The verdict is in: AI spending needs receipts

The verdict is in.

The market told us what it wants.

On Monday, I wrote that this week would help decide whether the AI trade still holds.

It is Friday.

Four of the world’s largest companies have reported.

The Federal Reserve has decided.

The Dow has fallen 1,150 points in one session and rebounded sharply in the next.

The answer is more precise than a simple yes or no.

The market is not rejecting AI investment. It is demanding receipts.

Microsoft and Amazon were rewarded.

Meta and Apple were punished.

The dividing line was not simply whether a company spent heavily, beat earnings or used the word AI.

It was whether the new evidence made future economic returns easier or harder to see.

Checking Monday’s thesis

My week-ahead research framed Microsoft, Meta, Apple and Amazon as four versions of the same test:

How much should investors pay for future growth when the capital bill, inflation path and discount rate are all uncertain?

Alphabet had already shown that exceptional cloud growth could coexist with a falling stock if the spending outlook and cash-flow pressure became more demanding.

At the time, the emerging lesson looked simple:

The market punishes AI spending.

This week’s results show why that conclusion was incomplete.

Amazon raised its 2026 capital-spending plan to $220 billion and its shares surged.

Microsoft continues investing aggressively and produced its strongest daily share-price gain in 18 years.

Meta is spending heavily too, but its shares fell.

The amount spent did not determine the reaction.

The evidence produced by the spending did.

MSFT

+15.5%

AMZN

+14.0%

META

−8.5%

AAPL

−9.5%

−16%0+16%

Sources: company results and market reactions reported by Reuters, 30 July and Reuters, 31 July. Values are deliberately rounded.

Microsoft: the cleanest receipt

Microsoft delivered the clearest answer of the week.

Azure and other cloud-services revenue grew 43%, above management’s 39–40% guide.

Azure also surpassed $100 billion in annual revenue for the first time, growing 41% for the full fiscal year.

Microsoft Cloud revenue reached $59.3 billion for the quarter, while commercial remaining performance obligations increased 84% to $678 billion.

Those figures connect the AI infrastructure build-out to current demand and contracted future business. Microsoft’s fiscal Q4 release

The market rewarded the evidence.

Microsoft rose more than 15% on Thursday, its largest one-day percentage gain since 2008. The move added roughly $450 billion to its market value.

That reaction was not proof that Microsoft’s spending is risk-free.

The company still faces rising depreciation, enormous infrastructure requirements and the need to convert demand into durable cash returns.

But the quarter moved the evidence in the right direction:

• Azure accelerated beyond guidance.
• Annual Azure revenue passed a major threshold.
• Contracted business expanded.
• Management’s forward cloud outlook remained strong.
• Capital requirements appeared better supported by visible demand.

Those are receipts.

Amazon: spending rose, but AWS made the case

Amazon made the week’s most important correction to the “capex is bad” thesis.

AWS revenue grew 37% to $42.2 billion, its fastest growth rate in 18 quarters.

AWS operating income increased from $10.2 billion to $16.6 billion.

That produces an operating margin of approximately 39.3%.

AWS represented about 21% of Amazon’s revenue but roughly 60% of its operating income.

Amazon then increased its 2026 capital-spending plan from $200 billion to $220 billion.

The shares still surged—about 14% in early Friday trading and approximately 15% at the latest check. Amazon’s Q2 release

The market did not ignore the spending.

It judged the spending alongside the operating evidence.

AWS growth accelerated.

AWS margins remained exceptional.

Management said demand continues to exceed available capacity.

That makes the capital commitment easier to connect to a revenue-producing business.

There is an important qualification.

Amazon’s company-wide free cash flow remains under pressure, and quarterly net income included a $53.4 billion non-operating gain primarily related to its Anthropic investment.

The headline profit is therefore not the receipt I would use.

The stronger evidence is inside AWS: revenue acceleration and segment operating profit.

Amazon has not proved that the entire $220 billion programme will earn an acceptable return.

It showed enough current operating leverage for the market to keep funding the argument.

