Amazon invests an additional $2.5 billion in Anthropic: It’s not buying AI models, but a moat of computing power

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1527 words

Amazon officially announced that it has recently made an additional $5 billion investment in Anthropic, with plans to potentially invest another $20 billion contingent upon the achievement of commercial milestones by Anthropic. (Context: Thawing relations between Anthropic and the Trump administration? Treasury Secretary and White House Chief of Staff meet with CEO Dario Amodei) (Background: Anthropic's new model Mythos is so powerful the company is hesitant to release it: capable of autonomously hacking global Linux systems and chaining complete exploit chains in hours) According to reports, the company has invested $5 billion in Anthropic, with the potential for up to $20 billion in additional funding in the future. Including the $8 billion previously invested, Amazon's total committed capital to this AI startup could reach $33 billion. The structure of this investment, officially confirmed by Amazon on Monday, reveals the company's true strategic intent. The $20 billion in additional investment is not a lump sum but is tied to Anthropic's "commercial milestones." While the specific metrics for these milestones have not been fully disclosed, the terms of the exchange are clear: Anthropic commits to spending over $100 billion on AWS (Amazon Web Services) over the next decade and using Amazon's self-developed AI chips, Trainium2 and Trainium3, as its primary source of training compute. The goal is to reach approximately 1 GW of compute power by the end of 2026, with a long-term target of 5 GW. In plain terms: Amazon gives money to Anthropic, and Anthropic spends that money back on Amazon's platform and chips. This is not a purely financial investment; it is a procurement contract combined with a valuation lock-in mechanism. Amazon does not need actual control over Anthropic's board; it only needs to bind the startup's capital requirements to its own cloud performance. To understand this deal, one must first grasp a key point: What is the problem Amazon has been most eager to solve in recent years? The answer is not "owning a more powerful AI model," but "getting more people to use its own chips." NVIDIA currently holds a near-monopoly on the AI training compute market. Cards like the H100 and B200 are in short supply, expensive, and entirely in the hands of a competitor. Amazon began developing its own AI training chips years ago, codenamed Trainium, designed specifically for deep learning with the goal of being cheaper than NVIDIA and more deeply integrated with AWS. The problem is that Trainium has long lacked a heavyweight flagship customer. For enterprise technology decision-makers, "migrating core AI training workloads to unproven chips" is a significant gamble that requires someone to test, run, and validate them first. Anthropic is that validator. If one of the world's top AI labs chooses to run its core training workloads on Trainium, the market signal sent by this endorsement is far more effective than any advertisement. Amazon CEO Andy Jassy specifically emphasized the "progress in collaboration on custom chips" in his statement, and it was no mere pleasantry. Amazon's capital expenditure for 2026 is projected to reach $200 billion, with the vast majority invested in AI infrastructure. With a self-bet of this scale, spending $25 billion to secure chip usage endorsements from a top AI lab is a calculation that makes perfect sense. Furthermore, this business model has a precedent. Since 2019, Microsoft has invested over $13 billion in OpenAI, in exchange for: exclusive hosting of OpenAI models on the Azure cloud platform, a 49% profit share for Microsoft, and priority integration licensing for the Copilot series of products. In terms of structure, Microsoft traded investment for compute procurement guarantees and model usage rights, while OpenAI traded valuation for training resources and enterprise customer channels. The two have formed a deep commercial lock-in, making it difficult for either side to unilaterally exit. The relationship between Amazon and Anthropic is moving toward a similar structure. The difference is that Amazon places greater emphasis on chip dominance rather than model application channel distribution. This indicates Amazon's judgment that the next bottleneck in AI competition is not at the cloud service layer, but in the compute infrastructure itself. The cloud wars have been fought for twenty years, and the outcome has often depended on who could turn potential customers into long-term users who have no choice but to stay. With this $25 billion, Amazon is applying that same logic.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-05-29

Anthropic closes $65 billion Series H funding round at a $965 billion valuation, surpassing OpenAI

Similarity 140%關鍵字 anthropic/億美元同分類 zh

2026-05-21

Anthropic will pay SpaceX $1.25 billion per month for xAI's excess computing power.

Similarity 140%關鍵字 anthropic/億美元同分類 zh

2026-05-21

Anthropic announces expected first-ever quarterly profit in Q2: Quarterly revenue doubles to $10.9 billion, hitting target two years ahead of schedule

Similarity 140%關鍵字 anthropic/億美元同分類 zh

2026-05-13