The market is now deciding which AI bets are credible and which are just expensive theater.

The Summary

The Signal

Four companies just committed three-quarters of a trillion dollars to AI infrastructure over the next two years. Google's cloud revenue growth outpaced both Microsoft and Amazon in Q1, the first time it's led the pack since the AI race began. Meanwhile, Meta's stock fell 6.5% despite beating earnings, purely on increased capex guidance. Same announcements, opposite reactions.

The difference comes down to revenue attribution. Google can point to actual cloud customers paying for AI compute. Microsoft can show Copilot seats and Azure AI consumption. Amazon has AWS inference workloads. Meta has... the promise that AI will eventually make Instagram and Facebook ads better. Investors trust Google more than Meta when it comes to AI spending because one sells AI infrastructure and the other is building it for internal use.

"The market is now deciding which AI bets are credible and which are just expensive theater."

This capital deployment dwarfs anything we've seen in tech. For context:

  • The entire U.S. semiconductor industry invested $67B in 2023
  • Global venture capital deployed $285B in 2023
  • These four companies will spend 2.5x that on AI alone in two years

The spending intensifies U.S.-China competition and potentially challenges NVIDIA's dominance as each company builds custom silicon. Google already runs TPUs. Amazon has Trainium. Meta is designing inference chips. When your AI budget hits nine figures, you vertically integrate.

The Q1 earnings painted a clear hierarchy. Google and Microsoft reported strong AI-driven earnings growth, with actual revenue increases tied to AI products. Amazon showed solid AWS numbers. Meta showed engagement metrics and ad performance that "benefited from AI" in ways analysts couldn't quite quantify. Wall Street noticed.

The Implication

Watch what happens to the companies selling picks and shovels to this gold rush. If $725B flows into data centers, power infrastructure, and custom chips over 24 months, the real winners might not be the hyperscalers. They're becoming their own suppliers. NVIDIA's GPU monopoly faces coordinated defection. Data center REITs and power providers face unprecedented demand. Compute arbitrage opportunities will explode as capacity comes online in waves.

For anyone building in the agent economy, this spending telegraphs where inference costs are heading: down, and fast. When Google outspends the GDP of Finland to drive down the marginal cost of intelligence, your AI agent startup's unit economics improve whether you use their cloud or not. Plan your burn rate accordingly. The era of expensive inference is ending faster than most models assume.

Sources

Crypto Briefing | RWA Times | Financial Times Tech