$GOOGL (-1,63 %)
$AMZN (-0,7 %)
$META (-0,81 %)

The day before yesterday, Alphabet announced that it will spend 175 to 185 billion dollars on capital expenditure (i.e. primarily AI) this year, a good 60 billion dollars more than expected. A gigantic announcement in the AI race! Yesterday, Amazon went one better: it plans to spend 200 billion dollars this year, 54 billion dollars more than expected. An escalation in the AI race between the mega-corporations. AI efforts in Europe look so tiny that you wouldn't even recognize them as crumbs on the kitchen table.

Four of the largest US tech companies have projected a combined capital expenditure of around $650 billion for 2026 - an incredible flood of money earmarked for new data centers and the long list of equipment they will require, including artificial intelligence (AI) chips, network cables and backup power generators. As Bloomberg reports, "The planned spending by Alphabet, Amazon, Meta Platforms and Microsoft, all seeking dominance in the fledgling market for AI tools, is a boom unlike any seen this century. The estimated spending by each of these companies for this year would represent a high for capital expenditure by a single company in the past decade, according to Bloomberg data.

To find a comparison for the lofty spending projections announced in the last two weeks of corporate earnings releases, you have to go back at least to the telecommunications bubble of the 1990s, perhaps even to the expansion of the U.S. railroad network in the 19th century, or to the U.S. federal government's investment in interstate highway expansion in the postwar years, or even to the New Deal-era aid programs.

The ever-growing numbers - an estimated 60% increase overall over the previous year - signify a further acceleration in the wave of data center construction taking place around the world. The race to build these sprawling facilities, which house racks of humming servers powered by expensive processors, has affected energy supplies, raised concerns about inflated prices for other users and brought developers into conflict with communities worried about competition for power or water. It also increases the risk that construction spending by a small group of wealthy companies, which already account for a growing share of economic activity in the US, will distort overall economic data.

The four companies "see the race to provide AI computing power as the next market where the winner takes all or nearly all," said Gil Luria, an analyst at DA Davidson. "And none of them are prepared to lose."

Last week, Meta announced that capital spending for the full year will rise to as much as $135 billion - a potential increase of about 87%. Microsoft reported a 66% increase in capital expenditure in the second quarter on the same day, beating estimates. Analysts expect the company to spend nearly $105 billion in capital expenditures for the fiscal year ending in June. The news triggered the second-largest one-day drop in the market value of a share.

Alphabet, founded in 1998 in a garage south of San Francisco, rattled investors on Wednesday when it announced a capital spending forecast that exceeded not only analysts' estimates but also the spending of much of the U.S. industry - the company plans to spend up to $185 billion. And Amazon topped that last night with planned capital spending of $200 billion for 2026, which also sent its shares tumbling in after-hours trading.

In contrast, the largest US carmakers, construction equipment manufacturers, railroad companies, defense contractors, mobile operators, parcel delivery companies, as well as Exxon Mobil, Intel, Walmart and the General Electric spin-offs - 21 companies in total - are expected to spend a combined $180 billion in 2026, according to Bloomberg estimates.

Each tech giant has taken a slightly different path to recouping its investment, but their spending is based on the same premise: that OpenAI's ChatGPT and competing AI tools that can generate text and represent elements of human thought will play an increasingly important role for people at work and at home.

Developing the cutting-edge software models that will enable this change is an extraordinarily expensive process, involving connecting thousands of chips that cost tens of thousands of dollars each. Hence the high cost. The expenditure is also based on the notion that the end products will lead to exponentially higher future revenues.

The spending is transforming companies that just a few years ago had a relatively small physical presence, even though their digital services reached billions of people. For much of their existence, Meta and Google parent Alphabet counted their luxurious corporate campuses and office space as a significant part of their real assets. Most of their spending went into salaries and stock options for the engineers and salespeople who worked there.

That is no longer the case. Last year, for the first time in six years, Meta spent more on capital projects than on research and development - mainly on engineers' salaries. The parent company of Facebook and Instagram owned real estate and equipment worth 176 billion US dollars at the end of last year, around five times as much as at the end of 2019.

While the numbers are rising, it is still unclear whether all companies will be able to realize their lofty goals. Since the expansion of data centers, they are already competing for the limited number of electricians, cement lasers and Nvidia chips coming out of Taiwan Semiconductor Manufacturing Co's factories. "There are and will continue to be shortages," Luria said.

There is also the question of how they plan to finance this. Meta and Google parent Alphabet, whose profits come mainly from digital advertising, Amazon, the biggest online retailer and cloud computing provider, and Microsoft, the biggest provider of business software, are each leaders in their industries and have plenty of cash. Their willingness to invest much of that cash in an AI-powered future means those reserves and investors' patience will be tested.