Amazon closed 15.3% higher on Friday, July 31, at $271.58, and one line in Thursday night's release did all the work. AWS revenue grew 37% to $42.2 billion, and the contracted backlog reached $496 billion. For most of this year the tape has punished AI capital spending. Amazon raised its 2026 capex guide to $220 billion on the same call and got bought anyway. Amazon stock jumped because the backlog showed the spending is already sold, and that is the only capex number the market rewards right now.
The setup in 3 lines:
- AWS backlog hit $496 billion. That is up $132 billion in one quarter, with growth accelerating to 37%.
- The 2026 capex guide went to $220 billion. Roughly $20 billion of that raise is memory-chip inflation, not extra racks.
- Apple reported the same night and beat on revenue. It still fell 7.3%, because the tape pays for capex that converts and sells capex that rents.
On the call Thursday evening, Andy Jassy told analysts that most of the cloud capacity Amazon has planned for 2027 is already reserved by customers, with a slice of 2028 committed too. Then he added that the company will still be short of capacity. That is an odd thing for a chief executive to volunteer. It is also exactly why the stock gapped the next morning.
AWS added $132 billion of contracted backlog in a single quarter
The backlog is the number that moved the stock, so start there. AWS ended the quarter with $496 billion of signed customer commitments, up from roughly $364 billion three months earlier. Revenue growth accelerated for the fifth straight quarter. At 37%, that is the fastest AWS has printed since 2021.
Think of the backlog like a restaurant with reservations booked solid through next year. Every one of those tables is already spoken for. The kitchen just isn't finished yet, so the money shows up as a commitment before it shows up as revenue. That gap between signed and delivered is what the capex is paying to close.
That framing changes what the $220 billion actually is. If the capacity were speculative, the spend would be a bet on demand that may never arrive. Here the demand arrived first, in writing, and the spend is the delivery cost. Investors who spent four quarters marking down every hyperscaler on capex headlines had to reprice that distinction in one session.
$132 billion. That is how much contracted cloud revenue AWS added in three months, and it is the number that turned a capex raise into a 15% rally.
The $62.6 billion net income is mostly a $53.4 billion accounting mark
Read the headline profit and it looks like Amazon earned $62.6 billion in three months. That is not what actually happened. Most of it, $53.4 billion, is a non-cash gain. It came from marking the company's stake in Anthropic to a higher valuation. None of that money is operating cash.
Strip that out and the real result is still good, just human-sized. Net sales came in at $200.6 billion, up 20%, and operating income rose 43% to $27.5 billion. Free cash flow went negative, because $220 billion of annual building does that to a cash flow statement.
Here is the part worth sitting with. The market did not pay for the $62.6 billion. If accounting profit had been the trigger, the stock would have gapped on the profit line, and it didn't: it gapped on cloud growth and backlog. In other words, the tape read straight past the biggest number in the release and priced the third biggest.
Apple's same-night 7.3% drop is the other half of this trade
Apple reported hours before Amazon on July 30. It posted a June-quarter record of $109.4 billion in revenue, up 16%. The stock still fell 7.3% on Friday to $308.91, on 132 million shares. So both companies beat, both traded on similar volume, and they went opposite ways.
The split came from what each company's spending buys. Tim Cook pointed to a chip-supply squeeze and a demand-forecast miss, and Apple guided September-quarter gross margin to 47.5%-48.5%, down from 49.3%. Apple pays for memory and it pays for compute. But it rents the AI infrastructure rather than owning it, so there is no backlog line to show for the money.
Translation: Amazon spends to fill orders it already has, and Apple spends to keep its products working. Both count as legitimate uses of shareholder cash. Only one of them prints a receipt the market can underwrite.
Same input, opposite output. Memory prices are rising for everyone, and we mapped that squeeze when it first showed up in Micron's contract pricing gap and again in the sold-out memory cycle. Amazon absorbed the cost and showed the offsetting revenue. Apple absorbed the cost and guided margins down.
The $220 billion capex raise is memory inflation, not extra racks
This is where the read gets uncomfortable, and it's the part Friday's rally skipped. Amazon's capex guide went up about 10%. Roughly $20 billion of that increase comes from higher prices for memory, drives and SSDs. Jassy said as much on the call. So the extra money mostly buys the same hardware at a worse price.
Now put that next to how AWS sells. Most large cloud deals are signed at set prices for the full contract duration, which Jassy also confirmed. That means the revenue side of the $496 billion is largely locked, while the cost side keeps floating with a memory market that has been tightening all year. New contracts get priced at current costs. The existing ones can't.
We're not going to pretend to know how that resolves. Our read is that AWS margins, at 39.4% this quarter, are more likely to compress in 2027 than in 2026. The locked contracts signed cheap still have to be delivered on expensive silicon. The numbers for 2026 already look safe. It's the following year that carries the risk, and nobody on Friday was pricing 2027.
