Microsoft and Meta both reported results after the closing bell on Wednesday. Both are spending enormous sums on artificial intelligence infrastructure. Investors rewarded one and punished the other, and the difference between them is the clearest statement yet of what the market now requires from this spending.
Microsoft delivered an acceleration. Revenue reached $90.01 billion in the quarter ended 30 June, up about 18% from a year earlier and ahead of the roughly $87.62 billion expected, with net income of $35.77 billion, or $4.81 a share. Its shares rose about 8% in after-hours trading, helped by capital spending for the quarter coming in below expectations. The number that mattered was Azure, where growth quickened to 43% in constant currency from 40% in the prior quarter, ahead of the roughly 40% analysts expected. Azure revenue passed $100 billion for the full fiscal year for the first time, up 41%.
Meta delivered the opposite pattern. Revenue came in at $60.80 billion against expectations near $59.50 billion, a clear beat. Earnings per share came in at $6.18 against roughly $7.14 expected, a substantial miss. The company also lifted the floor of its capital spending range for 2026 to $135 billion to $145 billion, from $125 billion to $145 billion, and raised its expense outlook to $165 billion to $169 billion. It then guided third-quarter revenue to a range of $61 billion to $64 billion, the lower end of which sits below the roughly $63.15 billion analysts expected. Shares fell about 11% in after-hours trading.
The same behaviour, two verdicts
It is worth being precise about what separates these two reports, because on the surface they look similar. Both companies are spending at a scale no software business has ever attempted. Microsoft full-year capital expenditure reached $115.95 billion, up nearly 80% from $64.55 billion the year before. Meta is committing a comparable sum.
The difference is what arrived alongside the spending.
Microsoft showed acceleration in the specific revenue line the spending is meant to produce. Azure did not merely grow; it grew faster than the quarter before, faster than expected, and management guided to 45% growth in the current quarter against a consensus near 41%. That is a company demonstrating that the capacity it is building is being consumed as fast as it comes online.
Meta showed the cost side moving faster than the profit side. Revenue was strong and the advertising business is healthy, with ad impressions up 14% and average price per ad up 12%. But earnings fell short, expenses rose, and the capital budget floor moved up rather than the ceiling moving down. For a company whose artificial intelligence spending is harder to trace to a specific revenue line, that combination is the least reassuring one available.
Why the revenue attribution problem matters so much
This is the structural asymmetry that has been building all year, and Wednesday made it explicit.
Microsoft sells cloud capacity. When it builds a data centre and fills it, the revenue appears as Azure consumption, in a reported segment, in the same fiscal year. An investor can divide one number by the other and form a view.
Meta does not sell compute. Its artificial intelligence spending improves recommendation systems, advertising targeting and content ranking inside products that were already profitable. The benefit is real, and it may be large, but it arrives blended into an advertising line that was growing anyway. There is no disclosure that isolates the return.
In a market that rewarded ambition, that distinction did not matter. In a market that has started demanding proof, it matters enormously. As our analysis of the capital spending question set out earlier this week, the burden of proof has moved from why are you not spending more to when does this pay for itself.
The Alphabet precedent
Neither reaction should have surprised anyone who watched last week.
Alphabet reported an excellent quarter, beating on revenue with cloud growth of 82%, and its shares fell about 5% after it raised full-year capital spending guidance to between $195 billion and $205 billion. The company also showed negative free cash flow for the first time in its history as a public company, a direct consequence of the building programme.
That was the moment the rule changed. Beating on revenue stopped being sufficient. The market began pricing capital intensity as a cost rather than a signal of confidence, and every subsequent report has been judged on that basis.
The depreciation question nobody answered
There is a second issue underneath both reports which neither addressed directly, and it will matter more each quarter.
Capital spending does not hit profit when the money is spent. It arrives later as depreciation, spread across the assumed useful life of the equipment. Microsoft $115.95 billion of full-year capital expenditure will convert into an annual charge against profit that depends entirely on how long the company assumes its accelerators remain useful.
If that assumption is six years, the charge is manageable. If the real economic life proves to be three, because a newer generation of chips makes the installed base uncompetitive for frontier work, the charge roughly doubles and reported margins fall accordingly. The cash has already gone either way. What changes is when the income statement admits it.
This is why the chip market matters to these two companies even though neither makes chips. A faster hardware cycle shortens asset lives. A slower one lengthens them.
What this means for the sector
- Companies that sell compute directly, and can show utilisation, are being given room to spend. Microsoft has earned that room for at least another quarter.
- Companies whose artificial intelligence returns are blended into an existing business face a disclosure problem rather than a performance problem. Meta advertising business is performing well; it simply cannot show the return on the specific investment.
- Raising the floor of a spending range is now read as a negative signal, where a year ago raising any part of it was read as confidence.
- Expense guidance has become as market-moving as revenue guidance, which is unusual for companies of this size.
- Apple reports on Thursday alongside Amazon. Apple has spent conspicuously less than its peers and, as our report on its rise past Nvidia noted, briefly touched $5 trillion this week on precisely that restraint.
The wider setting made it worse
Both reports landed at the end of a brutal session. The Nasdaq Composite closed 1.74% lower and finished the day more than 10% below its record high. Long-dated Treasury yields rose sharply after the Federal Reserve held rates with three officials dissenting in favour of an increase, with the thirty-year yield reaching its highest level since 2007.
Higher long-term interest rates are specifically hostile to companies whose value rests on cash flows arriving many years from now. A build-out justified by returns in 2029 is worth measurably less when the rate used to discount those returns rises. The macro backdrop and the earnings reaction were pushing in the same direction.
Outlook
The useful conclusion from Wednesday is not that artificial intelligence spending is wrong or right. It is that the market has stopped treating it as a single trade.
For three years, capital expenditure on artificial intelligence was assessed as a sector-wide bet, and every large technology company rose and fell together on it. That is no longer how it is being priced. Each company is now being asked, individually, to show what its spending produced this quarter. Microsoft could. Meta could not, at least not in a form an investor can verify.
That is a more demanding standard, and it is arguably a healthier one. It also means the next two days will be read the same way, one company at a time.
About the data: Microsoft figures are from its fiscal fourth-quarter 2026 results for the quarter ended 30 June 2026, including revenue growth, net income, earnings per share, Azure growth in constant currency, full-year Azure revenue and full-year capital expenditure with the prior-year comparison, together with management guidance for the current quarter. Meta figures are from its second-quarter 2026 results, including revenue, earnings per share, capital expenditure and total expense guidance ranges, and advertising impression and pricing changes. Analyst expectations referenced are consensus estimates compiled before the releases. Share price movements described are after-hours trading and are not closing prices. Alphabet figures are from its second-quarter 2026 results. Index levels and Treasury yields are closing values for 29 July 2026.
Reader note
This article is general information, not personalized financial advice. Read our Financial Disclaimer.