EN
Back to the archive

The archive · AI & Models · Strategic decision · 2023–2026

Parasail bets AI inference clouds beat chip owners: $32M Series A, 500B tokens/day

Ex-Groq executive Mike Henry's Parasail rents global GPUs for cheap AI inference; after stealth launch it raised a $32M Series A on 500 billion daily tokens.

Parasail

The betInference, not training, will make the next compute giant: developers running open models and agents want cheap tokens, and renting GPU supply can beat owning chips.Scaling

What the business is

Parasail is an AI inference cloud: it rents GPU capacity across 40 data centers in 15 countries, orchestrates workloads to cut cost and latency, and sells production AI endpoints to developers with no training workloads and no long-term contracts.

How it started

Mike Henry, a former Groq executive who earlier founded AI chip startup Mythic, started Parasail in 2023 with Tim Harris and began building in early 2024. From building Groq's cloud offering, Henry recognized that developers building software on generative AI would want cloud processing specialized to inference; the company launched publicly in April 2025.

What happened

Parasail mainly rents processing time at 40 data centers in 15 countries and buys more from liquidity markets, orchestrating workloads behind the scenes to drive down the cost of inference requests while running some of its own GPUs. In April 2026 it raised a $32M Series A co-led by Touring Capital and Kindred Ventures, with participation from Samsung NEXT, Flume Ventures and Banyan Ventures, bringing total funding to $42M to expand its AI Supercloud for deploying and scaling AI agents. Touring Capital's Samir Kumar told TechCrunch he expects inference to be at least 20% of the cost of building software in the future.

No ending yet — it is still running.

Background

Parasail is a San Francisco AI infrastructure company founded in 2023 by Mike Henry — former Groq executive and earlier founder of AI chip startup Mythic — with Tim Harris. It provides cloud computing for companies running AI models for inference: it rents GPU processing time across 40 data centers in 15 countries, buys more from liquidity markets, and orchestrates workloads to drive down the cost of inference requests.

The company launched publicly in April 2025 and says it now generates 500 billion tokens a day, with customers including Elicit, mem0, Gravity, Kotoba and Venice and 30% month-over-month revenue growth. In April 2026 it raised a $32M Series A co-led by Touring Capital and Kindred Ventures, with Samsung NEXT, Flume Ventures and Banyan Ventures participating, bringing total funding to $42M.

The bet is that the proliferation of open models and agents outside frontier labs will make cheap inference a giant market: Touring Capital's Samir Kumar told TechCrunch inference could become at least 20% of the cost of building software. Parasail differentiates by refusing training workloads and long-term contracts, competing with Fireworks AI, Baseten and the large clouds — but its customer base is concentrated among seed and Series B startups in an unpredictable AI sector.

What has to be true

  • 500 billion tokens a day and 30% month-over-month revenue growth give the thesis a measurable base.
  • Henry built Groq's cloud before founding Parasail, so the inference focus came from direct customer demand.
  • Renting 40 data centers in 15 countries instead of owning chips keeps capacity elastic and costs variable.
  • Analysts and investors expect agents and open models to make inference a large share of software cost.

What can be applied

Asset-light orchestration can beat incumbents locked into their own silicon and contracts — but only if the demand wave it bets on, open-model and agent inference, actually materializes at scale.

Aftermath

As of September 2026, Parasail is live and scaling its AI Supercloud: it processes over 500 billion tokens a day, reports 30% month-over-month revenue growth since its April 2025 launch, and lists Elicit, mem0, Gravity, Kotoba and Venice as customers. The Series A funds deeper orchestration and inference optimization, go-to-market efforts and GPU and data-center partnerships. Its bet remains concentrated — most customers are early-stage AI startups — and it competes with Fireworks AI, Baseten and the bigger clouds, with the risk that its demand wave does not grow as fast as expected.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases