% MoM) are unaudited company claims.

Questions to ask

  1. What share of tokens runs on owned GPUs vs. brokered capacity, and what are gross margins on each?
  2. Revenue run-rate and growth since April; runway at current burn; plans for the next raise or debt for hardware?
  3. Who leads engineering, how is the team organized, and how big is it today?
  4. Top-10 customer concentration, and how much revenue comes from model labs vs. app companies?
  5. How does on-call and reliability work across third-party data centers you don't control?
  6. What's the hybrid expectation in San Mateo, and how is the Europe team set up?

Sources