Amazon Gives Back the Gap as $220B of Spending Overwhelms a 39.4% AWS Margin

Amazon traded at $256.45 on Tuesday, down $2.06 or 0.80% from Friday's $258.51 close, after opening at $259.07 and running to $261.12 in the first hours of the session. The low printed $255.29. Volume ran 30.75 million shares against a 32.01 million daily average.

That intraday shape is the whole story. The stock gapped higher on the reopen after the Labor Day break, extended roughly 1% above Friday's close, and then reversed the entire move plus some. A gap that fills inside a single session is a market rejecting the higher price rather than accepting it.

The 52-week range spans $196.00 to $287.20. Amazon's all-time closing high was $284.02, set on August 3, 2026. Tuesday's print sits roughly 9.7% below that high and 30.8% above the twelve-month low.

The recent trajectory has been consistently negative. The stock is down 5.8% over the past 30 days and roughly 11% from the August peak. August closed at $259.77, an 8.54% decline for the month. Year to date, Amazon has returned about 10% against a 12% advance for the S&P 500 (SPX) — a rare stretch of underperformance for a company posting 20% revenue growth.

The broader tape offered no help. The Dow Jones Industrial Average (DJI) shed 570.78 points to 52,843.47 while the Nasdaq Composite (IXIC) gave up 0.27% to 26,435.53, with capital rotating hard into semiconductors and physical infrastructure. Intel (INTC) ripped 8.16% to $103.61 and Coherent (COHR) gained 10.73% on the same session that Amazon faded.

That rotation is the relevant context. Money is moving toward companies selling picks and shovels into the AI buildout, and away from the companies writing the cheques.

Amazon is writing the largest cheque in corporate history.

Down 11% From the High and Sitting Below the 100-Day Average

The technical picture deteriorated in late August and has not repaired.

The stock finished a recent session below its 100-day moving average, a level it had held through most of 2026. Losing a long-horizon average after an 11% decline from the high converts what looked like consolidation into a genuine correction, and it removes the most-watched dynamic support on the chart.

The structure now runs as follows. Immediate resistance is $261.12, Tuesday's high and the level at which sellers appeared. Above it, $265 marks the zone where the stock traded before the FTC news, and $284.02 is the all-time closing high with the $287.20 intraday peak just above.

Support begins at $255.29, Tuesday's low. Then $250, a round number that has not been tested since the post-earnings gap. The July 24 low of $232.01 is the next structural reference, and $230 sits at the bottom of the sell-side target range — the level the most bearish published forecast identifies.

The distances matter. From $256.45, the all-time high is 10.7% up. The $232.01 July low is 9.5% down. The pair is roughly symmetrical, which is unusual and reflects a market with no positioning skew.

Volume is the detail that argues against a capitulation reading. At 30.75 million shares against a 32.01 million average, Tuesday's reversal happened on below-normal turnover. That is drift rather than distribution, and it suggests the sellers driving the recent decline have largely finished.

The post-earnings move provides a useful anchor. Shares surged 9.15% after-hours to $257.04 on the July 30 report. Tuesday's $256.45 sits below that immediate post-print level, meaning the entire earnings reaction has been erased and then some.

Sell-side positioning remains uniformly constructive despite the price action. The average twelve-month target sits at $328.17, with a high estimate of $405 and a low of $230. Fifty-nine analysts recommend buying and none recommend selling.

A stock trading 22% below its consensus target with zero sell ratings and an 11% drawdown is a market disagreeing with the analyst community about something specific. That something is capital expenditure.

The Miami Crash: Flight 7598, a 32-Year-Old Airframe, and 21 Air

The most immediate news event is also the one least likely to matter financially, and separating those two facts is important.

A Boeing 767-300 operating under Amazon's Prime Air branding as Flight 7598 overran Runway 30 at Miami International Airport at approximately 2 p.m. local time on Sunday, September 6, after arriving from Luis Muñoz Marín International Airport in San Juan, Puerto Rico. The aircraft struck several vehicles and came to rest in a field beside the airfield, catching fire.

