Open-Source AI Models Process 80-90% of Enterprise Tokens

What happened

Jeffrey Morgan, CEO of Ollama, has indicated a significant shift in enterprise AI adoption, with open-source models now processing between 80% and 90% of all AI tokens used by businesses. This suggests a strong preference among organisations for deploying open-source AI solutions over proprietary alternatives for a substantial portion of their operational needs. The data points to a growing trend where companies are leveraging the flexibility and accessibility of open models for their artificial intelligence workloads.

Why it matters

This development is critical for AI governance and ethics, as the widespread adoption of open-source models introduces both opportunities and challenges. While open models can foster transparency, allowing organisations to inspect and understand their underlying mechanisms, they also necessitate robust internal governance frameworks to manage potential risks. Businesses must establish clear policies for model selection, customisation, and deployment, ensuring that these accessible tools are used ethically and responsibly. For AI practitioners, it underscores the importance of expertise in adapting and securing open-source technologies, moving beyond a sole reliance on black-box proprietary systems. Policymakers, too, must consider the implications for intellectual property, data privacy, and accountability as open-source AI becomes increasingly embedded in critical business functions.

The Institute take

The increasing dominance of open-source models in enterprise AI token processing is often framed as a win for democratisation and cost-efficiency. However, this perspective overlooks the heightened governance responsibilities it places squarely on the shoulders of organisations.

While open-source offers unparalleled transparency and customisation, it simultaneously removes the implicit governance assurances that come with proprietary, vendor-managed solutions. Responsible AI leaders must recognise that "open" does not equate to "governance-free." Instead, it demands a proactive investment in internal expertise, robust model validation pipelines, and clear accountability frameworks to manage the expanded surface area for ethical and operational risks. The real work begins when the model is open.