Thomson Reuters bets $40M on owning its AI instead of renting from OpenAI or Anthropic

What happened

Thomson Reuters is investing about $40 million over two years to develop its own AI language model called Thomson. This model is built on Alibaba’s Qwen technology but is focused on integrating the company’s proprietary content, especially its legal research platform Westlaw. The initiative marks a shift away from renting AI capabilities from giants like OpenAI or Anthropic.

Why it matters

Owning rather than renting AI gives Thomson Reuters tighter control over both data security and intellectual property in its specialized domain. The company’s CTO, Joel Hron, emphasizes the importance of owning only the intelligence that a business critically needs. This stance challenges the common industry approach of relying heavily on third-party AI providers, which can raise costs, dependency risks, and limit customization.

Thomson’s model outperforms in benchmarks primarily when it can draw on the company’s own data. That suggests the competitive advantage for enterprises in AI comes less from raw model intelligence and more from exclusive domain expertise embedded in the model. For businesses handling sensitive or proprietary content, owning a custom AI stack could boost accuracy, compliance, and differentiation.

What to watch next

Expect other companies with rich, unique data to weigh the tradeoffs between owning AI models tailored to their content and using rented, general-purpose ones. The $40 million price tag sets a baseline for investments needed to create proprietary AI with meaningful integration.

It will be key to see how Thomson Reuters handles model updates, cost-efficiency, and the evolving quality gap against large external models. Operators should track whether owning AI becomes a wider trend for critical enterprise workflows or remains practical mainly for high-value, data-differentiated incumbents.

AI Quick Briefs Editorial Desk