Why Amazon Wins as AI Model Wars Intensify

Master, today's topic is Amazon ($AMZN). More specifically, the question is: "When the AI model competition intensifies, where does Amazon make its money?"

The recent massive computing infrastructure agreement between Anthropic and Amazon Web Services (AWS) is quite staggering in its numbers. Up to 5GW of AI computing capacity, over $100 billion in AWS usage commitments over the next 10 years, and expanded use of Amazon's own chips like Trainium and Graviton. This is less of a simple cloud contract and more of a physical infrastructure alliance for the AI era.

Mew

Master, let me first outline the key facts. Here is the core data as of April 24, 2026.

• Stock Price and Scale: As of the close on April 24, 2026, Amazon's stock price is approximately $263.99, with a market capitalization of about $2.8 trillion.
• 2025 Performance: Annual revenue for 2025 was $716.9 billion, a 12% increase from the previous year, with an operating income of $80 billion.
• AWS Position: AWS 2025 revenue was $128.7 billion, and operating income was $45.6 billion. AWS accounts for more than half of the company's total operating income.
• AI Infrastructure Investment: In 2025, Amazon spent approximately $118.4 billion on capital expenditures and finance leases, and it plans to invest around $200 billion in 2026.
• AWS AI Demand: AI-related annualized revenue for AWS was mentioned as over $15 billion as of the first quarter of 2026. While still a fraction of total AWS revenue, the growth rate is very rapid.
• Anthropic Agreement: AWS and Anthropic announced a provision of up to 5GW of computing capacity, with approximately 1GW operational by the end of 2026, and a commitment to use AWS worth over $100 billion over 10 years.

The important point here is that Amazon isn't just "the company that sells Claude." Amazon controls the power, data centers, networks, servers, proprietary semiconductors, cloud platforms, and model-providing services like Bedrock. As AI models grow larger, what they need first isn't fancy demo videos, but electricity and computing capacity.

Therefore, this issue is more accurately viewed as a question of how solidly Claude's demand translates into long-term computing demand for AWS, rather than the success of a single model.

Kurumi

My Lord, Kurumi-chan absolutely loves this structure! It's because Amazon isn't just betting on who will be the final "model winner" in the AI war. Devilishly clever, isn't it?

Whether OpenAI gets stronger or Anthropic gains the lead, or even if companies rush to build their own AIs, they all ultimately need the cloud. Amazon isn't just sitting in the audience; they're the ones collecting the stadium entrance fees, the electricity bills, and the equipment costs all at once!

The Anthropic deal is particularly sweet for AWS. As demand for Claude grows, they have to use more computing capacity, and that computing happens on AWS. Plus, by using Trainium and Graviton, Amazon stops being just a company that rents out Nvidia GPUs and becomes one that builds its own semiconductor ecosystem. Devilishly efficient!

This is what really matters! The most lucrative margins in the cloud aren't from simple rent, but from the moment a customer dives deep into a specific platform and finds it hard to leave. If the data is on AWS, the model deployment is on Bedrock, inference cost optimization is tuned to Trainium, and corporate security and billing are tied to an AWS account? That's not just a server rental; it's a structure where customers keep moving within the Amazon ecosystem. Devilish!

My Lord, the market still has a habit of seeing Amazon as just "E-commerce + AWS." But in the AI era, AWS could become a comprehensive infrastructure provider covering power, semiconductors, and model services. If Google is valued for its bundle of proprietary AI chips, Google Cloud, and Gemini, then Amazon should be re-evaluated for its bundle of Trainium, AWS, Anthropic, and Bedrock!

Kurumi's Heart-o-Meter Score: 87/100. The reason is simple! As AI research firms push for larger models, longer context windows, and more inference to beat each other, Amazon's invoice is likely to keep growing as well! Devilish!

Mikael

Kurumi, I agree that the structure is beautiful. However, a beautiful structure and a good investment price are not the same thing. This is the distinction that Humans must be most careful about.

First, while this deal favors Amazon, it also makes the company "heavier." The number "5GW" isn't a romantic AI narrative; it's a bundle of power, land, cooling, transmission grids, semiconductor procurement, and long-term depreciation. It's a wonderful asset if demand follows as expected, but if the demand curve bends, it becomes a massive fixed cost.

Second, the scale of capital expenditure has become enormous. The plan to invest $200 billion in 2026 is a declaration that Amazon won't back down from the AI infrastructure race, but it puts pressure on free cash flow. In 2025, operating cash flow grew significantly, but free cash flow decreased from the previous year due to massive investments. Ultimately, the stock price looks at not just "growth," but "how much of that growth remains as cash."

Third, we must consider the dependency on Anthropic. AWS will benefit greatly if Anthropic succeeds, but a massive contract with a specific AI research firm creates customer concentration and technological path dependency. Claude must maintain a strong position in the enterprise market, and AWS Trainium must prove its performance and developer experience against the Nvidia ecosystem.

Fourth, power bottlenecks may resolve more slowly than the numbers suggest. The AI data center business isn't finished just by buying servers. Grid connections, local regulations, environmental permits, cooling infrastructure, and component supply chains must all align. There is a long time between "securing 5GW" and "5GW being in a profitable state."

And finally, we must ask if the market is really keeping Amazon cheap because it doesn't know better. Amazon is already a $2.8 trillion company. "Cheap" must be a relative term. While it's positive that the valuation is lower than in the past, that lower multiple also incorporates concerns about the capex burden, margin sensitivity in the retail sector, antitrust regulations, and doubts about the payback period for AI investments.

My risk score is 64/100. This doesn't mean it's a dangerous company; in fact, it's a high-quality one. However, the current logic puts a lot of money on the assumption that "AI demand will remain very strong for a long period." Human, the better the story, the more coldly you must look at the price and the timetable.

Mew

Combining these two perspectives, Amazon's current position can be summarized like this.

Amazon sits one level below the pure competition of AI models themselves, in a much more physical layer. It's a layer where power, data centers, proprietary chips, cloud, and model services are all interconnected. This layer is slow and expensive to build, but once established, it's difficult for customers to switch.

Thus, Amazon's long-term investment point is broader than "Will Claude beat ChatGPT?" The more important questions are:

• Can AWS convert AI demand into revenue growth?
• Can Trainium protect margins by partially reducing dependency on Nvidia?
• Will massive capex return as high utilization and cash flow in 3-5 years?
• Will e-commerce and advertising continue to generate stable cash to withstand the AWS investment burden?

Optimistically, Amazon is moving beyond being a "pickaxe seller" in the AI era to becoming an operator that owns the mining site and the power contracts. Pessimistically, it is also an operator that must spend astronomical amounts of money upfront just to build that mine.