Amazon's AI Equation: Sold-Out Cloud Capacity Meets a $220 Billion Bill

Amazon has little trouble finding customers for its cloud and AI infrastructure. The harder question, heading into the company's October 29 earnings report after the market close, is whether the economics of that business can keep pace with the capital being poured into it.

Andy Jassy, the group's chief executive, argued Thursday that inference costs — the expense of running already-trained AI models — need to fall to roughly a tenth of current levels before companies adopt the technology at scale. Amazon's in-house Trainium chips are central to that push. Jassy puts Trainium's price-performance advantage at 30–40% over comparable processors, and Amazon has already disclosed future Trainium revenue commitments exceeding $225 billion, including agreements with Anthropic and OpenAI.

Those figures represent contracted future revenue, not cash already collected. Turning them into actual sales depends on how quickly Amazon can bring promised computing capacity online.

Demand Is Not the Problem

By Jassy's account, AWS capacity is sold out for 2026 and largely committed for 2027, with a substantial share of 2028 already booked as well. Even so, he acknowledged that significant demand will go unmet despite the buildout. Reuters reported his remarks on Wednesday.

Should investors sell immediately? Or is it worth buying Amazon?

The scale of the spending matches the ambition. Amazon plans capital expenditures of about $220 billion for 2026, a forecast Morningstar notes was raised in July on the back of higher memory prices — meaning the added burden stems from more than infrastructure expansion alone.

That outlay creates capacity for the growing AI business while weighing on free cash flow and, through depreciation, raising costs in future periods. Morningstar data shows depreciation climbed 67% in the prior quarter, with trailing free cash flow slightly negative. Growth and investment therefore cannot be judged in isolation; the return on deployed capital is what matters.

Trainium's Software Story

Technical progress is broadening what Amazon's chips can do. SemiAnalysis reported Wednesday that Trainium2 and Trainium3 can natively run research training on PyTorch, extending their use beyond inference into model training itself. The Qwen3-32B model was used in the tests.

The same report flagged limits: the CUDA ecosystem remains far more mature, and the tests do not yet support sweeping conclusions about every large training workload. For Amazon, raw chip performance is only part of the battle — how easily developers can port existing workflows onto its stack matters just as much. Cheaper hardware and a friendlier software environment serve the same end, raising the incentive to run AI applications on Amazon's infrastructure.

Reading the October 29 Numbers

The cloud unit provides the clearest yardstick. AWS revenue grew 37% in the second quarter of 2026, with an operating margin of 39.4%. Wells Fargo, which on Wednesday named Amazon its preferred mega-cap stock ahead of the results, expects upward revisions to AWS revenue and operating income estimates on accelerating growth and strong margins, though it cited fuel costs as a near-term risk to earnings.

Amazon at a turning point? This analysis reveals what investors need to know now.

For the third quarter, Amazon has guided to group revenue of $197.0–202.0 billion. The figure needs context: the shift of Prime Day into the prior quarter will weigh on reported growth by roughly 4 percentage points, according to Morningstar, so a softer headline number would not by itself signal fading demand. Operating income is guided to $22.5–26.5 billion.

Below the operating line, the picture gets murkier. Of $120.7 billion in pre-tax profit in the first half of 2026, more than $69 billion came from unrealized valuation gains on the Anthropic stake — paper gains, not operating earnings. A net-income comparison alone can therefore obscure how the cloud and retail businesses are actually performing.

Morningstar expects margins to improve over the long run on the strength of AWS and advertising. The upcoming report must square that outlook with the current capital burden. Three things will matter most to investors: operating profitability, the trajectory of depreciation, and progress toward a sustainably positive free cash flow.

Ad

Amazon Stock: New Analysis - 24 October

Fresh Amazon information released. What's the impact for investors? Our latest independent report examines recent figures and market trends.

Read our updated Amazon analysis...