This latest disclosure indicates that Anthropic still urgently needs more computing resources to train and run its AI large models.

In a regulatory filing submitted on Tuesday, Anthropic stated that over the past month, it finalized new procurement commitments with Microsoft (NASDAQ: MSFT) and Alphabet (NASDAQ: GOOGL): from now through 2033, an additional approximately $10 billion will be spent on leasing cloud servers, an amount not included in previously disclosed spending plans.

At the same time, Anthropic said that a $34.9 billion server space leasing agreement with smaller cloud provider Lambda could be canceled because Lambda failed to complete data center construction by the deadline agreed by both parties. However, Anthropic stated that although the contract gives it the right to cancel the deal, the company "currently has no intention of canceling."

This disclosure, on the one hand, reflects that even as Anthropic is carefully evaluating the delivery and fulfillment progress of smaller cloud providers like Lambda, it still urgently needs to expand computing power to support the training and inference of large models.

On the other hand, it also reveals that Lambda, like many peers in the industry, is struggling to quickly complete computing capacity construction to meet surging market demand.

The updated procurement commitments are further additions to Anthropic's previous total planned cloud service spending of approximately $650 billion.

According to Tuesday's filing, as of 2033: Anthropic's total committed spending on Microsoft Azure cloud was raised from $44.9 billion last month to $49.2 billion; its committed spending on Google Cloud was raised from $114.5 billion last month to $121.6 billion.

Other large contracts already signed remain unchanged: $161.2 billion for equipment procurement from Broadcom, $120.5 billion for computing power from SpaceX, $110 billion for cloud resources from Amazon, plus several smaller agreements with other cloud service providers.

Anthropic also issued a risk warning: in order to meet business demand, the company may sign more short-term computing procurement contracts. Short-term contract prices are much higher than long-term agreements, which will squeeze the company's operating margin in the short term.

The company reported an operating loss of $567 million in the third quarter; in the previous quarter, it recorded an operating profit of $974 million. The company said the main reason for the profit reversal was precisely these high-priced short-term computing procurements.