The $700 Billion Question: Big Tech's AI Spending Meets a Tougher Crowd

Great numbers, angry crowd

Here's the thing that should make every AI investor sit up: last week Alphabet reported the biggest quarterly profit in its history — $112.1 billion, nearly four times what it made a year earlier — and the stock dropped more than 7%.

Why? Because Google's parent also hiked its 2026 spending forecast by $15 billion. The profit didn't matter. The spending did. That tells you exactly what mood Microsoft, Meta and Amazon are walking into when they report this week.

The Daily Upside compared it to 19th-century Paris, where opera houses hired 'claqueurs' — paid clappers — because the crowd was so hard to please. Right now, Big Tech could use a few. Blowout earnings aren't buying applause anymore. The question investors keep asking is blunt: when does all this spending actually pay off?

Alphabet posted the biggest quarterly profit in its history — and the stock fell 7% because it wanted to spend $15 billion more.

The eye-watering scale of the bet

Add up Alphabet, Microsoft, Meta and Amazon and you get roughly $700 billion in combined capital expenditure — that's spending on data centers, chips and infrastructure — on track for this year alone. Wall Street thinks 2027 could crack $1 trillion.

That's a genuine change in what these companies are. Moody's put it well: they're shifting from 'asset-light' businesses built on software and cloud services to 'asset-heavy' ones that require 'unprecedented levels of investment.' The ratings agency warned this could pressure credit quality across Alphabet, Microsoft, Amazon, Meta, Oracle and CoreWeave — and in some cases lead to negative or diminished free cash flow, meaning they'd burn more cash than they bring in. The six carry about $460 billion in direct debt already.

For now, the giants have pristine balance sheets, so their top-tier ratings aren't in doubt. The canary in the coal mine is Oracle, which Moody's rates just two notches above junk with a negative outlook, and which S&P recently downgraded to one notch above junk. If the boom sours, that's likely where the cracks show first.

Why 'modelmaxxing' changes the ROI math

The reason the spending question got so sharp this year is that the customers buying all this AI compute suddenly have cheaper options. Last year the fashion was 'tokenmaxxing' — employees competing to burn through as many AI tokens (the units used to measure and bill AI use) as possible. Now it's 'modelmaxxing': routing each task to the most efficient, cheapest model that can handle it. Using Claude or GPT for a simple job, as one analyst put it, is like calling an electrician to change a lightbulb.

And the cheap alternatives are increasingly Chinese. DeepSeek, Qwen and Moonshot's newer models typically charge under $5 per million output tokens, versus $25-$30 for US frontier models like GPT-5.5 and Claude Opus. An IDC survey found 47% of US decision-makers at large companies already use a Chinese model for at least one use case. Even Sam Altman admits: 'This is the first year where AI spend has been a big topic... everyone's asking what we can do to help reduce spend.'

The so-what: if customers can do most of their work on models costing a fraction of the US giants' prices, the revenue that's supposed to justify $700 billion in spending looks a lot shakier. There's a catch, though — using Chinese models raises real security and compliance worries (the US recently barred the Defense Department from contracting with firms like Alibaba and Baidu over alleged military ties). So this isn't a simple 'cheaper always wins' story. But it does mean the pricing power US labs were counting on is under real pressure.

What to watch — and why it touches your portfolio

Even if you don't own a single Big Tech share directly, you may be more exposed than you think. AI-themed ETFs exploded from $10.3 billion a year ago to $64.9 billion by the end of June, per J.P. Morgan Asset Management — now the single biggest thematic category, ahead of infrastructure, defense and cybersecurity. Some are doing well (iShares' BAI is up around 30% year-to-date; Global X's AIQ around 15%), but one CIO called them 'sentiment-driven vehicles' rather than set-and-forget retirement holdings.

The volatility is real. Nvidia alone sank to about $165 in April and rocketed above $230 in May. When earnings mood turns — as it did for Alphabet — that swing hits these baskets too.

This week, watch two things. First, capex guidance, not headline profit: markets are punishing spending hikes regardless of how good the quarter was. Second, any signal on whether the giants can defend pricing against cheaper models. UBS's advice for ordinary long-term investors is basically 'stay invested and diversified,' but even they flagged that 'limited visibility on capex beyond 2027' could keep weighing on sentiment. Translation: buckle up for a bumpy earnings week.