Five people were killed. Five others were injured, three of them critically. Emergency services deployed roughly 60 fire apparatus and nearly 200 personnel to manage the wreckage, rescue those trapped and contain a fuel leak. The incident forced a roughly three-hour ground stop at one of the country's busiest airports and disrupted more than 450 arrivals and departures by Sunday night.

The National Transportation Safety Board has opened an investigation with Federal Aviation Administration assistance, and the probe is scrutinising aircraft speed, weather conditions and the absence of an airport safeguard. Findings are published at ntsb.gov.

Two structural details govern the financial read. The airframe was 32 years old, built as a passenger jet and converted to a freighter in 2015. And Amazon did not operate the flight. 21 Air, a Miami-based cargo carrier, holds the operating certificate, employs the crew and carries the insurance. Amazon supplied the branding and the freight, and owns some of the 767s that 21 Air flies.

That structure limits direct liability substantially. The closest historical comparable produced a limited and short-lived share price reaction.

The genuine investor concern is second-order. Amazon Air has become central to the company's ability to move goods quickly without depending entirely on outside carriers, which makes reliability strategically important. If the inquiry produces inspections, operating limits or modifications to the air network, that carries incremental cost and delay.

A single accident is unlikely to affect the profitability of a logistics operation of this scale. Scrutiny of an ageing contracted freighter fleet across the whole network is a different matter, and that is the tail the stock is carrying.

The FTC Suit: Twenty-Two States and a $20 Billion Claim

The more consequential legal development landed a week earlier and goes directly at a business generating $19.8 billion a quarter.

The Federal Trade Commission, joined by 22 state attorneys general from both parties, filed suit alleging that Amazon's advertising practices overcharged its roughly 1.2 million advertisers by $20 billion from 2019 to the present. The claim centres on how the company sets prices in sponsored-product ad auctions on its marketplace, with regulators alleging that Amazon began participating in those auctions with its own "soft reserve" bid in 2018.

The stock fell roughly 2.8% to 3% on the filing.

The mechanics of the allegation matter for assessing the risk. A soft reserve bid is a floor placed into an auction by the auction operator. If Amazon was bidding into auctions it also ran, sellers paid more than a purely competitive clearing price would have produced. Whether that constitutes an antitrust violation or a standard reserve-price mechanism is the legal question, and it will take years to resolve.

The financial exposure is what investors have to size. Advertising revenue grew 26% year over year to $19.8 billion in the June quarter, making it an annualised business approaching $80 billion. It is also among the highest-margin revenue lines Amazon operates, because it monetises traffic the company already has.

A $20 billion damages claim is material but not existential against a company generating $200.6 billion of quarterly revenue. The structural risk is larger than the damages figure: a remedy that constrains how Amazon prices its own auctions would compress the margin on the fastest-growing high-margin segment in the business.

That sits alongside broader antitrust exposure and EU scrutiny of the cloud business. Case filings are published at ftc.gov.

The regulatory overhang is the second of the two factors driving the recent pullback. The first is capital expenditure, and it is by far the larger of the two.

Q2 by the Numbers: $200.6 Billion and 43% Operating Income Growth

The operating results underneath the share price decline were exceptional, which is what makes the current setup interesting.

Second-quarter revenue reached $200.6 billion, up 20% year over year from $167 billion and up from $181.5 billion in the first quarter. Growth was 20% in constant currency, and the result beat the high end of company guidance on both the top and bottom lines.

Operating income rose 43% year over year to $27.5 billion. Operating margin expanded to 13.7% from 11.4% — a 230 basis point improvement on a revenue base growing 20%. That combination of accelerating revenue and expanding margin is the hardest thing for a company of this size to produce.

Net income for the three months to June 30 came in at $62.6 billion per the quarterly filing, a figure inflated by non-operating items but nonetheless enormous. Filings are available at ir.aboutamazon.com.

Third-quarter revenue guidance runs $197 billion to $202 billion. That midpoint of $199.5 billion implies a sequential deceleration from $200.6 billion, but the comparison is distorted by the timing shift of Prime Day into the second quarter, which pulled revenue forward and creates noise in modelling.

Half-year revenue therefore stands at $382.1 billion. Adding the third-quarter midpoint puts nine-month revenue near $581.6 billion, with a full year approaching $790 billion to $800 billion depending on holiday-quarter strength.