Meta: growth without enough cash conversion

Meta’s operating business did not collapse.

Revenue grew 28% to $60.8 billion.

Daily active people increased 3%.

Advertising demand remained strong.

But the economic conversion deteriorated sharply.

Free cash flow fell 91%, from $8.55 billion a year earlier to $784 million.

Quarterly capital expenditure reached $31.1 billion.

Operating margin fell from 43% to 31%, although legal charges and severance also contributed to the decline.

The shares fell roughly 8–9% after the report. Meta’s Q2 release

Meta’s revenue growth is evidence.

It was not enough to answer the next question:

How quickly can the expanding infrastructure base produce returns that offset the cash and margin pressure?

The market did not conclude that Meta’s core advertising business was weak.

It concluded that the receipt for the next stage of spending remained less complete than the one Microsoft or Amazon provided.

That distinction matters.

Strong revenue does not automatically mean strong capital efficiency.

$8.55B

Q2 2025

$0.784B

Q2 2026

−91% year over year

Source: Meta Q2 2026 results. Free cash flow equals operating cash flow less purchases of property and equipment.

Apple: a good quarter with weaker forward evidence

Apple is different from the other three.

Its report was less a test of hyperscale AI infrastructure and more a test of ecosystem growth, supply and guidance.

The completed quarter was strong.

Revenue increased 16% to $109.4 billion.

Earnings per share rose 29% to $2.02.

iPhone revenue reached a record $54.25 billion.

But Services revenue of $30.74 billion came in below expectations, management warned about significant supply constraints, and the September-quarter revenue-growth outlook of 9–11% was below Wall Street’s approximately 12% expectation.

Apple shares fell roughly 9–10% in Friday trading. Apple’s fiscal Q3 release and Reuters’ outlook analysis

The headline quarter was good.

The forward evidence was not good enough for the expectations already embedded in the price.

That is the same mechanism I described after Alphabet:

A company can pass the business test and still fail the expectations test.

What counts as a receipt?

A receipt is not one specific number.

It is evidence connecting capital committed today to economic output that can persist.

Depending on the business, that may appear as:

• Revenue acceleration
• Backlog or contracted demand
• Higher utilisation
• Expanding segment margins
• Operating leverage
• Free-cash-flow growth
• Improved returns on invested capital
• Guidance supported by visible capacity demand
• Lower unit costs or better monetisation

Microsoft’s receipt was Azure acceleration, annual scale and contracted demand.

Amazon’s was AWS acceleration and segment profitability.

Meta showed strong top-line growth, but weaker cash conversion and lower consolidated margin.

Apple beat the completed quarter, but offered softer forward evidence.

No single metric explains every reaction.

The principle is broader:

Spending is neither bullish nor bearish by itself. The return evidence determines whether the market treats it as investment or cost.

Expectations still decide the price reaction

Receipts matter, but they are not the only variable.

Valuation and prior expectations determine how much evidence is required.

A company priced for modest growth may rise on a small improvement.

A company priced for perfection can fall after an excellent quarter.

That is why “beat” and “miss” are insufficient descriptions.

The useful questions are:

• What outcome was already priced in?
• Which operating assumption changed?
• Did the result improve the expected return on capital?
• Did guidance raise or lower the future hurdle?
• Is the market paying more or less for each unit of growth?

Microsoft and Amazon did not rise merely because their reports were good.

Their results exceeded the economic proof the market required at those prices.

Meta and Apple did not fall because their businesses suddenly became bad.

Their reports left important expectations insufficiently supported.

The week was built to create whipsaw

Earnings were only part of the story.

On Wednesday, the Federal Reserve held its target range at 3.50–3.75%.

The vote was 9–3.

Three members preferred a quarter-point hike—the most visible sign this week that the inflation debate remains unresolved. Federal Reserve statement

The Dow fell 1,152 points, or 2.2%, its worst session since April 2025.

Rising yields, renewed Iran tensions and another leg of technology weakness intensified the decline.

