AI Markets: Microsoft reassures, Meta disappoints, China hits back

Microsoft reassures markets with its results, while Meta worries them despite record revenue. Between China’s pushback on humanoid robots and doubts about AI data center debt, this edition takes stock of a tense earnings week.

Microsoft tops $100 billion in AI cloud, but the real story is in the accounting

On July 29, 2026, Microsoft posted quarterly results that beat expectations, powered by Azure and AI. But the most interesting signal for markets is tucked inside an accounting change that lightens data center costs.

Key points:

  • Quarterly revenue of $90 billion (+18% year over year), net income of $35.8 billion (+31%) and earnings per share of $4.81, above the analyst consensus (about $4.24 expected).
  • Azure grew 43% in the quarter. For the full fiscal year, Azure cloud tops $100 billion in annual revenue for the first time (+41%).
  • Microsoft signed more than $130 billion in new data center leases in this quarter alone, according to Bloomberg, bringing its future commitments not yet on the books to $329 billion.
  • The company is keeping its 2026 capital spending budget nearly unchanged, partly thanks to stretching the accounting lifespan of its data centers.
  • The stock jumped about 8% after the July 29 close.

Azure and AI drive solid results

Microsoft beat expectations on nearly every metric. Net income climbed 31%, helped by a $3.2 billion gain tied to its stake in Anthropic. Azure growth accelerated to 43% in the quarter, and for the first time it crossed $100 billion in revenue over a full fiscal year, up 41%. Microsoft 365 Copilot now has more than 30 million paying subscribers, up from about 20 million three months earlier. Capital expenditures hit $41 billion in the quarter alone, up 70% year over year. According to CFO Amy Hood, demand still exceeds available cloud and AI capacity.

The accounting trick that inflates reported profitability

The most important point for investors lies elsewhere. Microsoft is extending the accounting lifespan of its data centers and offices from 15 to 25 years, starting in fiscal 2027. In other words, the company spreads the wear and tear on its assets over a period almost twice as long. The result: lower depreciation charges each year, and therefore margins that look better, without actual spending changing by a single cent. Microsoft already pulled this lever in 2022, extending the lifespan of its servers from 4 to 6 years, for a $3.7 billion boost to operating income. This change also explains why the stated 2026 capital spending budget drops to about $175 billion instead of the expected $190 billion: some future data center leases move from an accounting category counted as capex to one that isn’t. Amy Hood confirmed it: actual investment plans aren’t changing. That’s a reassuring signal for markets, but also a reminder that AI’s reported profitability owes a lot to accounting rules. For a European investor, Microsoft can’t be held in a PEA, since it’s listed on the Nasdaq: exposure goes through a standard brokerage account or a UCITS-format US equity/tech ETF.

Sources:

Meta beats revenue expectations, and the market punishes it anyway

Meta reported better-than-expected revenue on July 29, 2026. But the stock fell after the close. The AI bill is worrying investors more than ever.

Key points:

  • Quarterly revenue: $60.8 billion, up 28% year over year, above the consensus of about $60.3 billion.
  • Earnings per share: $6.18, well below the $7.22 analysts expected, according to GuruFocus.
  • Meta generated $31.9 billion in cash from operations but kept only $784 million of it, after spending more than $31 billion on AI data centers, according to The Motley Fool.
  • The stock dropped several points after the close, wiping roughly $15 billion off Mark Zuckerberg’s fortune on paper, according to 24/7 Wall Street.

Where did all that cash go?

Of the $31.9 billion in cash generated by operations, Meta spent about $30.1 billion on servers, land and data centers, plus nearly $1 billion in lease payments. That left just $784 million in “free cash flow,” compared with $8.5 billion a year earlier, according to The Motley Fool. The company also paid out $1.35 billion in dividends, funded partly by debt rather than profits. For 2026, Meta raised its capital spending range to $130-145 billion (from $125-145 billion previously), and its total expenses to $165-169 billion, according to the company’s official press release.

A “generational opportunity” the market no longer takes at face value

Mark Zuckerberg defended his strategy in the press release: “AI is accelerating our business today, powering our next products and opening the door to new opportunities.” He frames this massive spending as preparation for what he sees as a generational shift toward highly advanced AI. But investors are skeptical: spending tens of billions is no longer enough without clear proof of a return on investment. The third-quarter revenue forecast ($61 to $64 billion) didn’t help either, with growth already slowing. That same evening, Microsoft posted results seen as more reassuring on the same issue of AI spending (see our dedicated article): proof that the market now judges each company case by case.

