Anthropic's legal plugin wipes off $300bn in value but investors miss the point (also, everything is a data business)
The release of Anthropic's new legal workflow tool battered data stocks, showing just how sensitive markets are to potential threats from AI - and how little they understand about data businesses
Ironically, this week’s newsletter was supposed to be about legal and patent intelligence provider Patsnap’s planned IPO. After the release of Anthropic’s new legal AI tool wiped off more than $300bn from software and data stocks, it seemed prudent to shift gears and share our view on the impact of AI on data businesses. Spoiler: we are long data (nothing against AI, of course).
More on that below.
Asymmetrix is no longer just a Substack newsletter; we are a full-fledged data and research company providing critical intelligence to stakeholders in the Data & Analytics industry. Our clients include private equity funds, investment banks / corporate finance houses, management consultants, equity analysts and more - all keen to stay up to date and informed on the Data & Analytics industry. If that sounds like you, feel free to reach out directly at asymmetrix@asymmetrixintelligence.com or via our website www.asymmetrixintelligence.com for a demo.
Its all about the data
Soon after ChatGPT was released in late 2022, an otherwise sophisticated investor asked: why do data providers have any value anymore? ChatGPT has it all.
The questioner laid bare not only his naivete but his fundamental misunderstanding of both how ChatGPT works (which only has access to public web data) and the value of proprietary datasets. This simplistic line of thinking seems to have repeated itself this week.
Earlier this week, Anthropic announced the release of a tool for its Claude Cowork assistant to assist legal professionals on tasks like document review, contract drafting and research. Software stocks, which have been feeling the heat for some time amid fears that AI will disrupt software companies’ competitive moats, took yet another hit on news of the Anthropic release. Data stocks - first those focused on legal data and workflow tools, like Thomson Reuters, RELX, and Wolters Kluwer, and then the broader information services market - took a hit as well.
Here is a look at Wednesday’s carnage across public data stocks:
One can argue that RELX, Thomson Reuters, and Wolters Kluwer are reasonably vulnerable to the impact of solutions like Anthropic’s new tool; all three focus on the legal industry and have product portfolios that include similar workflow solutions. In fact, Thomson Reuters and Anthropic have a partnership that integrates Claude with CoCounsel to power document reviews, legal contract analysis and legal research. If Anthropic can develop a competitive tool, then doubtless these providers are at risk of being replaced by lower cost solutions.
However, our view is that the market’s sensitivity and reaction to Anthropic’s release shows a fundamental lack of understanding of the value of proprietary data. Ultimately, the raw material that power quality LLMs, whether general answer engines or industry-specific AI tools, is data. Messy, publicly available data from government filings or web-scraping operations is surely sufficient for some workflows and some industries. Across most industries, however, businesses pay a premium for difficult-to-obtain, proprietary data that is more reliable, insightful, and useful in their day-to-day workflows.
And that is the difference between Anthropic’s shiny new legal tool and the Thomson Reuters’ CoCounsel tool. CoCounsel draws from Thomson Reuters’ comprehensive and highly regarded Westlaw legal database, which features proprietary data beyond public legal records, such as attorney-authored headnotes and its proprietary West Key Number System for topic classification. And Anthropic? They have formed an MCP connection with Midpage, a US legal research startup. To our knowledge, Thomson Reuters’ Westlaw is still regarded as industry leading; the jury is out on whether Midpage’s library is on par with Westlaw’s, truly proprietary, and reliable enough to support complex legal workflows. As our friends Andrew Steinerman and Alex Hess of JP Morgan equity research point out, there have been no signs that customers are defecting from Thomson Reuters, and particularly not from Westlaw.
To be sure, we at Asymmetrix certainly believe AI is impacting data businesses – we see it in our own business, but that’s a topic for another time – but it does not impact the value of a business if its core asset is proprietary data, even if that data is not accessed directly but via end point workflow solutions. In fact, we have found that the valuation of data providers with truly proprietary data holds up, if not increases, as public data becomes more easily accessible, commoditized and able to be productized.
The risk is on the margins – data companies relying on scraped public web data and doing little in the way of interpretation or transaction. Over the counter LLMs can easily find and analyze that same data or easily replicate datasets using a cluster of AI agents.
Research businesses like Gartner and Forrester, which have long been regarded as a cheaper alternative ($200,000/year) to hiring white shoe consulting firms ($2,000,000 engagement), are also at risk. Their abstract, high-level market research and strategic frameworks are increasingly thought of as generic and not deeply embedded in client workflows. Clever use of tools like OpenAI’s Deep Research can output research approximating Gartner’s research product – all at a fraction of the cost. It is no coincidence that Gartner’s and Forrester’s stocks have been battered for months, long before Anthropic’s legal tool came to market, with Gartner at ~1/3 of its market capitalization this time last year.
Sector-focused data businesses with truly proprietary content, be it data, research, numbers, or words, can deepen their moat and maintain pricing power, even in the shadow of ChatGPT, Claude, and Gemini. Building AI-powered endpoint solutions that draw from their datasets may be wise, but isn’t a necessity, so long as their information is available to clients in their preferred environment and it does not detract from the value of their own data or cannibalize their own business.
If LLMs want this data, they will have to pay for it.
As we predicted in our recent State of Data & Analytics 2026 report, we believe AI-native businesses are likely to acquire proprietary datasets to give them the edge to build more advanced products, enhanced by their AI models. We have seen the first signs of this, with OpenAI’s acquisition of healthcare startup Torch in January 2026, and anticipate more such acquisitions as AI giants recognize that proprietary data will unlock far superior products than publicly-trained AI solutions.
Alternatively, AI businesses may continue to license data and content. 2025 saw many licensing agreements signed, including OpenAI’s deals with Axel Springer, Financial Times, and News Corp, among others. Financial data providers like FactSet and LSEG have also signed deals with OpenAI, Anthropic and Perplexity. Licensing agreements like these come with lots of fine print and restrictions; owning the data assets and derivative products is a less risky path for AI companies.
At that point, the AI providers will have to put their hands in their pockets and pay a premium to acquire Data & Analytics businesses if they wish to create a true advantage against their competitors. And they will then become, in turn, Data & Analytics providers.
Everything is a data business.
📜 Interesting Content
Experian’s tech chief defends credit scores: ‘We’re not Palantir’ - The Verge