Amazon presented this Thursday its results for the second quarter of 2026 and the figure that has attracted the most attention does not come from e-commerce: AWS, its cloud division, grew 37% year-on-year to 42,200 million dollars, its fastest advance in 18 quarters. The cloud infrastructure business thus reaches an annualized revenue rate of 169,000 million dollars, and Amazon confirmed that both its artificial intelligence business and its own chips business already exceed 25,000 million annualized each.
A record quarter for Amazon’s cloud
Amazon’s total revenue grew 20% to 200,600 million dollars, with an operating profit of 27,500 million, 43% more than the previous year. But the star is AWS: its 16,600 million operating profit almost doubles the figure for the same period of 2025 and confirms that the cloud remains the biggest source of margin for the Seattle giant.
The most spectacular figure of the quarter, however, is an accounting one: net profit reached 62,600 million dollars, boosted by 53,400 million in non-operating income linked to the revaluation of Amazon’s investment in Anthropic. The firm, which has received nearly 8,000 million dollars from Amazon in recent years, has become a first-class financial asset for the company.
AWS’s strong growth comes amid massive investment: Amazon maintains its forecast of around 200,000 million dollars of capital expenditure for all of 2026, directed almost entirely at artificial intelligence infrastructure. Free cash flow has turned negative for the first time this year, down to -7,600 million, precisely because of that outlay.
Trainium and Graviton: the bet on in-house silicon
Andy Jassy’s message, Amazon’s chief executive officer, was unequivocal: AI and chips are now two businesses with more than 25,000 million dollars of annualized turnover and triple-digit growth. The Trainium accelerators, designed by Annapurna Labs, continue to gain traction: the two most important AI labs in the world, Anthropic and OpenAI, have signed multi-gigawatt multi-year capacity commitments, and a long list of startups —NEURA Robotics, Odyssey, TwelveLabs, Decart or Poolside, among others— has joined the platform.
On the general-purpose processor front, Amazon announced the general availability of Graviton5, which offers up to 25% more performance than Graviton4 and between 30% and 40% better price-performance than comparable instances. According to the company, 98% of the thousand largest EC2 customers already use some member of the Graviton family, making the chip based on the Arm architecture one of the most successful in-house silicon deployments in the industry.
Agents, security and the path towards cloud native
Beyond hardware, AWS took advantage of the presentation to strengthen its software arsenal. In Bedrock, its catalogue of managed models, more than ten foundational models were incorporated, including GPT-5.6 from OpenAI, Claude Opus 5 from Anthropic, Gemma 4 from Google DeepMind and Grok 4.3. The company assures that hundreds of thousands of customers already use the service.
It also presented AWS Continuum, a tool that discovers, prioritizes and corrects code vulnerabilities using agents and frontier models, and new capabilities for AI agents to execute payments and searches autonomously. In infrastructure, the Lambda MicroVMs promise instant startups and sessions of up to eight hours, designed for long-running agent loops.
The company also announced an investment of 1,000 million dollars in AWS Forward Deployed Engineering, a team of AI engineers that works directly with clients such as the NBA, the NFL, Southwest Airlines or Ricoh to deploy agentic solutions in days.
Pinterest, the example of the new era of commitments
Among the agreements highlighted in the quarter is that of Pinterest, which has committed 4,000 million dollars in cloud services until 2031, the largest infrastructure commitment in its history. The visual platform, with more than 600 million monthly users, will use Trainium and Graviton to train and run its visual search models, and plans to migrate its workloads towards EKS, AWS’s managed Kubernetes.
The pattern is repeated throughout the sector: OpenAI (38,000 million with AWS), Meta (more than 10,000 million with Google Cloud) or Snowflake (6,000 million) have opted for multi-year commitments in exchange for guaranteed accelerator capacity, in a market where chip scarcity continues to set the agenda. Analysts, who expected AWS revenue of around 40,500 million, saw their forecasts comfortably exceeded.
Conclusion
The quarter leaves two clear readings: the first, that the cloud remains the great growth and margin engine of Amazon, accelerating precisely when it demands the most investment. The second, that AWS has ceased to be a simple reseller of capacity to become a vertical integrator —chips, models, agents and engineering— that competes with Nvidia, Google and Microsoft on all AI fronts. With capital expenditure of 200,000 million planned for the year, Amazon is betting most of its balance sheet on that strategy continuing to bear fruit.
