By a Senior Tech Correspondent | Updated Aug 1, 2026
Amazon just hiked its 2026 capital spending to roughly $220 billion and CEO Andy Jassy says even that won’t be enough to meet demand. In the same breath on Thursday’s Q2 call, the company admitted AI capacity will stay constrained through 2027, with contracted demand already stretching into 2028.
Here’s the number that should make every investor pause: free cash flow swung to a $7.6 billion outflow over the trailing twelve months. That’s a sharp reversal from a year ago. Amazon is now bleeding cash at a time when its own CEO is asking shareholders to trust that the bill will be worth it “a few years” down the road.
Why It Matters
This isn’t just Amazon’s problem it’s the whole hyperscaler economy’s problem. AWS revenue jumped 37% year-over-year to $42.2 billion in Q2, its fastest growth in 18 quarters, with an annualized run rate of $169 billion. The backlog of contracted business sits at $496 billion, growing at triple-digit rates.
But that growth comes with a brutal trade: the cloud market grew 43% to $143 billion in Q2, and Amazon, Microsoft, and Google control 67% of it. When the incumbents collectively spend a trillion dollars on AI infrastructure, that’s not a bet on winning the future that’s the cost of not losing the present. This is the same dynamic behind our look at the $700 billion AI arms race and why big tech fears extinction more than losses. Constellation Research analyst Holger Mueller puts it plainly: “We are in the gold rush era you need to build it to sell.”
Technical Breakdown
The $20 billion increase over the prior $200 billion guidance is telling. Jassy attributes it less to a broader construction push and more to memory chip costs the volatile, supply-constrained components that are quietly becoming the new bottleneck in AI economics. Amazon flagged “resource and supply volatility, including for memory chips” as a business risk, which is corporate code for: our suppliers are holding us hostage.
The strategy underneath is twofold. Amazon is leaning hard on its in-house silicon — Trainium and Graviton chips now generate more than $10 billion in annual run-rate revenue to blunt its dependence on Nvidia GPUs. It’s a bet that the most interesting chip of the AI era isn’t the most powerful one, but the one you own. And AWS is even in talks to make Trainium available outside AWS, a move that would turn a cost center into a competitive weapon.
“The next phase of AI investment will be shaped as much by silicon and memory economics as by concrete, steel, and power,” said HyperFrame Research’s Steven Dickens. Amazon plans to double its power capacity by the end of 2027 compared to 2025. Much of that planned 2027 capacity is already reserved by customers. If you want to understand why Amazon is racing to design its own hardware instead of renting Nvidia’s, the hidden infrastructure behind AI agents explains the GB200 cost math that’s driving every hyperscaler decision right now.
The Catch / Friction
Here’s where the story gets uncomfortable.
First, the numbers only add up if AI demand stays sticky. Jassy’s own messaging is hedged: he says revenue growth will eventually outpace capex growth, “the resulting revenue, free cash flow, and return on invested capital is very compelling.” Translation: not today, not next year.
Second, there’s the question of whether this is offense or defense. Amazon is building partly because customers have already signed contracts $496 billion of them. But it’s also building because Microsoft, Google, and a pack of well-funded “neoclouds” like CoreWeave, OpenAI, and Oracle are eating into its territory. When your competitors are spending to your left and right, cutting capex isn’t prudent — it’s surrender.
Third, the memory crunch is a double-edged sword. Rising DRAM and HBM prices are inflating Amazon’s bill and its competitors’. But they also mean Amazon’s costs rise on every megawatt it adds. The same components that drove the $20 billion increase could squeeze margins just as hard next quarter the same surging Nvidia B300 prices that are crushing smaller players are hitting Amazon’s procurement team too. At some point, a skeptic has to ask whether the whole thing like IBM’s $10 billion gamble before it — is spending scale in the hope that ROI shows up later.
Looking Ahead
The honest read: Amazon has turned itself into a utility company with AI ambitions and utilities get judged on yield, not speed. The market already punished the company once this year, sending shares down 7-11% in after-hours trading when the $200 billion figure first landed in February.
The $220 billion revision buys capacity, not certainty. If the AI boom holds, Amazon’s backlog turns into a cash machine by 2028. If the froth deflates, Amazon owns the world’s most expensive parking lot of GPUs and a free-cash-flow problem that dwarfs anything the company has managed before. For users, this means AWS prices stay high and capacity stays scarce through at least 2027. For investors, it means Amazon is betting the house on a single question: does AI demand turn out to be a bubble, or a decade? We’ll start getting the answer in about 18 months when all that concrete, steel, and silicon is finally lit up.
source: reuters
Independent technology writer focused on artificial intelligence, emerging technologies, and digital innovation. Covers AI applications in sports, productivity, and online business.
