Thomson Reuters built its own AI on Qwen for $40M, ditching Anthropic
SumitGup · x · 2026-08-29
Thomson Reuters spent about $40 million over two years building its own AI model, with the final training run costing just $450K. Starting from Alibaba's open-source Qwen, they trained on proprietary legal and news content from Westlaw, Practical Law and Reuters, aiming to reduce dependence on Anthropic. Their CTO compared renting outside AI to being a permanent tenant versus owning the building.
The quoted thread argues enterprise customers are broadly cutting OpenAI/Anthropic spending for cheaper alternatives; if Thomson Reuters can do this, any large company can — meaning the enterprise licensing model of big AI labs is largely a bust, putting their datacenter-funding cash flows (and that trillion-dollar IPO) at risk.
Related event: Thomson Reuters Spent $40M Building Its Own AI Model on Qwen(2 posts)→
More from Venture
- AI-driven job losses could turn $10T consumer debt into a $1T banking hole — churchkey · 2026-10-16
- I/O Fund: the 17-year bull market has room to run, but a ~20% correction comes first — Beth_Kindig · 2026-10-16
- TSMC Q3 revenue hits $46.7B, up 41% YoY; Q4 guidance of $52-54B tops estimates — Beth_Kindig · 2026-10-16
- BytePlus-backed AI Film Award offers $1M grand prize for films 70% made with Seedance — DavidmComfort · 2026-10-16
- dunkhippo33 on startup timing: most great ideas only exist because they weren't possible before — dunkhippo33 · 2026-10-16
- Inbox Zero launches open-source Superhuman rival with AI drafts at half the price — CurieuxExplorer · 2026-10-16