The best quarter for corporate profits since 2021. Microsoft up 8%, Amazon’s first $200 billion quarter, Meta punished for a beat, Apple sold off for being boring. Roughly $700 billion of AI spending disclosed in a single week. And on Tuesday, SpaceX opens its books to public shareholders for the first time in twenty-four years. Here’s what actually happened, and what I think comes next.
By Zain-Ud-Deen Khan | 02/08/26 | ELT
There’s an old joke in trading circles about the only three rules you need: buy the dip, short the VIX, f*** Bitcoin. It’s funny because it’s stupid, and it’s stupid because it works right up until the afternoon it obliterates you.
This quarter, the joke’s been holding. Barely. But holding. And the reason it’s held is more interesting than the joke.
The Headline Number Is a Bit of a Lie
Start with the top line, because it’s genuinely remarkable and also slightly misleading.
With 61% of the S&P 500 having reported, blended earnings growth for Q2 sits at 47.4%. That’s the highest since Q1 2021. Some 86% of companies have beaten expectations, and in aggregate they have beaten by 31.4%, which would be the biggest surprise since FactSet started tracking it in 2008.
Quick definitions, because these get thrown around: an earnings surprise is the gap between what a company actually earned per share and what analysts collectively predicted. Blended growth mixes actual results from companies that have reported with estimates for those who haven’t.
Now the asterisk. That 31.4% is mostly two companies. Two of the biggest names. Alphabet posted EPS of $9.11 against a $2.88 estimate, but that included a $98 billion gain. Amazon posted $5.75 against $1.82, including $53.4 billion of non-operating other income, largely from its stake in Anthropic. Strip those two out and the surprise falls to 9.2%, and growth drops from 47.4% to 28.8%.
Still excellent. Still the seventh straight quarter of double-digit growth. But if you’d taken the headline at face value you’d have concluded corporate America had roughly doubled its earning power in twelve months. It has not. It’s rather revalued some investments and had a very good quarter.
Ten of eleven sectors grew. Healthcare was the lone decliner. Energy led, which, given what oil’s done since the Iran flare-up, surprises nobody.
The Week That Repriced the AI Trade
Between the 22nd and 30th of July, five of the Magnificent Seven reported inside about seventy-two hours. All five grew revenue by double digits. The market’s reactions had almost nothing to do with that.
Alphabet went first, and got punished. Cloud revenue surged. Didn’t matter. Management raised full-year capital expenditure guidance to $195 to $205 billion, up from $180 to $190 billion the week before, and the shares fell 7%. Free cash flow turned negative for the first time in the company’s public life, with management signalling debt and equity issuance to plug the gap. Capital expenditure, for the uninitiated, is money spent on physical assets, in this case data centres, chips, networking and power. Free cash flow is what’s left after that spending, and it’s the number that tells you whether a business is funding itself or borrowing to grow.
Microsoft went next and got rewarded, up 8.13%. Same enormous spending, $41 billion in the quarter. Different framing. Satya Nadella’s team said Azure demand exceeds supply, and Amy Hood extended the assumed useful life of data-centre buildings from 15 to 25 years, which lowers reported capex by roughly $15 billion. The market bought the story: this isn’t speculative spending, it’s demand-constrained spending. The distinction is everything.
Meta reported the same day and got hammered despite 28% revenue growth to $60.8 billion. EPS came in at $6.18 against $7.27 expected, dented by $2.4 billion in legal costs and $1.18 billion in severance for 8,000 layoffs. The number that should worry you: free cash flow collapsed from $8.5 billion to $784 million, because $31.1 billion of capex consumed almost all of the $31.9 billion in operating cash flow. Meta is now spending essentially everything that it earns.
Amazon delivered the first $200 billion quarter in its history, net sales of $200.61 billion, up around 20%, and rose about 10%. Apple fell roughly 4%, which tells you the market had already priced its restraint after that record-territory run in mid-July.
Three companies guided forward capex: Microsoft around $175 billion for calendar 2026, Meta $130 to $145 billion, Alphabet $195 to $205 billion. Add Amazon’s trailing $173 billion and you’re looking at roughly $675 to $700 billion of annualised AI infrastructure spending from four companies.
Which Brings Us to the Thing Nobody Wants to Call a Climate Story
Here is the part that I keep coming back to, and it’s the part most earnings coverage skip entirely.
Seven hundred billion dollars of capital expenditure isn’t an accounting line. It’s a concrete, steel, silicon and, above all, electricity. Every one of those data centres needs power, constantly, regardless of whether the wind is blowing. When Microsoft says demand exceeds supply, part of what’s in short supply is the grid itself.
Climate finance used to be a separate conversation, a panel at COP with its own vocabulary. It is no longer. It’s sitting inside the capex line of the most valuable companies on Earth, and It’s being discussed on every one of these calls under different names: energy costs, supply chain, power procurement, depreciation schedules on assets that run hot.
