Amazon Is Spending $200 Billion on AI. Its Free Cash Flow Just Fell 95%.

For two decades, the bull case on Amazon rested on a single, almost magical fact: the company could plow
staggering sums back into its own growth and still throw off rising mountains of cash. That fact just broke. Over
the trailing year, Amazon's free cash flow collapsed from roughly $26 billion to about $1.2 billion — a 95%
drop — and in the first quarter of 2026 it went outright negative, to the tune of eighteen billion dollars in
the red. The cause is a capital-spending plan of around $200 billion for the year, almost all of it aimed at
artificial intelligence. And the entire wager rests on one division — AWS — that funds nearly the whole company's
profit while growing slower than the rivals it is spending all that money to outrun. This is the anatomy of the
most expensive bet in Amazon's history, made to defend the one engine it cannot afford to lose.

---

Begin with the number that the market, strangely, waved through. In the first quarter of 2026 Amazon reported
quarterly capital expenditure of roughly $44 billion, and that single quarter of spending flipped the company's
free cash flow to negative $18.2 billion. On a trailing-twelve-month basis, free cash flow — the actual cash
a business generates after funding its own investment, the truest measure of whether a company is a money
machine or a money furnace — fell from about $25.9 billion to $1.2 billion. That is a 95% collapse in the
metric that, more than any other, justified Amazon's valuation for twenty years. And the stock barely flinched.

There is a reason it barely flinched, and the reason is itself the story. The market has decided, collectively,
that this is not a deterioration but an investment — that Amazon is doing what Amazon has always done, spending
aggressively today to own an enormous tomorrow, and that the cash will come roaring back once the AI build-out
matures. That interpretation may be correct. It is also, word for word, the interpretation that every company in
history has offered for every cash-incinerating capital binge, the true ones and the disastrous ones alike, and
the job of a forensic reader is to look underneath the reassurance at what is actually being bet, and on what.

What is being bet is around $200 billion of capital expenditure in 2026 — for AI data centers, custom
silicon, robotics, even a satellite constellation — and what it is being bet on is the continued dominance of
Amazon Web Services. This essay is about why that concentration matters, why the growth math is more worrying
than the headlines suggest, and why the depreciation bill from $200 billion of spending is a time bomb planted
inside the one number that makes Amazon work.

One engine pulls the whole train

To understand the stakes you have to understand a fact about Amazon that its sprawling retail empire is very good
at obscuring: the company barely makes money selling you things. The vast e-commerce operation — the
warehouses, the trucks, the two-day shipping, the everything store — runs on famously thin margins, sometimes
barely above breakeven, occasionally below. The profit that makes Amazon one of the most valuable companies on
Earth comes overwhelmingly from one place: AWS, the cloud-computing division.

The first quarter of 2026 makes the dependence vivid. Amazon's total operating income was about $23.9 billion.
AWS's operating income was roughly $14.16 billion of it. A single division, representing a minority of the
company's revenue, generated the majority of its profit — and in many recent quarters AWS has accounted for an
even larger share, sometimes essentially all of the company's operating income while retail hovered near zero.
AWS is not a business unit at Amazon. It is the business. Everything else — the Prime memberships, the
advertising, the devices, the groceries, the logistics — is in some sense scaffolding around the cloud division
that pays for it.

This is why the AI capital binge is so consequential and so fraught. Amazon is not spreading $200 billion across
a diversified set of profit engines. It is pouring the overwhelming majority of it into AWS and the AI
infrastructure that lives inside AWS — which means the company is concentrating its single largest-ever bet on
the single division that already carries the entire enterprise. If AWS thrives, the spending was visionary. If
AWS stumbles, there is no second engine to pull the train, because there never was one. The whole machine has
one cylinder, and Amazon just bet two hundred billion dollars that it keeps firing.

The growth gap nobody wants to discuss

Here is where the official narrative and the underlying data begin to diverge, and the divergence is the heart
of the forensic case. The headline from Amazon's quarter was genuinely good: AWS revenue grew 28% year over
year, to about $37.6 billion, its fastest growth in roughly four years — what the bulls called "escape
velocity." Twenty-eight percent on a business at a $150-billion-plus annual run rate is a remarkable number, and
nobody should pretend otherwise.

But growth rates do not exist in a vacuum. They exist relative to competitors, and relative to competitors AWS is
losing the race even as it accelerates. In the same period, Microsoft reported that its Azure cloud revenue grew
40%. Alphabet reported that Google Cloud grew 63%. Set the three side by side and the picture inverts:
AWS, the largest cloud at roughly 30% of global infrastructure spend, grew at 28%; Azure, at about 25% share,
grew at 40%; Google Cloud, at about 13% share, grew at 63%. The market leader is growing slowest, and its two
largest rivals are growing one and a half to more than two times faster. That is the signature of a leader
gradually ceding share — not collapsing, not in crisis, but steadily losing ground at the margin to hungrier
competitors, quarter after quarter.

