Big Tech's AI Capex Is Burning Cash: Who Wins and Who's Toast

There's a scene in every war movie where the general points at a map and says "we're winning" right before everything goes sideways. Big tech's AI spending narrative feels exactly like that right now. $AMZN, $MSFT, and $GOOGL all posted jaw-dropping earnings, but strip out the mark-to-market gains on their stakes in Anthropic, OpenAI, and SpaceX, and the S&P 500's apparent 50% year-over-year earnings growth collapses to something closer to 29%, per LSEG data covered in CNBC's recent earnings breakdown. That's still great. But it's not what the headlines implied.

The raw capex numbers are genuinely staggering. Hyperscalers collectively burned through over $700 billion on AI infrastructure this year. Alphabet reported negative free cash flow of $5.9 billion in Q2, the first time in its 22-year public history, after raising its 2026 capex target to $195-$225 billion. Meta's free cash flow cratered 91% year-over-year to just $784 million. Microsoft and Amazon showed more resilience, but even they are cash flow negative on AI capex and expected to stay that way. NYU valuation professor Aswath Damodaran, appearing on CNBC, put it plainly: even the Mag-7 are showing a notable drop-off in marginal returns on invested capital from heavy capex spending.

The bull case is real and not imaginary. AWS grew 36.7% year-over-year to $42.2 billion, its fifth straight quarter of acceleration and fastest growth in 18 quarters, pushing Amazon past a $3 trillion market cap for the first time. AWS CEO Matt Garman confirmed AI revenue is running at a $25 billion annual rate, driven by broad enterprise adoption across financial services, healthcare, retail, and media. This isn't one or two giant clients propping up the number. The demand is wide.

But here's the thing nobody wants to say at the strategy off-site: the classical yield on that $700 billion hasn't materialized yet. Gary Cohn, appearing on Bloomberg, noted the yield curve has steepened roughly 60 basis points in the two-to-ten spread, which is organically doing inflation-fighting work without a Fed rate hike. That's fine for now. It's less fine if capex keeps compounding and AI revenue growth decelerates. Damodaran's warning about consolidation among smaller AI players is the canary in the coal mine: when the shakeouts hit, it won't be Amazon that goes first.

The social sentiment picture across YouTube and X is rapidly bifurcating: retail is euphoric about Palantir's 150% U.S. commercial growth and Amazon's $3 trillion milestone, while institutional analysts are quietly flagging free cash flow compression and the distortion of VC mark-to-market gains in reported earnings. Both camps are partially right, which is the most annoying possible outcome for anyone trying to make a clean call.

The AI capex race isn't a bubble or a sure thing. It's an arms race where the survivors will look brilliant in hindsight, and the casualties will look obvious. The problem is nobody knows which is which until the shooting stops.