Thomson Reuters, the news and legal-information giant, has built its own artificial intelligence model — and the stated reason is money.
According to Quartz, the company has launched an in-house AI model in order to cut what it spends with Anthropic, the AI lab behind the Claude assistant. Fortune, in a report circulated via Google News, describes the same move as Thomson Reuters loosening its grip on Claude, framing it as the company reducing its reliance on an outside AI supplier it had been leaning on.
The available reports are thin on specifics. Neither source summary gives figures for how much Thomson Reuters was paying Anthropic, how large or capable the new in-house model is, or whether Claude is being dropped entirely rather than used more sparingly. What both agree on is the direction of travel: build internally, buy less externally.
That direction is the interesting part. Most companies that have adopted generative AI over the past few years did so by renting it — paying a lab like Anthropic or OpenAI per unit of text processed. That is cheap to start and expensive at scale, and a company like Thomson Reuters, which runs enormous volumes of legal documents and news copy through software, sits squarely at the expensive end.
Building your own model is a large upfront bet that pays off only if you use AI heavily and predictably enough for the fixed cost to beat the metered one. Thomson Reuters apparently thinks it does.
It matters because if large, AI-hungry customers start replacing rented frontier models with homegrown ones for their bread-and-butter work, the assumption that AI labs will simply keep collecting rising fees from enterprise usage starts to look less certain.