Now layer the spending on top of the growth. Amazon is deploying around $200 billion in capital — an
unprecedented sum — and the return on that spending, in the one metric that matters, is cloud growth that still
trails Azure and Google Cloud by a wide margin. The bull says the spending is what produced the
re-acceleration to 28% and will push it higher. The skeptic notes that the rivals are spending heavily too and
growing faster, which suggests AWS's slower growth is not a spending problem that more capex fixes but a
competitive-position problem — that Microsoft's enterprise lock-in and Google's AI-research halo are pulling AI
workloads toward Azure and GCP in ways that AWS, for all its scale and all its billions, is struggling to match.
If that is true, then $200 billion buys Amazon the privilege of running very hard to stay roughly in place.

The depreciation time bomb

Now the mechanism that turns this from a strategy question into an earnings risk, because capital spending does
not vanish when the check clears. It lands on the income statement, slowly, for years, as depreciation — and
depreciation eats directly into the AWS operating margin that funds the entire company.

Here is the dynamic. When Amazon spends $200 billion building data centers and stuffing them with chips, that
asset base must be depreciated over its useful life — and as AI hardware ages faster and the asset base swells,
the annual depreciation charge climbs relentlessly. That charge is a real expense that reduces operating income.
AWS's operating margin has been extraordinary — around 35% in recent quarters, a level of profitability that
is the envy of the industry and the source of Amazon's profit dominance. But that margin is now facing a visible,
growing headwind from AI-related depreciation, and the bigger the capex, the heavier the future depreciation, and
the greater the downward pressure on the very margin that makes AWS the company's profit engine.

This is the quiet trap inside the AI boom for every hyperscaler, and it is sharpest at Amazon because Amazon
depends on AWS margins more than its rivals depend on any single line. Microsoft has Office and Windows; Google
has search advertising; both have enormous, high-margin profit pools outside the cloud. Amazon has AWS, and AWS
has a 35% margin that the company's entire profit structure leans on. If $200 billion of annual capex, depreciated
over coming years, compresses that margin by even a handful of points — and depreciation of that magnitude
mechanically must press on it — then the profit that funds all of Amazon shrinks at the exact moment the company
has bet everything on expanding it. The spending is meant to grow AWS. Its accounting shadow may quietly shrink
AWS's profitability for years, and the market is paying almost no attention to the lag.

The second engine that isn't big enough yet

The reflexive rebuttal to the one-cylinder worry is that Amazon now has a second high-margin engine: advertising.
And it is true — Amazon's advertising business has grown into a genuine giant, tens of billions of dollars a year
of high-margin revenue from sponsored products and placements across its retail surface, and it is one of the
most encouraging developments in the company's recent history. A bull will say the profit base is diversifying,
that AWS is no longer the whole story, that the concentration risk this essay describes is yesterday's company.

There is something to that, but less than the bulls imply, and the reason is scale. Advertising is growing and
high-margin, but it is not yet large enough, relative to AWS, to be the engine that pulls the train if the cloud
division falters — and crucially, much of advertising's value is itself tethered to the health of the retail and
cloud ecosystems that surround it. Amazon's profit structure is migrating from "almost entirely AWS" toward
"mostly AWS, with a growing advertising contributor," which is real progress and a real mitigant. But it does not
change the central fact for the bet at hand: the $200 billion is going into AWS and AI, the profit that justifies
the company still leans overwhelmingly on AWS, and advertising — for all its promise — is not yet the kind of
second profit pool that Microsoft has in Office or Google has in search. It is a cushion, not a replacement. The
train still has one real engine, and a promising auxiliary motor that is not yet rated to pull the load alone.

The circle every hyperscaler is standing in

Step back from Amazon specifically and notice the shape of the whole AI-infrastructure boom, because Amazon is
one node in a structure that should give any investor pause. The hyperscalers — Amazon, Microsoft, Google, Meta —
are collectively spending something approaching half a trillion dollars a year on AI infrastructure, an enormous
share of which flows to a single supplier, Nvidia, whose record revenues are in turn cited as proof that the AI
boom is real and that the spending is justified. The spending validates the supplier; the supplier's results
validate the spending. It is a circle, and Amazon's $200 billion is part of its circumference.

