Big Tech's 2026 AI Capex Surge: Why Wall Street Split 2-2

Between July 22 and July 30, 2026, Alphabet, Meta, Microsoft, and Amazon all reported earnings, and all four raised or reaffirmed enormous increases to their AI infrastructure spending for 2026. If bigger AI budgets were simply good or simply bad news, all four stocks should have moved the same way. They didn't. Alphabet fell roughly 5%. Meta fell 8 to 10%, wiping out about $175 billion in market value. Microsoft rose on its results. Amazon jumped as much as 10% in after-hours trading. Same theme, four different verdicts from investors in the space of eight days.

That split is more revealing than any single earnings report, because it shows what Wall Street is actually pricing in 2026: not the size of the AI spending number, but whether a company can prove the spending is chasing demand it already has, rather than demand it hopes to have.

Four Earnings Reports, One Week Apart

The figures above come from earnings coverage by CNBC, Investing.com, and company disclosures reported across multiple outlets this week. Guidance has shifted more than once at some of these companies over the course of 2026, so treat the numbers as the latest snapshot rather than a fixed budget.

Why Alphabet and Meta Got Punished

Alphabet's quarter, on paper, was strong: revenue of $119.8 billion, up 24% year over year and above the $116.9 billion analysts expected, with Google Cloud revenue surging 82% to $24.8 billion, according to CNBC's earnings coverage. But the company also raised its 2026 capex forecast to $195-205 billion, a $15 billion increase at the midpoint from its April guidance and well above the roughly $188 billion analysts polled by Visible Alpha had modeled. The stock fell anyway, because the raise arrived without a matching signal of near-term payoff — investors read it as an open-ended acceleration rather than spending tied to visible, contracted demand.

Meta's case was more straightforward: it missed. Earnings per share came in at $6.18 versus $7.14 expected, and the company raised its 2026 capex guidance to $135-145 billion in the same report. Chief Financial Officer Susan Li attributed part of the increase to higher memory-chip pricing and added data center costs. JPMorgan analyst Doug Anmuth downgraded the stock to Neutral, citing concern that Meta is entering another cycle of heavy investment with an unclear path to profitability for AI outside its advertising business. When a spending raise lands next to a profit miss, investors have no beat to weigh it against — so the spending stands alone, and it read as risk.

Why Amazon and Microsoft Got a Pass

Amazon's raise looked, on the surface, like the largest of the four: $220 billion, up from an original 2026 guidance closer to $200 billion. But it came attached to a $496 billion backlog of contracted, not-yet-delivered cloud demand and AWS revenue growth of 37%, well past the 31% analysts had forecast. On the earnings call, CEO Andy Jassy said Amazon would "still not have enough capacity to meet all the demand we have in 2026," and expects the same shortfall in 2027, adding that demand already booked for 2028 is "striking." That framing turns the $220 billion into a number chasing demand that already exists on paper, which is a very different pitch to investors than a bet on demand that might materialize.

Microsoft's case is subtler. Its reported calendar-2026 capex figure of roughly $175 billion looked like a pullback from an earlier $190 billion guide, but reporting indicates the change is largely an accounting reclassification — a change in the expected useful life of data centers and office buildings, effective fiscal 2027 — rather than an actual reduction in AI investment. Microsoft has said its fiscal 2027 capex is expected to rise year over year. Combined with a solid quarterly beat, the market treated the headline capex number as noise rather than a signal, and the stock rose.

The Pattern: Spending Alone No Longer Moves the Stock

Eighteen months ago, essentially any capex raise from a hyperscaler was read as a threat to free cash flow and margins, and stocks fell accordingly almost by default. This week shows that reflex has broken down. What now moves the stock is whether the spending arrives with proof: a backlog, a beat, an accelerating cloud number. Without that proof, as with Alphabet and Meta, the same size of spending increase reads as risk. With it, as with Amazon, it reads as a company racing to keep up with revenue it has already secured.

How Much Money Is Actually on the Table

Add the four companies' latest guidance together — Amazon's $220 billion, Alphabet's $195-205 billion, Microsoft's roughly $175 billion in reported calendar-2026 terms, and Meta's $135-145 billion — and the combined figure lands somewhere around $725 to $745 billion for 2026 alone. Independent tracking from research firm Futurum and Statista put the four companies' combined 2026 AI capex at roughly $725 billion, up about 77% from $410 billion in 2025. However the exact total is sliced, every major cloud provider is spending at a scale that did not exist three years ago, and none has signaled a slowdown for 2027.

