The AI gold rush just hit the part where the pans run out and only the miners with real claims survive.

The Summary

The Signal

Vis Raghavan runs banking at Citi, which means he sees the deal flow before it becomes news. When he says AI needs $3.5 trillion through 2030, he's not speculating. He's looking at the pipeline of data centers, chip fabs, energy infrastructure, and model training costs that someone has to write checks for. That number is not evenly distributed. It never is.

The phrase "flight to quality" is banker-speak for: the dumb money phase is over. For two years, anything with "AI" in the deck got funded. That window is closing. Capital is consolidating around companies with actual revenue, defensible moats, or infrastructure so fundamental you can't build the future without it.

"When a Citi vice chair says survival of the fittest out loud, he's telling you the bodies are already in the hallway."

This matters because of what comes next. Anthropic's IPO is the canary. If public markets give it a monster valuation, every AI company with a path to profitability will sprint toward an exit. If it disappoints, private money stays private longer and the funding barbell gets more extreme:

  • The top tier (OpenAI, Anthropic, maybe three others) gets infinite capital
  • The infrastructure layer (compute, energy, chips) gets project-financed like utilities
  • Everyone else fights for scraps or dies

The $3.5 trillion figure also reveals something: this is not a software story anymore. Software scales cheaply. This is a physical buildout. Data centers. Power plants. Semiconductor fabs. That kind of capital doesn't come from venture funds. It comes from sovereign wealth, pension funds, infrastructure investors, and debt markets. Those players don't bet on 100 moonshots. They pick three winners and build around them.

The Implication

If you're building in AI, your next six months depend on whether you're in the quality tier or not. Can you show revenue? A path to profitability? Infrastructure someone else needs? If not, your funding window is narrowing fast. If yes, you've never had more leverage with investors.

For everyone else: watch Anthropic's IPO pricing. It will tell you whether public markets are ready to absorb AI's capital needs or if this stays a private game for another cycle. And if it stays private, that $3.5 trillion concentrates in fewer hands than you think.

Sources

Bloomberg Tech