Ollama makes the most popular free tool for running open AI models (such as Meta's, NVIDIA's or China's Kimi and DeepSeek) on your own computer, and since 2025 has sold paid cloud plans that run larger open models on rented NVIDIA hardware. That makes it a key gateway where millions of developers pick which open model to use; the company says 8.9 million developers use it each month and that it shows up inside 85% of the Fortune 500, though neither figure is independently verified. It has raised $88 million with just 14 staff and has disclosed no revenue or valuation. The key risks are whether a free, open-source tool can be turned into a paying cloud business, repeated security problems with Ollama servers left open to the internet, and friction with parts of the open-source community it depends on.
Timeline
Oct 13
Partners with cloud host Fly.io so AI agents running in Fly.io's sandboxed virtual machines can call Ollama's cloud models without handling API keys directly.
Aug 31
Switches its Pro ($20/month) and Max ($100/month) cloud plans from billing by GPU time to per-token pricing with monthly usage credits and opens a $500-a-month Team plan for businesses, saying GPU-time bills had become hard to predict as open models grew larger; the change came weeks after it temporarily paused new Max sign-ups because demand was outrunning capacity.
Jul 9
Raises a $65 million Series B led by Theory Ventures, bringing total funding to $88 million; with only 14 employees, it says 8.9 million developers use it monthly but discloses no revenue or valuation.
May 5
Security firm Cyera discloses "Bleeding Llama" (CVE-2026-7482), a critical flaw letting outsiders pull prompts and secrets from the memory of unprotected Ollama servers, and estimates about 300,000 internet-facing servers are exposed; Ollama had shipped a fix in February, about three weeks after Cyera reported it.
Apr 15
A widely shared essay, "Friends Don't Let Friends Use Ollama," and a popular "Stop using Ollama" thread on Reddit's main local-AI forum revive complaints about slow credit to the open-source llama.cpp project it was built on, lock-in and slower performance than rivals.
Jan 29
SentinelOne and Censys researchers count about 175,000 Ollama servers publicly exposed across 130 countries; a separate report the same week described criminals hijacking exposed AI servers and reselling access to them.
Jan 16
Adds compatibility with Anthropic's API so Claude Code can run on open models, a day after adding OpenAI's Codex; the CEO later said January, when coding agents such as OpenClaw took off, was the proof point for the business.