The Constraint Nobody's Budget Can Fix

What is striking across all four earnings calls is that none of the companies described capital as their limiting factor. Industry analysis attributes the surge to three overlapping pressures: training costs for frontier AI models that keep rising with compute, inference demand from deployed AI products outpacing existing data center capacity, and a parallel race to build proprietary chips — AWS's Trainium and Google's TPUs among them — to reduce dependence on Nvidia and cut the cost per unit of AI compute.

Power supply is the constraint that gets the least attention but bites the hardest. Large data centers need grid connections, transformers, and often dedicated power generation, and those timelines are set by permitting and utility buildout, not by a company's quarterly budget approval. That is why Jassy's comments about capacity shortfalls stretching into 2027 and 2028 read as a real constraint rather than corporate hedging: money can be committed on an earnings call, but substations cannot be built on the same schedule.

What It Means If Your Business Runs on These Clouds

For companies that consume AI infrastructure rather than build it, this earnings week points to a few practical takeaways:

• Reserved or committed-use pricing is likely to stay more attractive relative to on-demand pricing, since providers want predictable demand to plan capacity against.
• Lead times for large GPU or custom-silicon allocations may lengthen before they shorten — all four companies described 2026 and 2027 demand as running ahead of planned supply.
• Multi-cloud or multi-region flexibility is worth more as a hedge, because capacity constraints are not evenly distributed across regions or chip generations.
• Custom silicon options, such as AWS's Trainium or Google's TPUs, are being pushed harder by providers partly to manage their own capacity crunch, and are worth evaluating case by case against Nvidia-based instances on cost.

Not Everyone Is Convinced

The market's verdict this week was not unanimous even within the winning column. JPMorgan's downgrade of Meta pointed at a risk that applies, to varying degrees, to all four companies: a repeat of past over-investment cycles, where capital gets committed faster than a clear path to return on it. Data center hardware, GPUs especially, depreciates faster than the buildings that house it, and a slowdown in AI demand growth, even a mild one, would leave these companies holding capacity built for a boom that paused. None of the four has disputed that risk publicly; each has bet that under-building is the costlier mistake. Amazon's backlog and Microsoft's beat bought them the benefit of the doubt this quarter. Alphabet and Meta did not get it, at least not yet — itself a sign that investor patience for unproven AI spending is thinner than it was a year ago, not thicker.

The Bottom Line

The headline number from this earnings week was never really the spending — all four companies are now committing well over $100 billion apiece to AI infrastructure for 2026. What actually moved four stocks in four different directions was whether each company could show its spending was following demand it had already won, or simply hoping demand would show up. Amazon and Microsoft made that case. Alphabet and Meta did not, at least not clearly enough for investors in the moment. Whether that judgment holds will depend on whether the backlogs, bookings, and cloud growth rates behind these numbers keep converting into revenue over the next several quarters — the only test that will settle whether this week's winners and losers had it right.

Frequently Asked Questions

Why did Amazon's stock rise while Meta's fell, when both raised AI spending in the same week?

Amazon tied its $220 billion capex raise to a $496 billion backlog of contracted cloud demand and 37% AWS revenue growth that beat expectations. Meta raised its capex to $135-145 billion in the same report where it missed earnings-per-share estimates, so investors had no beat to weigh the spending against and read it as added risk.

How much are Alphabet, Amazon, Meta, and Microsoft spending on AI in 2026 combined?

Adding their latest guidance together puts the combined figure at roughly $725-745 billion for 2026, with independent tracking from research firms estimating about $725 billion, up roughly 77% from $410 billion in 2025.

Did Microsoft actually cut its AI spending?

Not in substance. Microsoft's reported calendar-2026 capex figure of about $175 billion looked lower than its earlier $190 billion guidance, but reporting attributes most of the difference to an accounting reclassification of how long data centers and office buildings are expected to last, not an actual pullback. The company has said fiscal 2027 capex is expected to rise year over year.

What is actually limiting AI infrastructure growth if these companies have the money to spend?

Executives and analysts point to power supply and chip availability rather than capital. Grid connections and dedicated power generation for large data centers take years to build regardless of budget, and GPU and custom-chip supply has not kept pace with all four major cloud providers expanding at once.