Back
Meta explores $10 billion Anthropic AI compute lease as 2026 capex rises

Meta explores $10 billion Anthropic AI compute lease as 2026 capex rises

Preliminary Anthropic talks emerged as Meta disclosed surging AI infrastructure obligations, including $278.99 billion of unrecognized lease commitments tied mainly to data centers, while lifting the low end of 2026 capex guidance to $130 billion-$145 billion.

Fact Check
All three components of the claim are corroborated by authoritative primary reporting. The preliminary Meta-Anthropic $10 billion compute lease talks are confirmed by Reuters and CNBC (July 17, 2026). Meta's raised 2026 capex guidance to $130B-$145B (up from a $125B floor) is confirmed by CNBC's July 29 earnings coverage and the 'AI capacity dilemma' article. Meta's surging AI infrastructure lease obligations, primarily data centers, are documented in the Q2 2026 10-Q summary (StockTitan), with the specific $278.99B data-center lease figure matching the filing disclosures. The framing linking the Anthropic talks to disclosed capex/lease obligations is accurate.
Summary

Meta is in preliminary discussions to lease roughly $10 billion of AI computing power to Anthropic, a potential new revenue stream as it decides how much excess capacity to monetize versus reserve for its own AI models and services. The talks surfaced alongside Meta’s July 29 earnings report and quarterly filing, which showed the financial scale of its AI buildout: Q2 2026 capital expenditures reached $31.1 billion, free cash flow fell to $784 million from $8.55 billion a year earlier, and operating and finance leases not yet started or recognized on the balance sheet jumped about 53% in three months to $278.99 billion, mainly for AI data centers. Meta raised the low end of its 2026 capital expenditure range to $130 billion-$145 billion from $125 billion-$145 billion, stopped Q2 2026 stock buybacks, added $24.91 billion of long-term debt, and reported $349.31 billion of non-cancelable contractual commitments. Revenue of $60.8 billion beat the $60.2 billion consensus, while earnings per share of $6.18 missed the $7.14 expected, and the shares fell about 8%-10% in after-hours trading.

Terms & Concepts
  • AI computing power: Processing capacity used to train and run AI models.
  • capital expenditures: Spending on long-term assets such as servers and data centers.
  • free cash flow: Cash left after operating costs and capital spending.