Amazon: Is the cloud leader also the AI leader?

In the early stages of the generative AI boom, market attention was almost entirely monopolized by OpenAI, Microsoft, and Google. Although Amazon still controls the world's largest public cloud platform, it faced questions about missing out on the model race. Now, AWS is launching a comprehensive counterattack with its self-developed chips, model platforms, and enterprise customer base. However, whether the "cloud leader" can naturally extend its dominance to become the "AI leader" depends on more than just traditional cloud market share.

According to Synergy Research Group, the global cloud infrastructure market reached $143 billion in Q2 2026, a 43% year-over-year increase; AWS held a 28% share, still leading Microsoft's 20% and Google's 15%. However, AWS's share has declined compared to a year ago, while Microsoft and Google are growing faster, reflecting how AI is redistributing new orders in the cloud market.

AWS's counterattack is first evident in its revenue. In Q2 2026, AWS revenue grew 37% year-over-year to $42.2 billion, the fastest growth in 18 quarters, with operating profit rising to $16.6 billion. Amazon further disclosed that the annualized revenue from AWS's AI business and the group's chip business has both exceeded $25 billion, maintaining triple-digit growth. This indicates that AWS's AI narrative is beginning to translate from product demonstrations into actual revenue, rather than relying solely on "natural upgrades" from existing cloud customers.

Its first card is the self-developed Trainium chip. The cost of training and inference for large models is extremely high. If AWS were to rely long-term on NVIDIA GPUs, not only would supply be constrained by others, but profits would also be easily siphoned off by upstream suppliers. The true significance of Trainium is that it allows AWS to integrate chips, servers, networking, software, and cloud services to compete for every dollar of computing revenue behind each AI token. Anthropic's Project Rainier, which currently trains and runs Claude using nearly 1 million Trainium2 chips, serves as both an important reference customer and a key case study for AWS to alleviate market concerns.

The second card is Amazon Bedrock. Microsoft gained a first-mover advantage through OpenAI, while Google boasts Gemini and TPUs. However, AWS has chosen to position itself as a "model-agnostic" platform, allowing enterprises to choose among models from Anthropic, OpenAI, Meta, Mistral, DeepSeek, and Amazon Nova. For enterprises, the most powerful model does not always come from the same vendor; the ability to control data, costs, security, and the risks associated with switching models is often more important than leaderboard rankings. If Bedrock can further integrate models into enterprise workflows via AgentCore, AWS's traditional customer relationships could once again become its moat in the AI era.

However, competitive pressure has not dissipated. During the same period, Azure and other cloud services revenue grew by 43%, with Microsoft 365 Copilot paid seats exceeding 30 million; Google Cloud revenue surged 82% to $24.8 billion. While disclosure methodologies vary across companies and direct comparisons are not entirely appropriate, Microsoft leverages its Office, GitHub, and Copilot entry points, whereas Google integrates Gemini, Search, data, and TPUs into a unified ecosystem. Both are closer to end-users than AWS.

The costs are equally significant and cannot be ignored. Amazon has raised its 2026 capital expenditure plan to $220 billion, while free cash flow over the past 12 months has shifted from positive to an outflow of $7.6 billion. Even with strong AI demand, data centers involve challenges related to power consumption, depreciation, chip iteration, and price competition. If customer demand growth fails to keep pace with capacity expansion, today's moat could become tomorrow's asset burden.

Therefore, while AWS remains the undisputed leader in cloud computing and has re-entered the ranks of AI leaders, there is currently no single definitive "AI leader." What investors should truly monitor is not how many models Amazon releases, but whether Trainium can expand beyond major clients like Anthropic, whether Bedrock and AgentCore can achieve large-scale production deployment, and whether AWS growth can consistently outpace the broader market. If these three conditions are met, AWS's late-stage counterattack could mirror its earlier rise in the cloud era; otherwise, it may remain merely the largest cloud platform rather than the defining platform of the AI age.

Yee Hap Holdings Investor Relations Department

(Chips and Computing Power Series #87)

$Amazon (AMZN.US)$ $Oracle (ORCL.US)$ $Meta Platforms (META.US)$ $SpaceX (SPCX.US)$ $Tesla (TSLA.US)$

72K Views