Amazon's $220B Capex Guide, the $496B Backlog, and the Fine Print Inside a Blowout Quarter
AWS delivered its fastest growth in 18 quarters, backlog surged to $496 billion, and advertising accelerated. The capex bill is enormous—but the growth engine is finally catching up.
July 30, 2026 · Amazon Earnings Deep Dive
Amazon’s press release was a very good quarter. The conference call an hour later was where the actual news lived, and none of it is in the filing.
Two numbers to take away. Amazon raised its 2026 capital spending plan to roughly $220 billion, up from about $200 billion, and CEO Andy Jassy blamed the increase on the rising cost of memory chips. And AWS is sitting on a $496 billion backlog of contracted future revenue, growing at triple-digit rates year over year. Neither figure appears anywhere in the earnings release. Read only the filing and you missed both.
The capex raise, and what it costs
Jassy’s words: “We now believe we will spend approximately $220 billion in cash CapEx in 2026. The higher cost of memory pushing this number up from our prior estimate of about $200 billion.”
Notice the reason. Amazon did not attribute the increase to newly found demand, it attributed it to what memory costs. That is a narrow, specific claim, and it is worth separating from two things it is not.
It is not a statement that demand is soft. The opposite: in the same breath Jassy said even $220 billion won’t cover 2026 demand, that 2027 looks the same, and that 2028 demand is “striking.” Strong demand and a cost-driven increase coexist here.
And it is not established that the physical build is unchanged. Amazon gave no cost bridge and no reconciliation of quantities, mix or timing. The one capacity metric on offer is a target to double power capacity by the end of 2027 versus 2025, which Jassy said remains on track, but a power target running eighteen months out is the wrong instrument for detecting a memory-price increase, and a company can hold that target while changing server mix, memory per server, or the timing of spend inside it. So: the raise is cost-attributed, not cost-verified.
Here is the arithmetic nobody did on the call. Amazon spent about $96.3 billion of cash capital in the first half. Hitting $220 billion means roughly $123.7 billion in the back half, call it $61.8 billion a quarter, about 17% above the pace it just ran in Q2. The spending curve gets steeper from here, not flatter.
And Jassy says even that isn’t enough: “Even at that amount, we will still not have enough capacity to meet all the demand we have in 2026, and I believe this dynamic will also be true in 2027. In fact, the demand we already have for 2028 is striking.”
The backlog
$496 billion of contracted AWS business, growing triple digits. One analyst noted the remaining-performance-obligations figure is roughly two and a half times where it stood in the third quarter of last year.
This is the single most important disclosure of the night, and it exists only as spoken words on a conference call. It is the number that makes a $220 billion spending year defensible rather than reckless, Jassy explicitly said the backlog is “all taken into account in our CapEx projections.”
The quarter itself
Revenue of $200.6 billion, up 20%. Operating income of $27.5 billion, up 43%. AWS grew 37% to $42.2 billion, its fastest in eighteen quarters and its fifth consecutive quarter of acceleration. Advertising grew 26% to $19.8 billion, a business now running near $79 billion a year. Headline earnings of $5.75 per share, up 242%.
All of that is real. Some of it is flattered, and it is worth knowing exactly which parts.
The $53.4 billion that isn’t operating income. Amazon disclosed, plainly, that “second quarter 2026 net income includes non-operating pre-tax other income of $53.4 billion, primarily from our investments in Anthropic.” That is a paper mark on a private company’s valuation, not cash from selling anything. It sits behind roughly two-thirds of the quarter’s net income. Strip it out and our rough estimate lands near $1.95 per share rather than $5.75. Amazon didn’t publish that adjusted figure and we’re clear it’s our arithmetic, not theirs, but the gap between $5.75 and something under $2 is the difference between what the headline said and what the business earned.
About $1.2 billion of one-off help inside operating income. Roughly $600 million of tariff refunds, and roughly $600 million from a favorable swing on energy contracts that get marked to market each quarter. The energy piece landed mostly in AWS. Amazon’s own CFO quantified the distortion: AWS margins were up 650 basis points year over year, but only 520 basis points excluding that accounting gain. So the celebrated 39.4% AWS margin is closer to 38% on an operating basis, and it isn’t a record either way, because the first quarter of 2025 printed 39.5%.
Revenue got a calendar gift too. Prime Day landed in the second quarter this year for most large markets. Last year it sat entirely in the third. Amazon quantifies the drag this creates on next quarter’s guidance but never quantifies the boost it gave this one. The boost is real; the size is undisclosed.
None of this makes it a bad quarter. It was an excellent one. It means the reported earnings lines overstate how excellent, while revenue growth, paid units and operating cash flow carry no such asterisk.
Free cash flow went negative, and debt is filling the gap
Trailing-twelve-month free cash flow came in at negative $7.6 billion, down from positive $18.2 billion a year ago, the fifth consecutive decline and Amazon’s first negative reading since 2023. The slide in Amazon’s own deck is still titled “Long-Term Goal – Optimize Free Cash Flows,” which is either institutional inertia or a very dry sense of humor.
The cash-generating machine is fine: operating cash flow was $161.4 billion over twelve months, up 33%. The problem is the bill. Capital spending rose $66.1 billion year over year, which Amazon attributes to artificial intelligence.