What about a European investor?

Meta is listed on the Nasdaq, so it isn’t eligible for a PEA. A European investor who wants exposure to this kind of stock generally goes through a UCITS-format US equity or tech ETF. These results are a reminder that beating revenue expectations is no longer enough to reassure if profitability and cash flow raise questions.

Sources:

Robots banned, chips defended: Washington plays both sides against China

Two days after the US ban on Chinese humanoid robots, Beijing is raising its voice and Nvidia is stepping up to defend its own interests in China — a contradiction that says a lot about Washington’s tech policy.

Key points:

  • China’s Commerce Ministry says the FCC ban on humanoid robots “severely damages” bilateral relations and threatens retaliation (rare earths, Chinese market access for Tesla and Nvidia).
  • BYD confirms the August launch of its first humanoid robot, nicknamed “Xiao Di,” in its network of “Di Space” centers — joining Unitree, Xiaomi and Xpeng in this market.
  • Jensen Huang (Nvidia) publishes his very first post on X: a letter to Congress co-signed by Microsoft, Meta, IBM, Palantir and Hugging Face, arguing against restrictions on open AI models.
  • These two fronts (robots banned, chips defended) reveal a tension between national security and commercial interests in US policy toward China.

Beijing threatens, BYD speeds up

According to China’s Commerce Ministry, quoted by CNBC and Benzinga on July 30, adding Chinese humanoid and quadruped robots to the FCC’s “Covered List” “severely damages” the stability of trade relations. In response, Beijing is floating restrictions on rare earths and on Chinese market access for US companies like Tesla and Nvidia. According to Benzinga, China holds about 85% of the global humanoid robot market — a figure to take as a rough order of magnitude, not an absolute certainty. This standoff comes ahead of a planned visit by Xi Jinping to the United States in September, which could complicate negotiations.

Meanwhile, Chinese competition is organizing regardless: BYD confirmed to China Securities Journal the August launch of its first humanoid robot, named “Xiao Di,” in its “Di Space” showrooms, initially designed as a sales assistant at car dealerships. It joins Unitree (targeted by the US ban), Xiaomi and Xpeng, already in the race for bipedal robots.

Nvidia defends its sales in China

On July 24, Jensen Huang published his very first post on X to share an open letter, “Open Weights and American AI Leadership,” co-signed by about twenty tech companies. In it, he writes that “just as open source showed that transparency can be more secure than obscurity, AI safety may depend on more people being able to test the models.” The move comes as Huang makes repeated trips to Washington to defend AI chip exports to China, an issue on which Donald Trump has floated possible sales to “approved” Chinese customers in exchange for a 25% surcharge.

For investors, this standoff mainly exposes Nvidia on the China sales front, while the ban on Chinese robots could, in theory, benefit Western players like Tesla (Optimus), Figure or Boston Dynamics in the US market — a hypothesis, not a certainty. On the European side, Nvidia and Tesla remain ineligible for a PEA (listed on the Nasdaq) and can be bought through a standard brokerage account. Indirect, diversified exposure to the robotics/AI theme is available through UCITS ETFs such as KOID (the European version listed in London) or BOTZ, though this is not a buy recommendation.

Sources:

SK Hynix: a record quarter, and a stock that plunges nearly 10%

The South Korean memory giant posted the best quarterly results in its history on July 29, 2026. That same day, its stock tanked. This disconnect reveals what markets really expect from the artificial intelligence boom.

Key points:

  • Record revenue of 79.3 trillion won (about $54 billion, +257% year over year) and record operating profit of 60.5 trillion won (about $41 billion, +557%), according to SK Hynix’s official results.
  • Both figures, historic as they are, came in below what analysts expected (about 84 trillion and 64 trillion won).
  • Net income reached 93.9 trillion won, or about $64 billion — the figure cited by the Wall Street Journal. It’s inflated by a one-time gain: the sale of SK Hynix’s stake in Japan’s Kioxia.
  • The stock closed down 9.6% on earnings day, after losing more than 15% intraday. Samsung Electronics, which had not yet released its own results, fell more than 13% the same day by contagion.