Segment detail shows breadth. North America revenue rose 16%. Advertising grew 26% to $19.8 billion. Grocery and everyday essentials produced perishable customer growth of 50% since the start of the year, with same-day delivery now covering 2,300 cities.

Earnings per share growth ran 89.5%, and the stock has returned 12% against 0.9% for its sector on a trailing basis.

None of that is a company in trouble. It is a company whose market has stopped paying for growth and started asking about the cost of producing it.

AWS at $42.2 Billion and 36.7% — the Fastest in Eighteen Quarters

The cloud segment is where the entire investment case now lives.

AWS revenue reached $42.2 billion in the June quarter, up 36.7% year over year. That was the fifth consecutive quarter of accelerating growth and the fastest rate in eighteen quarters. The annualised run rate stands at $169 billion.

Acceleration is the operative word. A business of that scale growing faster than it did a year ago, four quarters in a row, is behaving unlike anything in the history of enterprise software. Large numbers normally decelerate. AWS has done the opposite through 2025 and 2026.

The driver is AI workload migration, and the pull-through effect is broader than accelerator demand alone. Post-training, reinforcement learning and agentic tool use run substantially on CPUs, which pulls core cloud services along with the AI-specific consumption. Management has cited Graviton price performance of 30% to 40% better than alternatives as the reason that CPU-side pull-through lands with Amazon rather than elsewhere.

The capacity constraint is the limiting factor on how fast this can grow. Management stated explicitly that the company will not have enough capacity to meet all demand in 2026 and 2027, which caps the growth rate below what demand alone would support.

That is a genuinely unusual disclosure. A company telling investors it is turning away revenue because it cannot build fast enough is describing a demand environment that no forecast model captures, and it is the direct justification for the capital expenditure programme.

Executives noted on the July call that 2027 and 2028 capacity is already reserved, which extends demand visibility well beyond the normal planning horizon for a cloud business.

The risk sits in the same sentence. Capacity reserved is revenue that has not yet been recognised, delivered or paid for at scale, and reservations can be renegotiated if the AI capital cycle turns.

A 39.4% AWS Margin Is the Number That Changed the Thesis

The single most consequential figure in the June quarter was a margin, not a growth rate.

AWS generated $16.6 billion in operating income, a 64% increase from $10.2 billion a year earlier. The operating margin reached 39.4%, up from 32.9% — a 650 basis point year-over-year expansion. A $600 million derivative gain contributed to the figure.

Margin expansion of that magnitude occurring simultaneously with the largest capital expenditure increase in the company's history is the most important evidence available that the custom silicon strategy is producing real financial results at scale.

The mechanism is straightforward. Rather than buying accelerator capacity at market prices that reflect a supplier's own margin, Amazon builds a meaningful share of its compute in-house. Every workload that runs on Trainium instead of purchased silicon captures margin that would otherwise leave the company.

That is the difference between Amazon's capital expenditure and a pure infrastructure buildout. A company spending $220 billion on third-party hardware is transferring value to its suppliers. A company spending $220 billion where it controls the cost at the silicon level is building an asset with a structurally lower cost basis.

The margin trajectory is what turns the capex from a liability into an investment case. If AWS can hold near 39% while growing 37%, the incremental operating income from each dollar of new capacity is enormous, and the payback period on the infrastructure shortens accordingly.

Efficiency gains and capacity optimisation were cited alongside the silicon effect as drivers.

The counterargument is that 39.4% may not be sustainable. Margin at that level in a segment simultaneously absorbing record depreciation from new data centres implies either that the new capacity is not yet in service — and therefore not yet depreciating at full run rate — or that pricing power is unusually strong. Both conditions can change.

Depreciation from $220 billion of 2026 capital spending will flow through the income statement over subsequent years. That is the mechanical headwind sitting in front of the margin.

The $496 Billion Backlog Is 2.9 Times the AWS Run Rate

The number that justifies the spending, if anything does, is the contracted backlog.

Remaining performance obligations stood at $496 billion, up triple digits from a year earlier and up $132 billion in a single quarter. That is contracted work not yet delivered — commitments customers have signed rather than intentions they have expressed.

