AWS Grows 37% but Amazon Says $220 Billion Won't Buy Enough AI Capacity

Amazon Web Services grew 37% in the second quarter to $42.2 billion, its fastest pace in 18 quarters, and Amazon raised its 2026 capital spending plan to about $220 billion from $200 billion, the company said on July 30. The headline that matters most for enterprise buyers came from CEO Andy Jassy: even at that level of spending, Amazon expects to fall short of the capacity its customers want this year and next, according to Let's Data Science. Read together with Microsoft's results the day before, the quarter suggests cloud AI capacity will stay a constrained resource that enterprises need to plan around, not a commodity they can assume.

What did Amazon report for AWS?

AWS revenue of $42.2 billion beat the $40.6 billion analysts expected, Stocktwits reported. Operating income for the unit rose to $16.6 billion from $10.2 billion a year earlier, per Let's Data Science. Jassy said AWS is "booming," Shelf Awareness reported, and that Amazon's AI business and its chips business have each passed a $25 billion annual revenue run rate. Let's Data Science reported both were growing at triple-digit rates.

Amazon's total sales rose 20% to about $200.6 billion. Reported profit was inflated by roughly $53 billion of non-operating income, mainly an unrealized gain on Amazon's stake in Anthropic, Stocktwits noted, so the AWS operating figures are the cleaner read on the core business.

Why is Amazon spending $220 billion, and why is it still short?

Amazon put about $53 billion into property and equipment in the quarter, net of some sale proceeds, per Stocktwits. Jassy attributed most of the increase in the full-year plan to higher memory costs, Let's Data Science reported, and the total also covers non-AWS projects such as robotics, semiconductors and satellites.

On demand, Jassy was blunt. "We will still not have enough capacity to meet all the demand we have in 2026," he said, according to Stocktwits. Let's Data Science reported that he expects the same constraint in 2027 and described demand already visible for 2028 as striking, while noting these are management expectations rather than commitments on when capacity will arrive.

Microsoft told a similar story on July 29. It said demand for Azure capacity still exceeds supply after adding about a gigawatt of capacity in its June quarter, Directions on Microsoft reported.

Is the spending sustainable? The case for and against

The bullish reading. AWS growth reaccelerated sharply while its operating income rose by more than 60%, which suggests the new capacity is earning money quickly. Investors appeared to agree: shares rose in after-hours trading. Amazon management also said AWS could eventually become a business with a few hundred billion dollars in annual revenue, per Stocktwits, an aspiration rather than guidance.

The cautious reading. Amazon's trailing 12-month free cash flow swung to an outflow of $7.6 billion from an inflow of $18.2 billion a year earlier, Let's Data Science reported. The jump in the capex plan was driven by component costs, not only by new demand. And the company's own forecast is that it will stay capacity-constrained for at least another year, which means customers may be competing for the same GPUs and memory for some time.

Both readings can be true at once. Demand is real and profitable today, and the cost of meeting it is rising.

What does a capacity-constrained cloud mean for enterprise buyers?

Several implications follow for CIOs and chief AI officers.

Capacity becomes a negotiating item. When providers say they cannot meet all demand, reserved capacity and committed-use agreements become more valuable. Large inference workloads planned for 2027 deserve procurement attention now, with clear terms on what happens if capacity is late.

Watch for component costs in pricing. Amazon tied its higher spending to memory prices. Whether and how such costs reach customers is not yet clear, but it is a reasonable question to raise with account teams before multi-year renewals.

Efficiency is a capacity strategy. Every token a team does not need to generate is capacity it does not need to buy. Smaller models for routine tasks, caching and splitting CPU and GPU work all stretch scarce resources. Our report on AI infrastructure at scale and inference costs describes how Rhino Federated Computing runs only inference on GPU machines and keeps preprocessing on CPUs.

Keep cost monitoring tight. Scarcity and rising unit costs make unmonitored usage more expensive. Lockton's Varun Tottempudi, head of data and platforms engineering for global analytics, warned at COLLIDE 2024 that without monitoring and alerts, "in six months you might see a bill" as large as the past five years' spending. The panel's guidance on AI ROI across the project life cycle remains a practical checklist.

Weigh self-hosting carefully. Tight cloud supply can make on-premises or self-hosted models look attractive. "Hosting your own models is a really big investment," Alteryx principal engineer Jonathan Lounsbury cautioned in a AI Connect webinar, citing developer time, cost and scarce GPU instances. Our coverage of self-hosted LLMs and vendor lock-in lays out the trade-offs.

What to watch next

Amazon guided to third-quarter revenue of $197 billion to $202 billion and operating income of $22.5 billion to $26.5 billion, Stocktwits reported. The more telling numbers for enterprise buyers will be whether AWS growth holds near this level as new capacity comes online, and how large Amazon's 2027 spending plan becomes, given Jassy's expectation that supply will remain tight next year.

Frequently asked questions

How fast did AWS grow in Q2 2026?

AWS revenue grew 37% year over year to $42.2 billion, its fastest growth in 18 quarters, Amazon said on July 30, 2026.

How much will Amazon spend on capex in 2026?

About $220 billion, up from the $200 billion plan announced in February, with higher memory costs the main reason; the total includes non-AWS spending.

Why does AI capacity matter to enterprise buyers?

If providers cannot meet demand, reserved capacity becomes more valuable and unit costs may rise, so enterprises should plan inference capacity and costs ahead.

Filed under: Data & AI Infrastructure, AI Strategy & ROI