The green finance industry has a genuine problem with this and hasn’t solved it. Hyperscalers buy renewable power at enormous scales, which makes them look excellent on paper against the standard sustainability metrics. But procurement doesn’t create new generation fast enough. It reallocates existing clean supply, and the shortfall gets filled by whatever else is on the grid. A company can be simultaneously the largest corporate buyer of renewables and a net driver of fossil generation, and nothing in the current disclosure framework forces that contradiction into the open.
The number I’d want on every one of these calls, and which nobody asks for: how many megawatt-hours, from what source, at what marginal cost to the grid everyone else uses.
Tuesday: SpaceX Opens the Books
On the 4th of August, SpaceX reports for the first time as a public company. Twenty-four years private, IPO’d on the 12th of June, and now subject to quarterly scrutiny like everyone else. This is the one I’ll be watching most closely.
What we know from the S-1 and Q1: revenue of roughly $4.69 billion in Q1, of which Starlink was $3.26 billion, about 69% of the total. Starlink is the only segment making real money, $1.19 billion of operating income in Q1, with full-year 2025 operating income of $4.42 billion on an EBITDA margin around 63%. EBDITDA, roughly, is profit before interest, tax and the accounting charges for wearing out assets. Useful for comparing operating performance, dangerous if you forget those charges are real.
The rest is bleeding. SpaceX lost nearly $5 billion in 2025 on $18.7 billion of revenue, and roughly $4.28 billion in Q1 2026 alone. Rocket launches lose money. The AI segment, which now includes compute leasing to third parties, made $818 million in Q1 against enormous build costs.
The consensus for Q2 is about $6.87 billion. But here’s the trap, and it’s the most interesting number in this entire article: the banks’ full-year forecasts range from roughly $38 billion to over $50 billion. Hit even the conservative end and the second half needs to produce around $26.6 billion, more than double the first half’s pace. The optimistic case implies $34 billion, roughly $17 billion a quarter.
That acceleration isn’t coming from Starlink or Falcon 9. It’s compute leasing, renting out xAI data-centre capacity to companies like Anthropic and Google at extraordinary monthly rates. Which means SpaceX, a rocket company, is now partly an AI infrastructure landlord, and its valuation depends on it.
My prediction for Tuesday: Starlink beats, subscriber numbers look strong, and the stock still struggles unless management gives hard, contracted visibility on second-half compute revenue. Morgan Stanley wants subscribers going from 10.3 million to 16.8 million by year end. Even if they get there, the H2 maths only works if the compute contracts are signed, not hoped for. Watch for whether they name counterparties and contract lengths. If they don’t, that’s an answer in itself.
What Else I’m Watching, and Why
Beyond the obvious names, a few reporting soon that tell you about the wider economy rather than just their own patch.
Visa and Mastercard are the cleanest read on whether consumers are actually holding up, because they see transaction volumes before anyone else. NextEra Energy matters as the largest clean-energy generator in America, sitting exactly where the AI power demand story meets the renewable build-out. ASML is the chokepoint for the entire semiconductor supply chain; if its order book softens, the AI trade has a supply problem that nobody’s priced. And Exxon and Chevron will show whether the energy sector’s earnings surge is a genuine trend or just an oil-price artefact.
I’d add one framing that has served me well: read the guidance before the results. The quarter that just happened is history and mostly already in the price. What management says about the next two quarters is the thing that moves shares, and it’s usually buried three-quarters of the way down the release.
Why Any of This Reaches East London
Two ways, and neither requires you to own a single share directly.
If you have a workplace pension, you own a slice of all of this through index funds. When Meta’s free cash flow drops 91% in a quarter, some of that is your retirement. Most people in Newham and Tower Hamlets holding an auto-enrolled pension have never read an earnings call and have no idea how much of their future is currently riding on whether Azure demand exceeds supply.
The slower one matters more. That $700 billion of data-centre spending is landing somewhere physical, and increasingly that somewhere is the eastern edge of London, pushing into the Docklands and Essex. These facilities compete for grid capacity with everything else, including housing developments. The capital-allocation decisions being narrated on American earnings calls end up as decisions about which local projects get connected and when. That’s not an abstraction. It’s a queue, and East London is in it.
Where I Land
My honest read: this was a strong quarter dressed up as a spectacular one, and the market is finally doing the thing I’ve wanted it to do all year, which is distinguish between companies spending heavily with visible demand and companies spending heavily on faith. Microsoft got rewarded for the former. Alphabet and Meta got punished for looking like the latter, despite growing fast. That’s healthy. That’s a market waking up.
For the rest of 2026, analysts expect earnings growth of 27.4% in Q3 and 25.2% in Q4, with the forward P/E at 19.6, slightly below the five-year average of 19.9. Price-to-earnings, if you’re new to it, is what investors pay for each pound of expected profit. At 19.6 this market is not cheap, but it isn’t the dot-com fever dream either.
My prediction: the capex versus cash flow gap will become the defining story of the next two quarters, not AI revenue itself. Somebody, and I’d bet on Meta or Oracle before the mega-caps, has an uncomfortable moment about funding it with debt. And the first company to properly quantify the energy cost of its AI build-out, honestly and in public, will get more credit from investors than it expects.
Buy the dip. Short the Vix. And read the capex line before you do either.