Set that against the AWS annualised run rate of $169 billion and the ratio is 2.9 times. Amazon has under contract nearly three years of current-run-rate cloud revenue before writing a single new deal.

The quarterly increment is arguably more informative than the level. Adding $132 billion of backlog in three months, against $42.2 billion of quarterly revenue recognised, means the company signed roughly three dollars of new commitment for every dollar it delivered.

That is the metric that separates this capital cycle from a speculative one. The infrastructure investment is supported by actual customer commitments rather than by extrapolated demand curves.

The two largest named commitments come from AI labs. Anthropic and OpenAI have each made multi-year, multi-gigawatt capacity commitments on Trainium, with one commitment running to as much as 5 gigawatts of future capacity. Deals of that size are negotiated in gigawatts rather than dollars, because power rather than silicon is the binding constraint.

The qualification is concentration. A backlog whose largest components sit with a small number of AI labs carries counterparty risk that a diversified enterprise backlog does not. Those labs are themselves funded by capital markets, and their ability to honour multi-year power reservations depends on continued access to funding.

Amazon is also actively discussing selling Trainium chips to customers separately from AWS capacity, which would diversify the revenue base away from pure cloud consumption.

Backlog is visibility, not cash. It tells you the revenue is contracted. It does not tell you when it converts or at what margin.

$220 Billion of Capex and Free Cash Flow at Negative $7.6 Billion

This is the number the market is choking on, and the arithmetic explains why.

Amazon raised its 2026 capital expenditure guidance to approximately $220 billion, up from the roughly $200 billion it had held since February and up 76% from about $125 billion in 2025. Management attributed the increase primarily to higher memory costs, and stated that even at $220 billion the company will not have enough capacity to meet demand.

Second-quarter capital expenditure alone reached $54.2 billion against $32.1 billion a year earlier — a 69% increase in a single year.

The cash flow consequence has already arrived. Trailing twelve-month free cash flow swung to an outflow of $7.6 billion from an inflow of $18.2 billion a year earlier. That is a $25.8 billion deterioration in the company's cash generation, driven entirely by the build.

Financing has followed. Amazon has raised more than $31 billion in new bond and loan financing to support the programme, and is expanding its physical footprint with commitments including a $10 billion data center campus in Missouri.

The forward path is where the estimates get uncomfortable. One published projection puts 2027 capital expenditure at $320 billion and 2028 at $370 billion, with negative free cash flow of approximately $50 billion in each of those years.

If those figures are close to correct, Amazon would spend roughly $910 billion across three years and generate negative free cash flow of approximately $107 billion cumulatively. That is a scale of capital commitment without precedent in corporate history, funded from a combination of operating cash flow and debt.

Against roughly $790 billion to $800 billion of estimated 2026 revenue, $220 billion of capex represents about 27.5% of the entire top line going into the ground. For context, that ratio would be aggressive for a utility and is unheard of for a company that still derives the majority of its revenue from retail.

The bull framing is that this is an infrastructure build ahead of a massive demand cycle, validated by a $496 billion backlog. The bear framing is that if AWS growth slows materially, the market's focus shifts instantly from the revenue opportunity to the capital requirement.

Trainium, Graviton and Two Separate $25 Billion Run Rates

The custom silicon business has become material enough to analyse on its own terms.

Amazon disclosed two milestone figures in the June quarter. Its AI revenue run rate and its custom chips business — Trainium for AI training and inference, Graviton for general-purpose compute — each exceeded $25 billion on an annualised basis, with both growing at triple-digit percentage rates year over year.

A $25 billion chip business growing triple digits would be a significant standalone semiconductor company. Embedded inside Amazon, it functions as a cost-reduction engine for AWS and as an increasingly credible external product.

Graviton's claimed 30% to 40% price-performance advantage over alternatives is the mechanism pulling CPU workloads onto Amazon's infrastructure as AI moves from training toward inference and agentic systems. That architectural shift favours CPUs proportionally more than the training phase did, and it is the reason AWS growth has accelerated rather than plateaued.

