Enterprise AI: Own or Rent?

I read this news about Thomson Reuters spending $40 million over two years to build its own AI model.

Thomson Reuters provides tools that lawyers and tax professionals worldwide use to research case law, statutes and compliance material, alongside Reuters news.

The company built a model, internally called Thomson, on top of Alibaba’s open-source Qwen. It worked with Imperial College on safety and neutrality, then added its domain expertise through training on legal, tax and news content, including Checkpoint and Reuters material.

Its CTO, Joel Hron, described the thinking behind the investment:

“You’re building equity in something you own for the long term, and that compounds, instead of paying rent on someone else’s model forever.”

That word, rent, is basically what Palantir CEO Alex Karp was talking about on CNBC a few weeks earlier.

He argued that companies pay for tokens that create no value while risking the knowledge and competitive advantage embedded in their businesses.

On Palantir’s Q2 earnings call, he went further, raising the concern that enterprises may pay to let their intellectual property and know-how migrate into someone else’s model.

Satya Nadella echoed a related worry days later in a blog post, describing a reverse information paradox.

The concern is that you could pay for intelligence twice: once with money and again with something more valuable, the proprietary knowledge revealed through prompts, corrections and workflows.

Of course, the extent to which any provider retains or trains on enterprise data depends on its contracts, settings and technical controls. The broader question of who owns the knowledge and value created through AI use still matters.

Quite a few other companies we know are thinking in this direction.

This may be the real shape of where enterprise AI eventually goes: not a mass exit from frontier labs, but enterprises asking harder questions about their own data and considering models they can control.

Owning your own model beats renting one, even when renting is cheaper and faster.

References

• Business Insider: Thomson Reuters builds AI model to rely less on Anthropic
• Satya Nadella: Article on AI and information
• Palantir: Alex Karp on what customers pay for