Ollama just stopped selling its premium plan because open models are eating its cloud alive

Ollama hit pause on new Max subscriptions Monday. The reason: open models like GLM-5.2 are burning through its cloud capacity faster than it can add GPUs. The company needs to hold some headroom for "very large models" due next week.

GLM-5.2, a recent open-weight release from Zhipu AI, pushed Ollama's infrastructure to its limits. The company says it's adding compute to keep existing Max subscribers happy, but new sign-ups will have to wait. The cheaper Pro tier is still open.

The pause underscores just how fast the open model ecosystem is growing. Ollama raised $88 million earlier this year to make running open models boringly easy, and its token volume has more than doubled month over month. That kind of growth reflects the broader hunger for open-weight models, a trend that has driven models like Gemma 4 to 300 million downloads (Gemma 4 downloads milestone).

This isn't just an Ollama problem. Across the industry, demand for open models is reshaping cloud economics. For instance, a single open-weight model like Kimi K3 triggered a global revaluation of AI stocks (Kimi K3's market impact). Meanwhile, platforms like Hugging Face are doubling down on infrastructure ease, not model performance, to capture the same wave (Hugging Face's infrastructure bet). And open-source coding platforms like OpenCode have hit $40M ARR by betting developers want choices, not lock-in (OpenCode's growth story).

Ollama kept specifics under wraps, but "very large models" suggests parameter counts north of 100 billion. The company is betting that open models will keep driving traffic, even if it means turning away paying customers now. The economics of serving such models remain tricky, as Microsoft's internal testing of open models shows potential for massive cost savings (Microsoft's 89% cost cut).