Trainium's position is more contested. Enterprises whose AI workloads are built around the dominant accelerator ecosystem face genuine porting costs to migrate, and that friction is why the incumbent's hold on AI compute spending has not collapsed despite Amazon's cost-advantage claims.

For customers already running on Amazon's model platform or building new AI workflows from scratch, the friction is minimal. That is the addressable segment, and it is growing faster than the installed base of legacy workloads.

The strategic logic of the $220 billion is clearest here. Nearly all of it funds data centers, power capacity and AI chips — infrastructure where Amazon controls the cost at the silicon level rather than paying a supplier's margin.

Amazon is running two AI infrastructure strategies simultaneously: purchasing more third-party accelerators than almost any other organisation while building its own silicon to reduce that dependence over time. Those strategies are complementary in the short run and substitutes in the long run.

Product expansion continues alongside the infrastructure. The company has launched an AI security product built on frontier models that ingests existing vulnerability backlogs, prioritises by business context and validates fixes in a sandbox, and is rolling out consumer AI products including an expanded assistant in new markets.

Retail and Advertising Are Funding the Build

The segments that generate cash rather than consume it deserve attention, because they are what makes the capital programme financeable.

Advertising revenue grew 26% year over year to $19.8 billion, driven by engagement in streaming video advertising and live sports, plus AI-powered tools improving advertiser performance. Annualised, that is a business approaching $80 billion at margins substantially above the corporate average.

North America revenue rose 16%. The grocery and everyday essentials business showed perishable customer growth of 50% since the start of the year, and same-day delivery has expanded to 2,300 cities.

Those numbers describe a retail operation that is still gaining share and still improving its delivery economics after three decades. The 13.7% consolidated operating margin, up from 11.4%, reflects retail leverage as much as cloud margin.

The relationship between the segments is the underappreciated part of the story. Retail and advertising generate the operating cash flow that funds a substantial portion of the AWS build without requiring proportionate debt issuance. A pure-play cloud provider attempting a $220 billion programme would need a fundamentally different balance sheet.

That is also why the FTC advertising suit carries strategic weight beyond its $20 billion damages claim. Advertising is the highest-margin funding source for the infrastructure programme, and a remedy constraining auction pricing would reduce the cash available to build.

Same-day delivery expansion and grocery growth also matter for the AI thesis in an indirect way. AI-driven logistics and warehouse robotics are among the internal applications of the same infrastructure, which means a portion of the capital spending produces returns inside retail rather than only through AWS.

Amazon operates three reportable segments — North America, International and AWS — and the market currently values it almost entirely on the third.

Valuation: 27.5% of Revenue Going Into the Ground

The valuation debate reduces to a single question: what multiple do you pay for revenue that requires 27.5% of itself in annual capital expenditure to grow?

At $256.45, with roughly $790 billion to $800 billion of 2026 revenue in prospect, Amazon trades at a modest multiple of sales by megacap technology standards. On earnings, the picture is distorted — operating income of $27.5 billion in a single quarter annualises well, but depreciation from the current build has not yet flowed through.

The sell-side consensus sits at $328.17, implying roughly 28% upside from the current print. The range runs from $230 at the low to $405 at the high — a 76% spread that honestly reflects how binary the capital cycle outcome is.

Fifty-nine analysts recommend buying. None recommends selling. That unanimity coexisting with an 11% drawdown from the all-time high indicates the price decline is being driven by fund flows and factor rotation rather than by fundamental downgrades.

The bear case is precise rather than vague. Amazon is committing $220 billion in 2026 with free cash flow already at negative $7.6 billion on a trailing basis and projections of negative $50 billion in each of 2027 and 2028. The valuation therefore depends on investors believing the infrastructure spending produces substantially higher cash flows later. That creates a higher bar for execution than a company generating cash while growing.

The bull case is equally precise. The $496 billion backlog makes at least part of the future revenue pipeline visible. AWS operating margin expanded 650 basis points to 39.4% while capex was rising. Two separate $25 billion businesses are growing triple digits. Capacity is reserved into 2028.

The pivot point is AWS growth. If it holds near 37% with margins near 39%, the $220 billion looks like an aggressive build ahead of a demand cycle. If growth slows materially, attention shifts to the capital requirement within a quarter.