Two things the headline numbers hide. Amazon has $29.3 billion of equipment it has acquired but not yet paid for, real commitments that haven’t hit cash flow yet. And it is funding this with debt at serious scale: $67 billion of long-term debt raised in the first half alone, taking long-term debt from $65.6 billion at year-end to $128.9 billion at June 30. It roughly doubled in six months.
Asked directly how the build-out gets funded, Jassy said: “You’ve seen us issue debt this year. We have a lot of options available to us… nothing to share today.” The balance sheet is less coy than he was.
His defense of the spending is genuinely substantive and deserves a fair hearing. Data-center shells absorb capital for about two years before servers go in, then generate revenue for thirty-plus years across five or six generations of hardware. Servers themselves are bought only months ahead of deployment, with demand already visible, “if the demand isn’t there, we won’t spend the capital.” They break even in a little under three years against five-to-six-year useful lives, and most AI capacity is contracted on five-year terms.
He also said the quiet part directly: Amazon will “encounter free cash flow headwinds until these data centers come online, can be monetized, and we get a few years into these servers being utilized.” Nobody should be surprised later.
Asked whether data-center spending might slow in 2027, he didn’t name a year. He talked about how much capacity is already reserved instead.
Two strategy disclosures that weren’t in the filing
Amazon is building its own frontier AI model. Jassy said it outright: “we are pursuing our own frontier model,” motivated by cost control, prioritization and speed. His forecast, within a few years there will be at least half a dozen comparably capable models, all available in Bedrock, “one of them will be ours.”
Trainium may be sold as a standalone chip. Asked about selling Amazon’s AI silicon to third-party data centers, Jassy said customers want it “separate from our cloud,” that Amazon is “actively having those conversations,” and that “I expect there’s a real chance we’ll do that in the future.” That would put Amazon into direct competition with Nvidia as a merchant chip vendor, while, in the same answer, reaffirming a “deep partnership” with Nvidia and noting customers will run Nvidia hardware “for as long as we can foresee.”
One correction to the coverage: OpenAI’s commitment to Trainium is not new. It was announced in February, part of a package that included Amazon investing $50 billion in OpenAI and expanding their compute agreement by $100 billion over eight years. It also explicitly includes Nvidia hardware alongside Trainium, and OpenAI’s largest compute relationship remains with Microsoft. Tonight’s mention was a restatement, not a reveal.
The pattern worth noticing
Three of America’s largest technology companies reported inside nine days, and all three posted earnings materially inflated by paper gains on equity stakes.
Alphabet booked a $99.0 billion gain on equity securities on July 22, adding $6.26 to earnings per share, and never said what the securities were. Microsoft booked $3.2 billion from Anthropic on July 29. Amazon booked $53.4 billion on July 30, the largest such mark any company has explicitly tied to an AI lab.
Only two of the three told shareholders what got marked. And across a full hour of questions from six analysts, not one asked Amazon about the largest non-operating item in its income statement.
What could go wrong
The bear case was already in motion before Amazon reported. Roughly $890 billion came off the largest tech names in the week of July 20-27, with Alphabet’s selloff pinned squarely on its own capital-spending increase. Alphabet’s free cash flow is negative $5.9 billion. Oracle’s is worse. Amazon’s is now negative too. This is starting to look less like one company’s investment cycle and more like a sector paying cash today for revenue it hopes arrives later.
The cost-driven nature of Amazon’s raise cuts against it as well. If components keep getting more expensive while servers take just under three years to break even, the payback math gets tighter. To be fair to management: signed contracts hold their pricing, and new contracts get priced against current costs, which protects new business but leaves the existing backlog absorbing the inflation.
And a valuation mark that adds $53.4 billion at a $965 billion private valuation can subtract on the next round.
Bottom line
The call carried the news, not the filing, a $220 billion spending plan, a $496 billion backlog, a first-party AI model, and merchant chip sales under consideration. None of it in the release.
The reported earnings lines overstate a genuinely excellent quarter. Operating income up 43% is closer to 37% excluding one-off items; the 39.4% AWS margin is nearer 38% and isn’t a record; $5.75 in earnings is closer to $1.95 without the Anthropic mark.
The cash story has quietly become a debt story. Free cash flow is negative, long-term debt doubled in six months, and Amazon’s own CEO says the pressure continues until the data centers earn their keep.
AWS is the biggest cloud, not the fastest-growing one. Its 37% trailed Google Cloud’s 82% and Azure’s 43% this season, off a far larger base, which is the honest frame.
The most useful thing here for anyone outside Amazon: memory. One of the largest technology buyers on the planet just said memory prices are high enough to move a $200 billion plan by 10%, named memory chips in its own risk disclosures, and echoed a complaint Microsoft made in April. That’s confirmed cost pressure running through the biggest buyers, which points at the memory and storage suppliers, Micron, SK Hynix, Samsung, and the drive makers, and lines up with what we wrote about memory supply agreements on July 25. One caution: buyers complaining about prices confirms today’s shortage, not that it lasts. Memory is the most cyclical input in semiconductors, and nothing disclosed here tells you where in that cycle we are.
Disclaimer: This article is for informational and educational purposes only and does not constitute financial, investment, tax, or legal advice. Any opinions, scenarios, price targets, or market observations reflect my personal views and may change without notice. Investing and trading involve substantial risk, including the possible loss of principal. You are solely responsible for your own investment decisions, position sizing, risk management, and trades. Conduct your own research and consult a qualified professional where appropriate.