The price of AI FOMO

Ever paid extra to get your hands on something before it sold out? SoftBank just learned how expensive that strategy can get when the stakes are in the billions.

The Japanese tech investor reported a $7 billion quarterly loss today, including a $3.7 billion hit tied to equipment it rushed to secure for US data centers, in its first net loss in nearly two years.

As it turns out, getting ahead of the AI gold rush can come with a hefty bill. In today’s investment blitz, companies are competing for more than chips—they’re also racing to secure the electricity and equipment needed to power AI, often years before those investments generate returns.

The early bird gets the bill

SoftBank made $9.5 billion in advance payments for power equipment and manufacturing capacity for US data centers, including a massive Ohio campus intended to serve OpenAI.

After reassessing its plans, the company wrote down $3.7 billion in nonrefundable payments and related costs, though future rebates could recover up to $2.1 billion.

The fallout reached Wall Street, where shares of GE Vernova, a potential turbine supplier for the project, fell following the news. But AI hardware demand hasn’t disappeared: TSMC reported record October revenue, up nearly 60% from a year earlier.

The ripple effect

SoftBank’s bigger gamble involves borrowed money. The company raised $56.5 billion through financing arrangements last quarter, while its financing costs more than doubled.

Its biggest bet is OpenAI. SoftBank has invested $64.6 billion for a roughly 13% stake in the ChatGPT maker, hoping an eventual IPO will deliver a substantial payday. But AI valuations are facing a reality check, with rival Anthropic’s reported IPO valuation expectations falling to $1.4 trillion today, just a shade of its earlier $2 trillion ambitions.

That leaves SoftBank exposed on two fronts: Its infrastructure bets depend on projects moving forward, while its startup investments depend on future valuations holding up. Rising borrowing costs or construction delays could make both harder to justify.

You don’t need to own SoftBank to feel the consequences. US-listed chipmakers, power-equipment suppliers, utilities, and cloud companies are exposed to the same AI spending cycle, while many tech giants occupy a sizable share of Americans’ retirement portfolios. For investors, record chip sales are proof of strong demand, not guaranteed returns. The real question is whether today’s enormous AI spending will generate tomorrow’s profits.

FOMO is expensive. AI FOMO just comes with a few extra zeros.—HC