39.4%. The AWS operating margin that now has to survive a memory market it signed contracts ahead of.
Amazon erased a three-month correction in one session
Now the structure, because the tape tells a cleaner story than the release does. Amazon topped at $278.56 in early May. Price then spent eleven weeks working lower in a choppy, overlapping decline that bottomed at $225.55 in late June. It bounced to $254.96 by mid-July, faded again, and put in a second low at $226.16 on July 29.
Those twin lows, roughly sixty cents apart across five weeks, are the whole technical case. A corrective decline that ends on matched lows and then gaps above its entire range is behaving like a completed correction, not a continuation. And the gap came on 128.9 million shares against a three-month average near 47 million.
The May high sits just overhead, which is the honest complication here. Friday's session stopped at $273.23, a couple of percent short of that level. So the first real test is whether buyers can absorb the supply parked at the prior peak. Gaps of this size and volume usually get retested within a few weeks, and how that retest behaves matters more than Friday's close did.
Amazon is one of the assets on the Kunkel Capital rotation: members get the full structure map with entry, exit and invalidation refreshed on a fixed cycle. That is the layer this post deliberately doesn't publish.
For the pattern behind that retest, our gap-and-shelf work on Microsoft covers how earnings gaps tend to build a base. The post-earnings drift study covers how long the move usually keeps running after the first session.
Five things Friday's gap actually settled
- Capex is no longer a sell signal by itself. The market now sorts spending by whether it has contracted revenue behind it.
- Backlog is the new scoreboard. AWS at $496 billion gave the bulls a number they can hold up against every capex complaint.
- Memory cost is a real 2027 margin risk. Locked contract pricing plus floating input costs is a squeeze, not a rounding error.
- Accounting profit doesn't move mega-caps. A $53.4 billion mark-to-market gain got ignored in favour of a growth rate.
- Same-night comparisons are the cleanest tell. Apple and Amazon shared the macro, shared the memory bill, and split by 22 points of return.
The rest of the hyperscaler group is running the same experiment. Four of them together have now guided to something like $720 billion to $745 billion of 2026 capital spending. That is why the Alphabet capex selloff and the Oracle sell-the-news top both turned on backlog disclosure rather than on spending totals. When a company shows the bookings, it gets paid. When it shows only the invoice, it gets sold.
Where this thesis breaks
The read is simple enough. Amazon's gap is a re-rating of AI spending that arrives with signed demand attached, and the structure supports continuation while the market keeps paying for conversion over raw capacity. The macro backdrop is not helping, though. Long yields backed up hard last week. The 30-year sits at its highest since 2007, and long-duration growth names don't love that.
So here's where we're wrong. If Amazon closes a full day back below the July 30 close of $235.50, the gap has failed and the twin-low read goes with it. A softer version of the same signal: price holds the gap but AWS backlog growth decelerates sharply next quarter, which would say the $496 billion was a pull-forward rather than a trend. Either one kills this thesis.
That is the discipline the whole approach runs on. We publish what is happening and why it matters, and we name the behaviour that would prove us wrong. For the wave context behind a move like this, our note on wave 3 extensions covers what usually follows a gap out of a completed correction.
Frequently asked questions
Why did Amazon stock jump 15% on July 31, 2026? AWS revenue grew 37% to $42.2 billion. The contracted backlog reached $496 billion in signed commitments. That told investors the AI capital spending is already matched by customer demand.
What is the AWS backlog and why does it matter? It's the total value of cloud contracts customers have signed but not yet consumed. It matters because it turns capex from a bet on future demand into the cost of delivering booked revenue.
Why did Apple fall while Amazon rallied on the same night? Apple beat on revenue but guided September-quarter gross margin lower on component costs. It also has no equivalent backlog to show for its AI spending, so the market paid for conversion and sold the rest.
Is Amazon's $62.6 billion profit real? The operating income of $27.5 billion is. Most of the headline figure is a $53.4 billion non-cash gain from revaluing Amazon's Anthropic stake, and it has no cash effect.
What is the main risk to the bull case? Memory prices are the main risk. AWS sells at fixed contract prices while its input costs keep rising, so the margin pressure lands in 2027 rather than in this year's numbers.
Know your entry, your exit, and where you are wrong on Amazon
Friday's 15.3% gap and the $496 billion AWS backlog are the surface signal, and every free chart shows you the same two lows in late June and late July. Amazon is on the Kunkel Capital watchlist: the full research maps the current wave count to a defined entry zone, an exit target and the exact invalidation level, refreshed on a fixed rotation, with alerts when levels hit. €19.99 first month, then €34.99. Cancel anytime.
Sources: Amazon Q2 2026 earnings release and SEC filings, Amazon Q2 2026 earnings call (July 30, 2026), Apple Q3 FY2026 earnings release (July 30, 2026), Reuters, Bloomberg.
Last updated: 2026-08-01
