As the cost of AI infrastructure continues to increase, Morgan Stanley says Big Tech capital expenditures are on track to hit new records
Meta is on track to spend $250 billion on AI in 2028, according to Morgan Stanley.
Investors who worry that Big Tech companies will pull back on their artificial-intelligence spending should think again.
The cost of the AI race is on track to rise even higher as companies like Meta Platforms (META), Amazon.com (AMZN), Alphabet $(GOOGL)$ $(GOOG)$, Microsoft $(MSFT)$ and SpaceX $(SPCX)$ scramble to bring more capacity online amid supply-chain bottlenecks, according to Morgan Stanley analyst Brian Nowak.
In a Monday note, Nowak raised his capital-expenditure forecasts for those five hyperscalers - and is now projecting a combined $1.2 trillion in 2027 and $1.4 trillion by 2028. The new estimates are up 9% and 10%, respectively, from previous levels.
This level of investment is expected to quadruple the amount of total available hyperscale compute capacity to 120 gigawatts by 2028, Morgan Stanley predicts.
Meta and Amazon, in particular, are driving the entire group's spending higher. Morgan Stanley raised Amazon's total companywide capital expenditures for 2027 and 2028 by 15% and 29%, respectively - with the company now projected to spend $308 billion in 2027 and $318 billion in 2028. Meta's capex estimates rose 29% for 2027 to reach $225 billion. and 22% for 2028 to $250 billion.
On Monday, Meta raised the expected cost of its data center being built in Louisiana to $50 billion, from the $27 billion that was announced in October.
With hardware shortages increasing the time between breaking ground to opening data centers to as much as three years, and impending political uncertainty heading into the 2028 presidential election, Nowak believes hyperscalers are increasing their budgets to overcome these obstacles. Increasing social and political backlash to data-center development could be leading Big Tech companies "to start building sooner to emphasize job creation," Nowak noted.
Alphabet, Amazon, Meta and Microsoft are already on track to spend $700 billion on their AI ambitions in 2026 alone. However, the cost of building AI infrastructure is on track to rise due to a shortage of high-bandwidth memory. Nowak estimates that memory components could comprise up to 25% of the total cost for a rack of Nvidia's (NVDA) latest Vera Rubin chips. Beyond the chips, increasing lead times for electrical and mechanical equipment and a shortage of skilled labor are pushing up prices as well.
As these companies spend more, they'll need to deliver "materially incremental, durable and profitable revenue growth" to justify their investments, Nowak wrote. And as they begin reporting their second-quarter earnings results later this month, capex levels and commentary surrounding return on investment will be in the spotlight.
Investors haven't been convinced thus far, as fears of AI overspending led to a broad selloff in the "Magnificent Seven" names in June. The Roundhill Magnificent Seven ETF MAGS sank 9.1% last month - its worst monthly performance since it dropped 10.5% in March 2025 - before bouncing 4.5% so far in July.
Earlier this month, investors speculated that Meta would be scaling back its AI spending on reports of the company developing a new business line to sell its compute to third parties. Some interpreted such a move as a signal that Meta was seeing lower-than-expected demand for its own AI products.
But on the contrary, Nowak has named Meta a top pick among the hyperscalers.
"Meta still has the most call optionality, where the market is penalizing them for the spend and not giving them credit for potential revenue from the spend," he wrote.
Nowak sees new monetization opportunities in the company's recent Muse Spark 1.1 model. It's Meta's first competitive agentic coding model and is available for developers to use through an API. He estimates that every 100 megawatts of compute allocated toward Meta's API could generate as much as $8 billion of revenue and roughly $2 of earnings per share by 2028.
-Christine Ji
This content was created by MarketWatch, which is operated by Dow Jones & Co. MarketWatch is published independently from Dow Jones Newswires and The Wall Street Journal.
(END) Dow Jones Newswires
July 13, 2026 14:36 ET (18:36 GMT)
Copyright (c) 2026 Dow Jones & Company, Inc.
Comments