The uncomfortable question inside that circle is the one about end demand. All of this capacity is being built on
the conviction that enterprises and consumers will pay, durably and at scale, for AI services sufficient to earn
a return on half a trillion dollars a year of annual infrastructure spending. Some of that demand is real and
arriving — AWS's own AI revenue run rate has crossed roughly $15 billion, a genuine business. But $15 billion of
AI revenue against $200 billion of annual capex is a ratio that only works if the AI revenue compounds
ferociously for years, exactly as projected, with no air pocket, no enterprise spending pause, no realization
that the productivity gains are slower to monetize than the keynote promised. Amazon is better positioned than
most to survive an air pocket — it has the retail cash flows, the AWS base, the balance sheet. But it is building
into the same demand assumption as everyone else, and the free-cash-flow collapse is what that assumption costs
in the present tense while the future tense remains a forecast.

What the bulls get right

In fairness — and the fairness matters, because Amazon is a genuinely formidable company and this is not Lucid —
there is a serious bull case, and parts of it are strong enough to give any skeptic pause.

Start with the most powerful historical pattern: Amazon has gone free-cash-flow negative before, during previous
investment cycles, and each time it emerged with a larger, more dominant, more profitable business, and the stock
rewarded the patient enormously. "Amazon is spending so much it has no free cash flow" has, three or four times
in the company's history, been precisely the moment to buy. The bears who fixated on the cash outflow in 2001, or
2014, or 2022 missed the build-out that followed. There is a real, evidence-backed argument that a 95% FCF
collapse driven by growth investment is a fundamentally different and more bullish thing than a FCF collapse
driven by deteriorating operations — and on the numbers, this is the former.

Then there is the silicon story, which is genuinely impressive. Amazon's custom chips — Trainium for AI training
and inference, Graviton for general compute — now exceed $10 billion in combined annualized revenue, growing
at triple-digit rates, and the company believes Trainium can save tens of billions of dollars in annual capex
and deliver a several-hundred-basis-point margin advantage on inference versus renting Nvidia silicon. If that
plays out, Amazon is not just spending on AI; it is vertically integrating the most expensive input in AI, which
could turn the capex burden into a durable cost advantage and a defense of exactly the margins this essay warns
about. That is a real and potentially decisive counter to the depreciation worry.

The honest synthesis is this: the bull case is not a fantasy, and Amazon has earned the benefit of the doubt
more than almost any company alive. But "earned the benefit of the doubt" is not the same as "the bet is safe."
The previous build-outs that vindicated the cash incineration were investments in businesses where Amazon was
winning and pulling away. This one is an investment in a business where Amazon is the leader but growing
slower than its rivals — and that distinction is the whole question.

Why this build-out is different

It is worth dwelling on that distinction, because it is the single most important difference between this capital
cycle and the ones that made Amazon bulls rich. In the earlier cycles — fulfillment centers, AWS's own
creation, Prime — Amazon was building out capacity for a business it was decisively winning, into demand it
largely owned, against competitors who were behind. The spending was aggressive but the competitive position was
commanding, so the capacity reliably filled with profitable revenue and the cash came back multiplied.

This cycle is different in its competitive geometry. The AI-cloud build-out is a race in which Amazon is not
pulling away but, on the growth numbers, slipping — spending enormous sums into a contest where Microsoft and
Google are spending comparably and growing faster. When you build ahead of demand in a market you dominate, the
capacity fills and the bet pays. When you build ahead of demand in a market where your rivals are growing twice
as fast, you risk building capacity that fills more slowly than you projected, depreciating assets against
revenue that arrives later and smaller than the plan assumed. That is the scenario the $1.2 billion of trailing
free cash flow should at least make an investor consider: not that Amazon is failing, but that for the first
time the cash-incineration-equals-buy-signal reflex is being applied to a build-out where Amazon is the
defender, not the conqueror. The pattern that worked when Amazon was winning is being trusted in a contest Amazon
is, at the margin, losing.

The kicker

The genius of Amazon, for two decades, was that it made spending look like saving — that every dollar poured into
the business came back as three, and the free cash flow rose even as the investment soared, and the bears who
counted the outflows always looked foolish in hindsight. That reflex, earned over twenty years of being right, is
now doing enormous work: it is the reason a 95% collapse in free cash flow and an $18 billion quarterly cash
outflow were greeted as a buying opportunity rather than a warning. Maybe the reflex is right again. Amazon has
earned the presumption. But the bet is no longer the bet that worked before. It is $200 billion poured into one
division that already funds the entire company, to defend a lead that is shrinking against rivals growing twice
as fast, with a depreciation bill that will press for years on the one margin Amazon cannot afford to lose.

The cash machine that defined a generation of investing just produced $1.2 billion, down from twenty-six, and the market called it a buy. Somewhere in a data center the size of a town, a hundred billion dollars of silicon hums in the dark, depreciating by the second, waiting for the revenue the spreadsheet promises is coming — and one division, carrying the weight of the whole company, races a little slower each quarter than the two giants racing it.