Amazon Reports $200.6 Billion in Quarterly Revenue as AWS Grows 37%; Raises 2026 Capital Spending Plan to About $220 Billion

The Number That Went Two Ways at Once

Amazon reported $200.6 billion in quarterly sales on July 30, 2026 — the first time the company has ever cleared $200 billion in three months^[1][2]. That is up 20% from $167.7 billion a year earlier. Operating income jumped 43%, to $27.5 billion^[1].

The cloud unit, Amazon Web Services, grew 36.7% to $42.2 billion, its fastest pace in 18 quarters^[2][3]. AWS also posted a 39.4% operating margin, meaning nearly 40 cents of every dollar of AWS revenue became profit^[3]. The stock jumped the next trading day, with reported gains ranging from about 9% to 15% depending on the outlet and the moment measured^[7][8][20].

Here is the other half of the same report. Amazon raised its 2026 spending plan on data centers, power hookups and AI chips to about $220 billion, up from roughly $200 billion^[1][2]. That spending now outpaces the cash coming in. Free cash flow — the money left after running the business and buying the equipment to grow it — turned negative over the past year, and Amazon's long-term debt nearly doubled, from $65.6 billion to $119.1 billion^[5][10].

Both of these things are true about the same company in the same quarter. A record-breaking business and a company borrowing heavily to keep building are not opposites. They are the same bet, described from two different vantage points.

Why Amazon Would Rather Overbuild Than Wait

To understand why Amazon is spending faster than it earns, you have to understand how cloud computing capacity actually works. When a company needs AI computing power, it goes to whoever has room available right now. If Amazon does not have that capacity built and waiting, the customer signs with Microsoft or Google instead — and rarely switches back once they are set up elsewhere.

That is why Amazon executives told analysts that capacity for 2027 is already largely reserved, and some 2028 capacity is spoken for too^[4]. The company points to a backlog of about $496 billion — contracts customers have already signed but have not yet started using — as proof this is not a guess^[4]. In Amazon's telling, this looks a lot like the early years of AWS itself, when heavy spending in 2008 looked reckless and later became the company's biggest profit engine.

That argument has a specific number behind it now, not just a story. AWS's own AI business, and the custom chips Amazon builds to run it, each passed $25 billion a year in revenue, more than doubling from a year earlier^[3]. Those in-house chips matter because Amazon avoids paying a supplier's markup on the AI hardware — which is part of why margin rose even as spending rose^[3].

Wall Street analysts mostly bought the argument, raising price targets and keeping buy ratings after the quarter^[6]. But that same appetite for capacity is what pushed Amazon into debt. Building for 2027 and 2028 demand means spending money now that revenue from that demand has not yet delivered — and Amazon chose to close the gap by borrowing rather than waiting.

The Answer Andy Jassy Did Not Give

On the earnings call, an analyst asked how Amazon would pay for the extra $20 billion in spending. CEO Andy Jassy said he had "nothing to share" on that point^[9]. Amazon's peers in the AI buildout — Microsoft, Google, and others — are largely funding their spending out of cash they already generate. Amazon is the outlier, closing its funding gap with borrowed money instead^[23].

The scale of that borrowing is why credit analysts are paying closer attention than usual. Amazon's free cash flow swung from roughly $18 billion to $26 billion positive a year ago to negative now^[5][21]. Morgan Stanley analysts project Amazon will end 2026 nearly $17 billion in the negative; Bank of America models a $28 billion shortfall.

Debt itself does not make a bet wrong. But it changes what happens if the bet goes slowly instead of badly. A company spending its own cash can quietly slow down if demand disappoints. A company that borrowed the money still owes the interest either way, whether AI revenue shows up on schedule or not.

There is also a mismatch in how long things last. AI chips can become outdated within three to five years. The debt used to buy them runs much longer than that. Skeptics are not arguing AI demand is fake — they are arguing about what happens if it merely grows slower than Amazon has bet it will^[10][23].

