AI Spending Is Top-Heavy: Top 1% of Users Drive 80% of Revenue

Everyone is bullish on AI. The semis are ripping, $SPCE-adjacent data center plays are printing, and Elon Musk is out here casually floating a $3.5 trillion merger like it's a Tuesday. But a pair of CNBC segments this week quietly dropped a data point that should make every AI investor at least a little uncomfortable: the top 1% of AI users account for 80% of all AI spending, averaging nearly $10,000 per month. The median user? Under $20.

That is not a growth curve. That is a barbell, and one side is almost empty. CNBC's coverage of AI's "top-heavy" spending problem, sourced partly from leaked S-1 documents and Apollo data, reveals that roughly 80% of those power spenders are venture-backed private startups, names like Ripple, Sierra, Lovable, and Cursor. Not Fortune 500 enterprises. Not governments. Startups burning VC cash. And per Apollo's most recent data, growth in AI spending has essentially flatlined for the most recent month, possibly because the big labs have been cutting prices to chase volume that isn't materializing the way the bulls expected.

This connects directly to the CoreWeave credit drama that's been circling the AI infrastructure trade. The NeoCloud sector's funding model leans heavily on expensive project financing, and as Bloomberg Podcasts noted, supply of AI-infrastructure debt is already far outpacing demand, with hundreds of billions more expected to hit the market. When your biggest customers are VC-dependent startups and your capital structure looks like a leveraged buyout, any tightening in private funding markets becomes a systemic risk, not just a stock-picker problem.

The bull case isn't dead. AI revenue is reportedly doubling every 7.2 months per Seeking Alpha, CPU lead times are tight, and cloud providers from AWS to AMD are raising prices, which signals real demand pressure, not manufactured scarcity. $META's blowout quarter, with $64.2 billion in Q3 revenue up 26% year-over-year, shows that when AI actually improves ad targeting and engagement at scale, the monetization is very real. That's enterprise AI working. The question is whether the 99% of users paying under $20 a month ever graduate to the power-user tier, or whether the whole model depends on a narrow band of well-funded startups that could evaporate with one bad rate cycle.

Fundstrat's coverage on YouTube this week added another uncomfortable macro layer: the US deficit is running at 6-7% of GDP during full employment, a level historically reserved for wars and recessions. If fiscal conditions tighten and VC funding dries up, the AI spending pyramid loses its base faster than any bear case currently prices in.

Broadening from the few to the many is how every transformative tech eventually justified its valuation. The internet did it, mobile did it. AI still has to prove it can.