Anthropic Says It Has Crossed Its AI Research Threshold Based On Future Models

A missed Lambda lease deadline also left an approximately $34.9 billion computing agreement open to termination by either party.

Share
Anthropic Says It Has Crossed Its AI Research Threshold Based On Future Models

Anthropic, which develops Claude’s AI models and work applications, determined that it has met its threshold for accelerated AI research and development, replacing an earlier assessment that it might soon reach it. The company said in Tuesday’s amended IPO filing that the decision rests on expectations for future models and its development pace, rather than capabilities demonstrated by current models.

Claude Helps Build Its Successors

Anthropic’s Responsible Scaling Policy allows it to make the determination from either observed or expected acceleration. The distinction matters: the company has changed its assessment of the development trajectory, without presenting that decision as a finding about what its current models can do.

Claude already participates in developing subsequent generations. Anthropic uses it to run experiments, prepare training data, write and review software, and automate evaluations and safety audits. The company said those uses have accelerated research and that its internal estimates point to a further increase in capability gains.

Safety has already affected that work. Anthropic previously reported that it delayed initial training of a new model for more than a month while strengthening protections against theft of model weights—the parameters that encode a trained model’s behavior. It restricted the initial April release of Mythos to selected organizations through Project Glasswing because of its advanced cybersecurity capabilities.

Anthropic warned that further safety restrictions could leave its development moving more slowly than competitors’ and hurt revenue growth. It plans additional reports as it studies the implications of crossing the research threshold.

Lambda Contract Can Now Be Terminated

Anthropic also reported that Lambda missed a deadline to secure a facility lease. That failure gives either company the right to terminate their August agreement, which covers approximately $34.9 billion in computing resources.

Neither party has issued a termination notice, and Anthropic said it does not currently intend to end the agreement. The contract is therefore still in place, but either side can now walk away.

Computing capacity supports both model development and the products Anthropic sells. Customers pay for access to Claude through its platform and applications, including Claude Code for software development and Claude Cowork for office tasks. Businesses can also integrate Claude into their own products and processes, with access available through Amazon Web Services, Google Cloud Platform and Microsoft Azure.

As of September 30, 2026, more than 11,900 customers had each generated at least $100,000 of revenue for Anthropic over the preceding 12 months. More than 1,900 had each generated at least $1 million.

Earlier Vesting, Conditional Partner Purchases

The board on Monday moved the service-based vesting date for approximately 67 million restricted stock units—share awards earned by meeting specified conditions—to November 19 from November 20. Anthropic expects the modification to produce material additional stock-based pay expense.

In November, Anthropic signed purchase agreements for its previously described $26 billion private placement. Participants include Google, AMD, Nvidia, Broadcom, Micron Technology, Microsoft and TSMC Partners. Their purchases remain contingent on completion of the IPO and are scheduled immediately afterward, at the IPO price.

Google’s $10 billion purchase would leave $10 billion available under its existing investment agreement. AMD’s $2.5 billion purchase would reduce its remaining equity commitment to $2.5 billion; further draws would depend on chip-delivery and deployment milestones.

Anthropic’s proposed $70.00–$78.00 per-share range was carried forward, rather than set in this amendment; it has applied to list on the Nasdaq Global Select Market as ANT.

The research assessment is now firmer, while Anthropic’s own warning remains: safety constraints could slow the development that supports its revenue growth.

Sources