US unicorn Anthropic secures $15 billion investment commitment from Microsoft and NVIDIA, subtly altering the landscape

On November 18th, Eastern Standard Time in the United States, Anthropic, a globally renowned startup in the field of large models, announced a strategic partnership with Microsoft and NVIDIA.

In this strategic cooperation, NVIDIA and Microsoft have promised to invest $10 billion and $5 billion respectively in Anthropic. In return, Anthropic has committed to purchasing at least 1GW of computing power from NVIDIA (GW, or gigawatt, is a unit of power. A 1GW computing power cluster can accommodate 200,000 NVIDIA GB200 chips).

Before this investment commitment, Anthropic had raised over $31.2 billion in cumulative financing and was valued at $183 billion. Based on this valuation, it was already the second - most highly valued large - model startup globally. Sources said that this investment will push Anthropic's valuation up to $350 billion, second only to OpenAI's $500 billion.

Dario Amodei, the CEO of Anthropic, Satya Nadella, the CEO of Microsoft, and Jensen Huang, the founder and CEO of NVIDIA, issued a joint statement via video.

In the statement, Amodei expressed his strong belief in "choice." He said, "I'm very glad to bring Anthropic to Microsoft Azure." This means that Anthropic will be the first model available on all three major global cloud providers (Amazon AWS, Google GCP, and Microsoft Azure).

Amodei stated that Anthropic will jointly optimize the model with NVIDIA. Starting with Blackwell (NVIDIA's flagship chip on sale in 2025), and then moving on to Rubin (NVIDIA's next - generation flagship chip, expected to enter mass production in the second half of 2026). The two sides have a computing power capacity plan of 1GW, and this is just the beginning.

Nadella said in the joint statement that Microsoft and Anthropic will increasingly become customers of each other. Customers of Microsoft Foundry (Microsoft's AI service platform) will be able to use Anthropic's Claude series of models. Microsoft's Copilot (AI office assistant) product family will also integrate Anthropic's Claude series of models. Anthropic has committed to using Microsoft Azure's computing power capacity.

Jensen Huang said in the joint statement that NVIDIA can't wait to use Anthropic's Blackwell to accelerate Claude. This will help Anthropic expand its computing power more quickly and reduce Token costs (Token is a text unit in large models. A Token can be a word, punctuation, number, symbol, etc.), truly spreading AI everywhere.

01

Signs of Loosening in the Amazon - Anthropic Partnership

Establishing a strategic partnership with NVIDIA and Microsoft means that Anthropic's "strong - binding" relationship with Amazon is weakening. The market is worried about whether there will be changes in their cooperation.

After OpenAI emerged in early 2023, Amazon invested $4 billion in Anthropic in September 2023 and an additional $4 billion in November 2024. This accounts for more than 25% of Anthropic's publicly disclosed financing of $31.2 billion (see "Amazon Invests Another $4 Billion in Anthropic, the Biggest Challenger to OpenAI"). It is also Amazon's "bridgehead" in the AI race against Microsoft (see "OpenAI and Anthropic Battle in the Open, Microsoft and Amazon Confront Behind the Scenes").

It can be said that Amazon is Anthropic's closest ally, and the two are most closely bound. Their relationship is similar to that between Microsoft and OpenAI.

Anthropic officially emphasized that Amazon remains Anthropic's main cloud service provider and training partner.

In the second half of this year, the cooperation between Anthropic and Microsoft became delicate. At the end of October, Anthropic announced that it would spend tens of billions of dollars to purchase 1GW of Google's self - developed TPU (Tensor Processing Unit, an AI chip) chips. This time, Anthropic has also committed to using Microsoft Azure's cloud services and NVIDIA's chips.

Previously, Anthropic had been deeply relying on Amazon AWS's cloud services and self - developed chips. According to Caixin, Anthropic is still using Amazon's self - developed Trainium 2 chips to train large models.

A person from Amazon AWS told Caixin in 2024 that the strategic cooperation between Amazon and Anthropic is deeper than what the outside world sees. Since Amazon's self - developed Nova series of models are not strong enough, Anthropic's Claude series of models hold a high position in Amazon AWS's sales system. Although Amazon offers customers dozens of mainstream models to choose from, it often recommends Anthropic's Claude series of models first.

Tom Brown, the co - founder and chief computing officer of Anthropic, announced the "Project Rainier" at Amazon AWS's re:Invent 2024 conference in December 2024. Anthropic will use a computing power cluster with hundreds of thousands of Amazon's self - developed AI chips (see "Anthropic to Use Hundreds of Thousands of Amazon's Self - developed Chips").

With Amazon's support, Anthropic has rapidly grown into OpenAI's most powerful competitor in just two years. Anthropic disclosed in September this year that its annualized revenue in January 2025 (monthly revenue × 12. Software companies usually charge on an annual/monthly subscription basis, so this statistical method is applicable) was $1 billion, and it climbed to $5 billion by the end of August, a 400% increase.

The performance of Anthropic's Claude series of models has always been on par with OpenAI's GPT series. Claude even has an advantage in the code field.

OpenRouter is a global large - model API (Application Programming Interface) aggregation platform that integrates mainstream global models and is used by developers worldwide. It can represent the usage preferences of startups and cutting - edge developers. OpenRouter data shows that as of November 19th, the Token call volume of Anthropic's models ranks third globally, accounting for 14%.

02

Microsoft and Amazon Poach from Each Other's Competitors

At the end of October this year, OpenAI announced a capital restructuring (see "OpenAI Restructures, Clarifies Microsoft's Major Shareholder Rights, and Reinforces Non - profit Organization Position").

After that, OpenAI and Microsoft also ended their strong - binding relationship. Microsoft no longer has the priority to be OpenAI's computing power service provider. Taking this opportunity, Amazon took away an important customer from Microsoft - OpenAI. In November this year, Amazon AWS and OpenAI signed a contract worth at least $38 billion over seven years. These cloud services will start to be deployed by the end of 2026 (see "OpenAI Signs a New $38 - Billion Deal with Amazon. How to Fulfill the Nearly $600 - Billion Computing Power Deal Remains a Mystery").

The competition between Microsoft and Amazon for Anthropic will make it a beneficiary. Although Anthropic has received investments from Amazon and Google multiple times, it has always been seeking strategic independence.

Avoiding being completely controlled by giants is an important design in Anthropic's governance structure. Anthropic was established as a Public Benefit Corporation (a special form of company under the laws of the state of Nevada in the United States). Legally, the board of directors and management of a Public Benefit Corporation are allowed to prioritize social interests and AI safety over simple shareholder returns.

On this basis, Anthropic has adopted a special governance structure. Shareholders such as Amazon and Google do not have seats on the board of directors or voting rights, but only observer seats.

A front - line salesperson from an international cloud service provider told Caixin that Anthropic's market strategy also conforms to the rules of the B2B market - providing customers with choices.

Anthropic's "multi - cloud strategy" allows it to use Amazon AWS, Google Cloud GCP, and Microsoft Azure simultaneously. The "multi - cloud strategy" also sends a signal - not taking sides. No matter which cloud their data and applications are deployed on, Anthropic can provide normal services. This makes it easier to gain the trust of enterprise customers.

This article is from the WeChat official account "Caixin AI Pulse." Author: Wu Junyu, Editor: Xie Lirong. Republished by 36Kr with permission.