Thomson Reuters builds in-house AI on Chinese tech to cut costs

What happened

Thomson Reuters developed an internal large language model, Thomson-1, based on a "realigned" version of Alibaba's Qwen3.5 to reduce reliance on Anthropic's Claude for AI features. The initiative is primarily cost-driven, as TRI invests heavily in AI while facing pressure on margins from competitive pricing and the need to scale AI across its product suite. The company clarified that Thomson-1 is not intended to replace Claude entirely, suggesting a multi-model strategy where different models serve different use cases based on performance, cost, and data sensitivity. This development occurs as TRI stock trades at $88.53, down 47% over the past year, with the market weighing AI disruption risks against TRI's recurring revenue and cash generation. The use of a Chinese-origin model raises potential regulatory and data security questions, particularly for legal and professional content, which could become a new risk factor if not managed carefully.

Implication

Building an internal model on Qwen3.5 could improve TRI's gross margins on AI-enabled products by lowering per-query costs, directly supporting the EBITDA margin expansion target of ~100bps in FY2026. However, the decision to base the model on Alibaba's technology raises concerns about data security, compliance with client confidentiality, and potential geopolitical backlash, especially in Western legal markets where TRI holds a dominant position. The clarification that Claude will still be used indicates TRI is hedging its bets, possibly because Qwen-based Thomson-1 may not yet match Claude's quality for high-value professional tasks, limiting immediate margin benefits. Investors should monitor future disclosures on model performance, adoption rates within TRI's products, and any client pushback or regulatory inquiries, as these will determine whether this strategic shift is value-accretive or a hidden liability. Overall, this news slightly increases uncertainty in the short term but could be a positive for long-term cost control if TRI manages the data and trust issues effectively.

Thesis delta

The core thesis remains unchanged: TRI's ability to monetize AI through paid attach and renewals is the key driver, and managing AI costs is part of that. However, this development introduces a new variable: the choice of a Chinese-origin model could complicate TRI's positioning as a trusted, secure platform for legal and professional data, potentially undermining customer retention. Therefore, the thesis risk profile has shifted slightly: while cost savings support margin expansion, the potential erosion of trust adds a negative offset that must be monitored.