Ollama Runs Models on Your Laptop. Its $65M Series B Is Really a Bet on the Cloud.

TL;DR

On July 9, Ollama, the free tool that 8.9 million developers use to run open-weight models on their own hardware, announced a $65 million Series B led by Theory Ventures, with Benchmark, 8VC, and Y Combinator also in. That brings total funding to $88 million for a 14-person company. The headline reads "local AI raises big." The part that matters for builders is quieter: the business the money is chasing is not the thing on your laptop. It is Ollama Cloud, and it bills for GPU time instead of per token.

What actually got funded

The numbers are real and the growth is not subtle. Ollama reports 8.9 million monthly active developers, more than 67,000 integrations, 176,000 GitHub stars, and usage inside 85% of the Fortune 500, including government, healthcare, and finance shops that cannot send prompts to someone else's API. The company was founded in 2023 by Jeff Morgan and Michael Chiang, who previously built Kitematic and sold it to Docker. This $65 million round follows a $15 million Series A led by Benchmark's Peter Fenton.

What they did not share: the valuation and the revenue. In venture terms, declining to state both at once is the polite way of saying the number is either very good or not yet a number.

Why 8.9 million is the number to stare at

Ollama's whole trick is turning "run this open-weight model" into a single command, instead of a weekend of compiling llama.cpp flags and hunting for the right quantization. Download, ollama run, done, on your own GPU, with your data never leaving the box.

That last part is why 85% of the Fortune 500 shows up in the stats. If you work somewhere that treats a customer record leaving the building as a reportable incident, a model that runs air-gapped is not a preference, it is the only option that clears legal review. In those shops, "just send it to OpenAI" is a sentence that ends meetings.

The reflex read is that local models are a hobbyist thing while the real work happens on frontier APIs. The adoption numbers say the opposite is quietly true across regulated industries, where the compliance problem, not the model quality, is what picks the runtime.

The business is a cloud

Here is the twist. A free local tool does not, on its own, justify $88 million. Ollama Cloud does. It runs the same models behind the same ollama run command and the same OpenAI-compatible API, except now on Ollama's GPUs, with paid tiers from free up to $100 a month.

The pricing is the tell. Where the incumbent inference clouds like Together AI, Fireworks AI, and Groq meter you per million tokens, Ollama Cloud bills for GPU time. Token pricing is a metered taxi: you pay for every mile of output the model writes. GPU-time pricing is renting the car by the hour, where you hold the keys, drive as far as you want, and pay for the time. For agent workloads that spew huge volumes of tokens in tight loops, which meter you sit under is not a rounding error.

The strategic bet is continuity. Prototype locally for free, then move the exact same code to the cloud when you outgrow your GPU, without swapping SDKs or rewriting a single call. That is a genuinely sticky wedge, and it is what $65 million buys: GPUs, plus the runway to make the local-to-cloud path frictionless before anyone else does.

The part worth watching

You have seen this movie. A beloved open-source tool takes a big round, and eighteen months later there is a "team plan," a telemetry prompt, and a feature that only works when you log in. Morgan's public framing is careful about exactly this: "Open models should be easy to run, easy to build with, and available wherever people need them: on your own machine, in the cloud, or both."

The healthy version of this story is that cloud revenue funds the free local tool the community actually loves. The other version is that the free tool becomes the funnel and the roadmap slowly tilts toward the part with a credit-card form. Nothing announced this week decides which one you get. But the incentive gradient just changed, so it is worth watching where the good features land first.

Key Takeaways

• The raise: Ollama closed a $65M Series B led by Theory Ventures, taking total funding to $88M, for a 14-person company founded in 2023.
• The scale is real: 8.9M monthly active developers, 67,000+ integrations, 176,000 GitHub stars, and usage inside 85% of the Fortune 500.
• Local wins where data cannot leave: regulated industries adopt Ollama because on-machine inference sidesteps the compliance problem of shipping prompts to an external API.
• The business is Ollama Cloud: same command, same OpenAI-compatible API, but billed on GPU time rather than per token, which favors token-heavy agent loops.
• Continuity is the moat: the pitch is prototype local, scale to cloud, no code changes, competing with Together, Fireworks, and Groq on switching cost rather than raw price.
• Watch the incentive drift: a VC-funded free tool with a paid cloud beside it is a familiar arc, so watch whether new features ship to local or cloud first.

Sources: TechCrunch, SiliconANGLE, Ollama blog, AIwire