Anthropic is signing one of the largest data-center leases ever written, and simultaneously getting sued — again — for allegedly training Claude on pirated books.
TeraWulf’s $19 billion, two-decade lease with Anthropic turns a crypto-mining also-ran into an AI landlord overnight, and it lands the same day Nscale lines up a $900 million credit facility to keep pace with GPU demand. Debt and decades-long leases are now standard financing tools for compute — which may tell you the industry has fully priced in that training frontier models is a capital-markets business first and a research project second. OpenAI made that explicit by hiring a compute financier to run capital markets ahead of IPO chatter.
Compute is getting Wall Street treatment. Data, meanwhile, is still getting treated like an afterthought.
That’s the problem exposed by the new $75 million suit against Anthropic over pirated book text. It’s the latest entry in a mounting docket testing how far last year’s fair-use wins actually stretch. Every dollar poured into Kentucky data centers is a bet that the underlying training corpus survives legal scrutiny — a bet that looks shakier with each new complaint. Building $19 billion of physical infrastructure while your data provenance is still being litigated is not risk management, it’s denial.
Contrast that with where the smarter capital is actually going: verticals with clean, defensible data moats. Norm’s $120 million Series C at a $1.2 billion valuation is a bet that narrow, high-stakes domains with licensable data beat scraped generality. Scale AI’s MoU with Singapore’s IMDA is the same instinct at government scale — evaluation-as-a-service with a sovereign stamp of legitimacy, which is worth more than another scrape of the open web. And Amazon quietly closing Mechanical Turk to new sign-ups is the tombstone for the pennies-per-task era that built those scraped corpora in the first place. The cheap-labor pipeline is dying right as the lawsuits over its output are multiplying. That is not a coincidence; it’s a market correcting.
On the enterprise side, the same logic holds: SAP buying Dremio is a bet that agentic AI needs governed, queryable data more than another model — exactly the discipline the frontier labs are skipping. Meanwhile Snowflake’s $80 billion narrative remains long on story, short on numbers — a reminder that “data strategy” headlines don’t always survive a hard look. And down at the other end of the market, Nomerra’s $2 million seed and dxFeed’s OPRA and CME expansion show the unglamorous, fully-licensed data plumbing still grinding along, unbothered by any of this drama — which is exactly why it’s investable.
The lesson of the day: infrastructure spend is not a substitute for clean data rights, and the bill for skipping that step is coming due in courtrooms, not data centers.
Watch tomorrow for whether Anthropic’s legal exposure gets any mention in how it frames the Kentucky buildout — silence would be its own kind of answer.
Stories covered
- dxFeed Adds OPRA, CME Futures Options Data to Overcharts (Data Markets)
- OpenAI Hires a Compute Financier as IPO Chatter Builds (AI Training Data)
- Legal AI Startup Norm Crosses Unicorn Line With $120M Khosla-Led Round (Deals & Funding)
- Nscale Lines Up $900M Credit Facility for AI Buildout (Deals & Funding)
- Scale AI Signs MoU With Singapore’s IMDA on AI Evaluation (AI Training Data)
- Snowflake’s $80B Data Bet: Thin on Details, Heavy on Narrative (Data Markets)
- SAP Buys Dremio to Feed Its Agentic AI Ambitions (Deals & Funding)
- Nomerra Bags $2M Seed to Untangle Private Markets’ Paper Trail (Deals & Funding)
- Anthropic’s $19bn Kentucky Bet Turns TeraWulf Into an AI Landlord (Deals & Funding)
- Anthropic Hit With New $75M Suit Over Pirated Books Used to Train Claude (Licensing & Legal)
- Amazon Quietly Shutters Mechanical Turk to New Sign-Ups (AI Training Data)
