Anthropic’s new compute story is not a reason to leave Claude or expect immediate throughput relief. It is a contract-review signal.

The signed part is specific: TeraWulf says Anthropic has a 20-year lease for about 401MW of critical IT capacity at its Justified Data Campus in Hawesville, Kentucky, with initial capacity expected in H2 2027, full ramp by early 2028, and about $19 billion in expected contracted lease revenue. The preliminary part is separate: CNBC reported that Anthropic is in very early talks to lease Meta compute, after The New York Times reported a possible arrangement worth roughly $10 billion over two years. Meta declined to comment; no deal has been signed.

Key takeaways

• Recommendation posture: ask sharper vendor questions. Do not migrate away from Claude because of this news.
• TeraWulf is signed, but the capacity is future-dated. It does not change Claude production limits this quarter.
• Meta remains reported preliminary talks only. Treat it as a conditional diligence item, not a current data-isolation risk.
• The useful action is to review DPA, subprocessor, residency, isolation, notice, and fallback language before infrastructure changes become urgent.

What is real, and what is not

TeraWulf’s SEC filing says its subsidiary Raylan Data LLC entered a 20-year lease with Anthropic PBC for high-performance-computing capacity in Kentucky. That is the hard fact. The filing and company release point to phased delivery beginning in late 2027 or H2 2027, not a near-term Claude capacity fix.

The Meta story is weaker evidence. CNBC described the talks as “very preliminary,” and The Next Web summarized the same reporting. There are no disclosed terms, no primary Anthropic statement, and no Meta confirmation. Operators should not treat Meta as an Anthropic infrastructure provider unless a deal is announced.

Why Claude customers should still care

Enterprise AI buyers increasingly depend on infrastructure they do not directly see. A Claude contract may feel like one vendor relationship, but the delivery chain can include clouds, data centers, logging systems, support tooling, and security vendors.

That does not mean TeraWulf or Meta are Anthropic subprocessors today. Anthropic’s DPA defines a subprocessor as an entity Anthropic engages to process Customer Personal Data. It says Anthropic gives reasonable notice before a new subprocessor gets access to Customer Personal Data, with a 15-day objection window. Anthropic’s public Trust Center subprocessor list, checked in the research handoff on July 19, listed Google Cloud Platform, Amazon Web Services, and Microsoft Azure as cloud-infrastructure subprocessors; it did not list TeraWulf or Meta in the visible public entries.

The operator question: if Anthropic adds new infrastructure paths before TeraWulf goes live, will your contract say what changed, where data may be processed, who can access logs or telemetry, and what rights you have if the answer matters?

Questions to send legal, security, or Anthropic

Ask for a current provider map: which clouds, data centers, support, logging, and telemetry providers can touch prompts, outputs, files, abuse-monitoring data, support metadata, or Customer Personal Data?

Confirm the subprocessor trigger. Does notice apply only when a provider processes Customer Personal Data, or also when it can access operational metadata, logs, or support records?

Check residency and isolation. Are workloads restricted to named regions or clouds? If a future provider also competes in AI models, what prevents provider personnel, tooling, logs, or metadata from crossing into product or model teams?

Plan fallback language. If a provider change affects latency, limits, residency, compliance approval, or customer commitments, can the workflow route to another approved model or region?

Operator call

Keep using Claude where it fits. Do not treat the TeraWulf lease as immediate capacity relief, and do not treat preliminary Meta talks as a current data event. The practical move is to add infrastructure-provider review to the next AI vendor-risk cycle: current subprocessors, notice and objection rights, residency commitments, isolation controls, security evidence, and fallback options.

What to watch next: an Anthropic Trust Center update, a signed Meta arrangement, or any Anthropic statement tying new capacity to Claude Enterprise/API limits, regions, pricing, or product availability. Until then, this is a useful contract checklist, not an operational emergency.