Thomson Reuters builds its own AI for $40 million

Thomson Reuters is spending $40 million to create its own language model called Thomson. Instead of relying on outside providers like OpenAI, the company built this system using Alibaba’s Qwen technology. It will take two years to fully develop.

The model works best when it accesses the company's private databases like Westlaw. This means the AI is specifically tuned to handle legal and professional data rather than general internet queries.

Company leaders believe owning their technology is a smarter long-term strategy than renting access to other models. They argue that control over their data and systems matters more than general intelligence.