Wall Street grows skeptical as AI giants reveal soaring capital costs

Google's latest earnings season revealed a troubling signal for investors: the company raised its capital expenditure forecast to as much as $205 billion from a previous projection capped at $190 billion. Even more striking, the lower bound of the new range surpasses what had been Google's maximum estimate. For shareholders, this represents not merely a budget increase—it signals that one of the world's largest tech firms cannot reliably predict its own costs.

The situation grows more complex when considering that Google is now spending more money than it's making, while simultaneously facing pricing pressure from competitors including Chinese AI startups and pressure to keep model costs competitive.

The broader ecosystem strain#

These pressures extend beyond Google alone. Meta, Amazon, and Microsoft are all reporting earnings this week, and market observers anticipate similar announcements of higher-than-expected infrastructure spending.

Several concurrent developments suggest investor nervousness is spreading:

• SpaceX's public shares have dropped to nearly half their peak value
• Oracle faces credit concerns tied to AI-related debt—particularly significant since Oracle functions as the market's proxy for OpenAI
• Nvidia has finalized financing arrangements worth approximately $750 billion combined, including a deal where Nvidia guarantees OpenAI's $250 billion debt

That last arrangement is revealing. According to Billy Leung, a tech strategist at Global X Management, such financing "is as much a reminder of funding strain in the AI build-out as it is a demand signal."

Cracks in the cost advantage myth#

Further unsettling investors, a Chinese startup recently unveiled a competitive AI model despite theoretical constraints on accessing advanced chips. If Chinese developers can achieve competitive results without the same GPU access available to US firms, it undermines a key assumption: that expensive hardware dominance guarantees market leadership. This raises an uncomfortable question about whether the industry is building far more data centers than it actually needs.