Google raised its 2026 capital expenditure guidance to as much as $205B this week, up from a $190B ceiling it set just one quarter ago. The lower end of the new range, $195B, already sits above what the company previously called its top-end spend. The $15B swing landed in the middle of earnings season and set the tone for what Wall Street expects from Meta, Amazon, and Microsoft when they report in the coming days.
The problem isn't just the size of the number. It's that Google moved the number by $15B in ninety days, which reads to investors as an admission that the company cannot pin down its own AI infrastructure costs. Google is also spending more than it is earning on that build-out, while facing pricing pressure from Chinese models that are shipping competitive results despite tighter access to Nvidia GPUs.
“Google has essentially said that it can't accurately forecast its costs, which is a scary thing.”— Elizabeth Lopatto, Senior Reporter, The Verge
That combination — rising spend, static or falling model prices, and credible low-cost competition out of China — is what has flipped the mood on AI capex from bullish to nervous. Every extra dollar of data-center spending has to be recouped from a market where the per-token price is trending down, not up.
Key facts
- 01Google raised its 2026 capex projection to as much as $205B, up from a $190B ceiling last quarter — a $15B increase.
- 02The lower end of Google's new range, $195B, already exceeds last quarter's top-end forecast.
- 03Nvidia has been in deal talks worth a combined roughly $750B, including a $250B guarantee of OpenAI debt.
- 04SpaceX shares have fallen to roughly half their peak value following its public listing.
- 05Meta, Amazon, and Microsoft report earnings this week and are widely expected to raise capex guidance as well.
Google is not alone. Meta, Amazon, and Microsoft are all expected to disclose higher-than-forecast AI infrastructure spending when they report this week. The hyperscaler cohort has spent the past 18 months in a capex arms race, and the market has largely rewarded it. This is the first earnings cycle where analysts are openly modeling the downside.
Around the same numbers, Nvidia has been running rounds of deal talks worth a combined roughly $750B, including a $250B guarantee of OpenAI's debt. The scale of that guarantee is what has some investors reading the arrangement as strained rather than strong.
“as much a reminder of funding strain in the AI build-out as it is a demand signal”— Billy Leung, Tech sector investment strategist, Global X Management
Billy Leung, tech sector investment strategist at Global X Management, told Bloomberg the Nvidia-OpenAI deal is as much a reminder of funding strain in the AI build-out as it is a demand signal. The concern is circular financing: Nvidia sells chips to customers whose ability to pay depends on financing that Nvidia itself is now backstopping. If underlying demand were as strong as the topline suggests, the guarantee would not be necessary.
Oracle sits in a similar spot. Its data-center buildout is heavily leveraged, and in public markets Oracle functions as a proxy for OpenAI, since OpenAI's compute commitments run through it. Investor unease about Oracle's debt load is, in practice, unease about whether OpenAI can grow into the capacity being built for it.
SpaceX offers another data point that has nothing to do with AI directly, but a lot to do with market sentiment. Its shares are trading at roughly half their peak, a signal that the market is willing to reprice a Musk-branded high-flyer sharply and quickly. Investors watching AI names are noting the pattern.
The Chinese variable makes the math harder. Chinese start-ups keep releasing models that perform competitively with US frontier systems despite export controls on the highest-end GPUs. If Chinese labs can approach US performance at a fraction of the compute, the case for spending $200B a year on hyperscaler capacity gets weaker, not stronger. It also implies the industry may be overbuilding data centers relative to what monetizable demand actually requires.
Even AI bulls concede that some overbuild is coming. The bullish thesis has always been that a wave of AI companies will die when the correction arrives, and the survivors will more than compensate. That thesis is intact. What has changed this week is that the question of when the correction begins is being asked out loud, on earnings calls, by sell-side analysts who three months ago were still cheering every capex increase.
One quarter's guidance revision does not make a cycle top. Google could grow into the $205B number, Meta and Microsoft could reassure the market later this week, and the Nvidia-OpenAI arrangement could look like ordinary vendor financing in retrospect. Chinese competition may also plateau as the export-control regime tightens further.
But the AI trade has been priced for a straight line up and to the right, and Google just introduced a wobble that the market cannot ignore. The pressure now shifts to Meta, Amazon, and Microsoft to show that their own capex increases come with revenue that scales alongside — and to Nvidia to demonstrate that its deal-making reflects demand rather than propping it up. If the next two weeks of earnings look anything like Google's, the phrase that gets repriced isn't AI itself. It's the assumption that infrastructure spending is free money for whoever holds the GPUs.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