The Bill That Shows Up Somewhere Else

There is a second dispute running alongside the first one, and it involves people who own no Amazon stock at all. Data centers use enormous amounts of electricity, and someone has to pay to build the power lines and generating plants that supply them.

Under the rules most states use, when a utility builds new transmission or power capacity to serve a big customer like a data center, those costs get folded into the utility's "rate base" — the pool of costs spread across everyone's monthly bill — unless regulators specifically carve out an exception. That is the mechanism at the center of the fight: without a rule saying otherwise, a household that will never set foot in a data center can end up helping pay for the power line that feeds one.

The dollar figures are already showing up. Utilities nationwide have sought roughly $31 billion in rate increases^[19]. In the PJM grid region, which covers 13 states and Washington, D.C., regulators have projected rate increases of up to 60% over five years as data centers add demand^[13]. In response, lawmakers in more than 30 states introduced over 300 bills in 2026 covering data center siting, taxes and power rules^[13].

A bipartisan bill in Congress, the Ratepayer Protection Act, would push states toward standards that put those grid costs on the tech companies building the data centers, rather than on households^[15]. Some AI companies have already volunteered similar commitments — Microsoft has pledged to cover its own electricity costs, and Anthropic has promised to cover data-center-related price increases where it operates^[13]. Amazon and its defenders counter that the industry pays for the power it uses and that new infrastructure benefits the whole grid, not just data centers^[13].

Two Very Different Complaints, One Shared Target

Coverage of this earnings report split less along the usual left-right line and more along who the writer was picturing as the reader. Business and market outlets treated the quarter mostly as a stock story. Forbes headlined that Amazon's stock "surges as AWS growth outpaces heavy AI spending" — the word "outpaces" quietly settles the debate in Amazon's favor before the article even begins^[7]. Yahoo Finance called it a "home run" quarter with the stock up 15%, the high end of the reported range^[8].

Benzinga took a different angle, building its story around Jassy's refusal to detail the financing rather than the results themselves — a legitimate angle, though "won't say how it's paying for it" can imply concealment where Amazon may simply be treating it as routine^[9]. On the political right, the populist-leaning Daily Caller focused on state lawmakers and rising power bills, framing tech companies as a cost imposed on ordinary ratepayers^[17].

On the political left, Fortune ran a similar concern from the opposite direction, with the headline "The AI boom's hidden electricity bill — and why you're paying it." The direct address to the reader turns a genuinely contested cost-allocation question into a settled fact about who pays^[14]. That two outlets on opposite ends of the spectrum arrived at nearly the same worry — ordinary people footing AI's power bill — says something the earnings number alone does not.

Outside the U.S., the coverage gap is itself telling. AWS holds roughly 28% of global cloud infrastructure, ahead of Microsoft's 21% and Google's 14% — meaning three American companies run about 63% of the world's cloud computing spending^[18]. Governments in Europe, India and elsewhere read a quarter like this less as a stock story and more as a marker of how dependent their banks, hospitals and public services have become on infrastructure they do not control.

What Gets Built Either Way

Strip away the framing on both sides, and the physical facts remain. The concrete is being poured, the power lines are going in, and the chips are being installed right now, in specific counties, on specific power grids. That does not depend on whether the bulls or the skeptics turn out to be right.

If AI demand arrives on the schedule Amazon's backlog implies, the spending looks exactly like the early AWS buildout: expensive now, dominant later. If demand merely grows slower than expected, the data centers, the transmission lines and the $119.1 billion in debt will still be there, and someone will have to absorb the gap — shareholders, bondholders, or the ratepayers whose bills are already climbing^[10][23].

That question will not resolve this quarter, or probably next year. What is resolving faster is the political one: with the 2026 midterms approaching and utility bills arriving monthly in mailboxes, state legislators in both parties are moving to decide, ahead of time, who pays if the bet does not pay off on schedule^[13][15].