Then Microsoft helped drive a sharp rebound on Thursday.

The S&P 500 rose 1.7%.

The Nasdaq gained 2.8%.

Microsoft surged more than 15%.

On Friday, Amazon rose while Apple fell and higher Treasury yields reduced some of the broader market’s early strength. Wednesday’s market recap and Friday’s market recap

WEDNESDAY Dow −1,152

Fed held at 3.50–3.75%. Three members preferred a hike. Yields and geopolitical pressure intensified the selloff.

THURSDAY MSFT +15.5%

Azure receipts changed the AI-spending narrative and helped drive a sharp technology-led rebound.

FRIDAY AMZN ↑ · AAPL ↓

AWS evidence was rewarded while Apple’s softer outlook disappointed. Rising yields limited the broader follow-through.

Sources: Federal Reserve and Reuters market recaps for Wednesday, Thursday and Friday.

The narrative changed several times in roughly 48 hours:

• Fed hold
• Three hawkish dissents
• Dow shock
• Microsoft-led rebound
• Amazon surge
• Apple decline
• Renewed yield pressure

A trader reacting to each headline had several opportunities to become perfectly positioned for the move that had just ended.

That is whipsaw.

Volatility provided information.

It did not provide a standing instruction to trade.

My portfolio through the verdict

Teradyne was one of the positions I opened during the previous Thursday’s scheduled deployment.

It was underwater shortly after entry.

Then it reported strong earnings on Wednesday and rose.

I am still holding it.

I did not buy Teradyne because I predicted the earnings beat.

I bought a partial position because it cleared the quality screen, ranked highly enough and fit the portfolio’s diversification rules.

The result arrived after the entry.

It was not the reason for it.

Lesson 21 explains why a winning outcome cannot travel backward and validate the original decision.

What the report did was add new evidence to the Teradyne thesis.

That evidence can influence what happens next.

It cannot rewrite what information existed before the report.

The rest of the deployment moved through the 1,152-point Dow decline without forcing a single decision.

That was possible because none of the positions was sized as though it needed to work immediately.

Cash remains around 57% of the account.

Some quality companies moved closer to my marked zones during Wednesday’s decline, then moved away during Thursday’s rebound.

That is fine.

A zone does not expire because price bounced for one day.

A trader does not need to chase because the market removed an opportunity before it became actionable.

Activity was available. Action was not required.

This week offered constant stimulation.

There was always another report, price move, rate interpretation or geopolitical headline available to trade.

The portfolio still required very little from me.

Hold the qualified positions.

Preserve cash.

Update the evidence.

Keep the levels marked.

Let price come to the system.

That is the bridge to Lesson 22: Boredom is not a reason to trade.

The temptation to manufacture activity does not appear only during quiet markets.

It can also appear after an exciting week, when every large move makes waiting feel like missed participation.

The control is the same:

If the system has no valid action, activity does not create one.

Closing the week—and July

Monday’s question was whether the AI trade still holds.

Friday’s answer is conditional.

The market will still fund large AI programmes.

It will even tolerate rising capital expenditure and near-term cash-flow pressure.

But the tolerance is not unconditional.

Companies must show credible operating receipts:

Revenue growth.

Demand visibility.

Margins.

Cash conversion.

Or a path that makes future returns easier to underwrite.

Microsoft and Amazon moved that evidence forward.

Meta’s cash-flow pressure left the case less complete.

Apple’s completed quarter was strong, but its forward evidence disappointed.

The market spent the week rewarding proof over promises.

That happens to be how I want to run the portfolio too.

No prediction.

No reaction to every headline.

Qualified positions.

Partial size.

Cash retained.

Evidence updated.

July ends today.

I will publish the full monthly recap, including portfolio numbers, early next week.

The verdict is not that spending wins or loses. The verdict is that spending must produce receipts.

This is a record of my process and opinions, not investment advice. I hold positions in securities mentioned or related. Copy trading involves risk, including loss of capital. Past performance is not an indication of future results.