Amazon's Earnings Never Stop Rising. Its Cash Flow Just Went Negative.

Position disclosure: Long AMZN at time of writing. Mythic Market Research is a proprietary research publication, informational and educational purposes only, not investment advice or a solicitation. Do your own research.

Prices and figures as of the July 16, 2026 close. AMZN $249.89.

Here is the sentence that defines Amazon right now:

Clean EPS rises every single year from FY25 to FY28 ($6.3 → $7.5 → $9.4 → $11.8) while free cash flow goes to negative $25 billion this year.

That is not a contradiction. It is an accounting fact about a company spending $200 billion on capex it depreciates over five to thirty years. Depreciation is non-cash. Capex is cash. Earnings look through the build; cash flow eats it.

Which means the market is being asked to own an earnings inflection before the cash inflection prints. That is a harder thing to hold than it sounds, and the tape tells you the market hasn’t fully agreed to do it: AMZN is up 10% over twelve months, up 0.6% over three, sitting below its 50 day and 9.1% off its high, while META and AAPL put up +12% in the last month alone.

Amazon leads neither its mega-cap cohort nor the AI trade it is levered to.

That is the whole post. The business is inflecting. The chart is just starting. The debate is timing.

The number that should have moved the stock and didn’t:

AWS grew 28% year over year in Q1 2026 (the fastest in fifteen quarters) at a 37.7% operating margin.

The five-quarter sequence: 17% → 17% → 20% → 24% → 28%. That is not noise. That is a reacceleration off a $150bn run rate, from the largest cloud in the world at roughly 30% global share. Twenty-eight percent growth on that base adds more absolute revenue dollars than either faster-growing rival does off a smaller one.

Underneath it:

• Custom silicon inflecting. Trainium and Graviton crossed a $20bn run rate. Trainium2 is sold out.

• Contracted, not speculative. AWS remaining performance obligations reached $364 billion. OpenAI committed ~2GW from 2027. Anthropic up to 5GW.

• AI monetization is real revenue. An AI services run rate above $15bn growing triple digits; Bedrock customer spend up 170% quarter over quarter.

• Consolidated operating margin hit 13.1% record!

Management’s framing is explicit and demand-led: they are monetizing capacity as fast as they can install it.

Free cash flow collapsed to $1.2 billion trailing twelve months, from $38 billion. Capex hit $147bn, up 67%. Long-term debt jumped from $66bn to $119bn in a single quarter, a $53bn raise to fund infrastructure.

Then the harder problem: AWS at +28% is still the slowest-growing of the big three. Azure is running 40%. Google Cloud 63%. Amazon is reaccelerating and still ceding growth-rate leadership, particularly at the AI high end.

So the bear case isn’t that the business is bad. It’s that Amazon is spending the most, growing the slowest of the three, levering up to do it, and asking you to wait until FY27–28 to find out whether the capital earned its return.

Reported FCF says Amazon generates roughly nothing. That overstates the distortion, it charges the entire AI growth build against a business whose existing revenue base doesn’t need it.

The common fix is gross owner earnings: operating cash flow minus D&A. That gets you $78bn today rising to $150bn by FY28, and a headline 19x FY28E. Cheap!

It’s also the optimistic bound, and it cheats twice: it assumes servers need no more replacement than today’s depreciation of a smaller historical base, and it treats $20bn a year of stock-based comp as free.

MMR anchors to the conservative line instead, operating cash flow, minus maintenance capex proxied at 1.3× D&A, minus SBC:

Owner earnings conservative: $37B - $90B

Gross owner earnings (upper bound): $78B-$150B

On enterprise value of $2.84tn (including $119bn net debt, not market cap), conservative owner earnings put AMZN at 41x FY27E and 32x FY28E.

Roughly in line with the earnings multiple. Fair, not cheap.

The reported picture overstates the distortion. The gross picture overstates the value. The line between them is the honest read: a genuine 2.4× inflection over four years that supports an A− business grade and does not make Amazon a bargain.

And one assumption is pivotal for all of it: five-to-six-year server lives. If AI accelerators economically obsolete on Nvidia’s roughly annual cadence — two to three year lives — then lives get cut or silicon gets written down, maintenance capex and depreciation both jump, and conservative owner earnings fall back toward reported FCF. That single assumption is the difference between 32x and un-ownable. It is not a footnote. It is the thesis.

Not vague risk language. Specific conditions that end the thesis:

1. Server useful-life reset. AWS margins are flattered by 5–6 year depreciation. If AI silicon obsoletes on a 2–3 year cadence, lives get cut or assets written down margins compress, maintenance capex rises, owner earnings collapse toward reported FCF.

2. Capex climbs with no ceiling. Every guide has stepped up: $83bn → $132bn → $200bn, the last coming in $54bn above Street. Management won’t give an FCF floor. A higher FY27 guide slides the inflection a full year right.

3. Backlog quality cracks. The $364bn RPO leans on OpenAI and Anthropic and Amazon funds Anthropic, which buys AWS and Trainium. That circularity means a renegotiation hits revenue and the equity mark simultaneously.

4. AWS growth stalls back toward 20% or below on Jul 30. Breaks the reacceleration thesis outright and confirms the market’s capex skepticism.

5. A close below the 200-day ($234) ends the multi-year uptrend and moves AMZN out of Stage 2. That is the technical line, and it is not a soft one.

AMZN is a Grade C+ setup on a Grade A− business, and the whole trade is the timing of one inflection.

AWS grew 28% at a 37.7% margin. The $200bn capex peaks this year and is demand-led and contracted. It should convert to AWS revenue roughly doubling by FY28 with margin toward 40% and conservative owner earnings compounding from $37bn to $90bn.

But on enterprise value that’s 32x FY28E owner earnings (fair, not cheap) and it rests on 5 - 6 years server lives holding.

What keeps this at C+ is the tape: up 10% over twelve months, flat over three, below the 50-day, still the slowest-growing of the big three.

This is a high-conviction inflection to stalk for a Stage-2 reclaim confirmed by earnings not to chase in the range.