The $64 billion figure, and what it really hides

Be careful not to confuse two things. SK Hynix’s second-quarter net income comes to about $64 billion, but that amount includes a one-off financial gain from selling a stake, not just memory chip sales. The profit that reflects the actual business, operating profit, is closer to $41 billion — and that’s the figure that came in below the analyst consensus. Another possible source of confusion: in early July, SK Hynix had already announced a separate investment plan of about 100 trillion won (nearly $64.4 billion) to build new fabs in Cheongju. Two different “$64 billion” figures, two different realities, not to be mixed up.

Why the market has doubts, despite AI demand

HBM memory, used in Nvidia chips to train AI, remains SK Hynix’s growth engine: shipments of the new HBM4 generation began this quarter. But some deliveries were pushed back to the second half, which partly explains the shortfall versus consensus. More broadly, after a stock that had already climbed a lot this year, investors are wondering: will Big Tech’s massive AI spending really pay off, and won’t memory supply eventually outstrip demand, as in previous cycles? Samsung, which released its own record results the next day and beat expectations, didn’t escape this sector-wide jitteriness either.

For a European investor, SK Hynix trades on the Seoul stock exchange and isn’t eligible for a PEA: you need a standard brokerage account, or indirect exposure through a global semiconductor UCITS ETF, where the Korean weighting generally remains a minority.

Sources:

CoreWeave’s CEO sells shares while AMD bets $14 billion on Core Scientific, the company he couldn’t buy

In late July, two pieces of news shed light on the fragility of AI compute “landlords”: repeated share sales at CoreWeave, and a giant contract signed by AMD with Core Scientific, the company CoreWeave tried to acquire nine months ago.

Key points:

  • Michael Intrator, CEO of CoreWeave, sold 307,692 shares on July 14 for nearly $25 million, his third sale in a month, through an automatic 10b5-1 plan.
  • On July 28, AMD signed a 15-year agreement with Core Scientific: up to 2.5 GW of capacity, for a stated $14 billion in base contract revenue.
  • Core Scientific is the company whose shareholders rejected a $9 billion buyout offer from CoreWeave in October 2025.
  • AMD also received warrants on 30 million Core Scientific shares — hence TheStreet’s take: the deal is “bigger than the headline number.”

An insider sale that feeds the jitters, without being a sure signal

According to The Motley Fool, Michael Intrator sold 307,692 shares on July 14 for nearly $25 million, at an average price of $80.85. It’s his third sale in a month: about 370,000 shares ($30.8 million) in early July, then $24 million on July 21. All of them fall under an automatic selling plan (10b5-1) adopted in November 2025, which schedules sales in advance to avoid any suspicion of insider trading. The CEO still holds more than 2.6 million shares directly and tens of millions of options. An automatic plan doesn’t, in itself, signal a loss of confidence. But with the stock down 30% in a month, every sale draws attention. Analysts remain divided: Jefferies and Citigroup downgraded the stock in late July, while Rosenblatt maintains a $250 target, well above the current price.

AMD bets on the one CoreWeave couldn’t afford

On July 28, AMD and Core Scientific announced a 15-year partnership covering 529 megawatts of capacity, with an option to reach 2.5 gigawatts by 2028, for a stated $14 billion in base contract revenue. But TheStreet points out that this figure understates the deal’s real scope: AMD also obtained warrants on up to 30 million Core Scientific shares, at $23.47 apiece, 6.5 million of which have already vested. So AMD isn’t just renting space, it’s taking a potential stake in its partner. The irony: Core Scientific is the company whose shareholders chose to keep it independent in October 2025, rejecting a $9 billion buyout from CoreWeave. Nine months later, that bet seems to be paying off, even though the stock first jumped 11% before falling back, as investors worried about dilution from the warrants.

For an investor based in France, CoreWeave and Core Scientific are listed on the Nasdaq: neither is eligible for a PEA, and only a standard brokerage account gives access to them. Indirect exposure is still possible through thematic AI or data center UCITS ETFs.

Sources:

Edition prepared by our monitoring agent (data and sources as of Thursday, July 30, 2026), reviewed in line with our editorial policy. General information, not investment advice: investing carries a risk of capital loss.