For now the numbers put Amazon on the right side of that trade. The company is spending heavily because demand exceeds capacity, not because it is chasing a hypothetical market.

Competition: The CUDA Moat and the Hyperscaler Race

Amazon competes on two fronts that behave differently, and its position is stronger on one than the other.

In cloud infrastructure, AWS growing 36.7% with a $169 billion run rate and a $496 billion backlog is not a company losing share. Acceleration for five consecutive quarters against competitors of comparable scale indicates AWS is capturing at least its proportional share of AI workload migration and probably more.

In AI silicon, the position is harder. The incumbent accelerator ecosystem retains a substantial software moat, and enterprises with workloads built around it face porting costs to migrate. That friction is real and it is why the incumbent's hold on AI compute spending has not weakened as fast as price-performance comparisons would suggest.

Nvidia (NVDA) fell 1.46% to $227.00 on Tuesday against a $5.481 trillion market capitalisation, while Intel rose 8.16% and the optical and interconnect complex ran double digits. That rotation within AI — away from the accelerator leader and toward suppliers and second sources — is directionally favourable for Amazon's silicon strategy, because it reflects the market pricing more competition in compute rather than less.

The broader risk that emerged this week is not competitive at all. Data center opposition is becoming a recognised constraint on AI infrastructure, with local resistance to new campuses adding permitting and siting risk to every hyperscaler's expansion plan. Power availability rather than capital is increasingly the binding limit, which is why capacity deals are now negotiated in gigawatts.

Amazon's counter is vertical integration. Owning the silicon, the data centers and the power contracts gives it control over more of the cost stack than any competitor that buys compute at market prices.

The financing environment is the other emerging risk. Debt funding the AI buildout has drawn increasing scrutiny, and Amazon's $31 billion of new bond and loan issuance places it inside that conversation. At a 4.80% ten-year Treasury yield and a 5.27% thirty-year, the cost of financing a multi-year infrastructure programme is materially higher than it was when the programme was conceived.

Verdict: Exceptional Operations, an Unprecedented Bill, and a Market That Has Stopped Paying

Amazon at $256.45 is a company whose fundamentals accelerated and whose stock did not, and the gap between those two facts is the entire analysis. The June quarter delivered $200.6 billion of revenue at 20% growth, operating income up 43% to $27.5 billion, margin expansion to 13.7% from 11.4%, AWS revenue of $42.2 billion growing 36.7% for a fifth consecutive quarter of acceleration and the fastest rate in eighteen quarters, an AWS operating margin of 39.4% up 650 basis points, advertising up 26% to $19.8 billion, and a contracted backlog of $496 billion that grew $132 billion in three months — nearly three times the $169 billion AWS run rate. Against all of that, the stock is down 11% from its $284.02 August 3 high, down 5.8% in a month, sitting below its 100-day average, and trading 22% beneath a $328.17 consensus target that no analyst has attached a sell rating to. The reason is the bill: $220 billion of 2026 capital expenditure, up 76% from $125 billion in 2025, representing roughly 27.5% of estimated full-year revenue, with trailing free cash flow already at negative $7.6 billion against a positive $18.2 billion a year ago, more than $31 billion of new debt raised, and published projections of $320 billion in 2027 and $370 billion in 2028 with negative free cash flow near $50 billion in each. Layer on an FTC suit joined by 22 states alleging $20 billion of advertiser overcharges since 2019 against the highest-margin funding source in the business, and a fatal Prime Air runway overrun in Miami on a 32-year-old airframe operated by 21 Air rather than Amazon. The path from here: hold $255.29 and reclaim $261.12 and the stock has a clear run at $265 with $284.02 as the target, roughly 3% to 10.7% of upside. Lose $255.29 and the sequence runs $250, then the $232.01 July low, then the $230 bottom of the target range — 9.5% to 10.3% down. Setup is constructive on operations and unresolved on capital allocation, and the honest read is that this stock does not re-rate until either free cash flow inflects or AWS growth proves it can hold 37% through a full depreciation cycle — with Tuesday's gap-and-fade on below-average volume suggesting the market has not yet decided which way that goes.

That's TradingNEWS