Anthropic Restricts OpenClaw Amid AI Monetization Crunch
SkimNews Take
AI labs are beginning to internalize the compute costs of their most advanced models, forcing a strategic shift towards more cost-effective AI agents even for popular applications.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic restricted third-party agent tool OpenClaw's access to Claude, with head of Claude Code Boris Cherny writing on X that "our subscriptions weren't built for the usage patterns of these third-party tools," forcing users to pay significantly more
- Gartner's Will Sommer projects roughly $6.3 trillion in capital investment in AI data centers from 2024–2029, requiring large providers to hit a 7% return on invested capital to avoid write-downs — a threshold he called an "unmitigated disaster" if missed
- To reach even that 7% minimum, Gartner forecasts large AI companies must earn a cumulative ~$7 trillion in AI-driven revenue through 2029, approaching $2 trillion per year by period's end, or ~$8.2 trillion for "historic returns"
- OpenAI has already committed $600 billion in spending through 2030 — described by Sommer as a "massive step down" from its prior $1.4 trillion plan — yet even best-case compound growth projections show it reaching only a fraction of the spend needed for 7% ROIC
- To generate the projected $2 trillion annually, providers would need to process a cumulative ~10 sextillion tokens per year (21 zeros), a 50,000–100,000x increase over current consumption even assuming a generous 10% per-token margin, per Sommer
- AI agents like OpenClaw powered by reasoning models are vastly more token-hungry than earlier chatbots — Georgia Tech's Mark Riedl noted agents may "talk out loud to itself for thousands and thousands of tokens" on a single user prompt, with inference now outweighing training costs at scale
- Sommer predicted market consolidation is "virtually inevitable," with no more than two large language model providers surviving in any regional market, while both Anthropic and OpenAI are reportedly racing to IPO by the end of 2026
Why it matters: Free or near-free AI is ending because labs must generate roughly $2 trillion in annual revenue by 2029 just to clear a bare-minimum 7% return for investors who have already committed $6.3 trillion to data-center buildouts. End users and developers — who can switch models at zero cost — will bear the squeeze through restricted access, higher tiers, and shrinking free plans, while Sommer's prediction of at most two surviving LLM providers per region means the current crowded market likely collapses before investors see their returns.
