Artificial intelligence is expected to pay off big for tech giants including Microsoft (NASDAQ:MSFT) and Alphabet (NASDAQ:GOOGL) someday. But expect deeper investments before gains trickle to the bottom line, the companies said on Tuesday.

According to Microsoft, expenses have surged significantly due to the construction of new data centers to bolster AI capabilities. They anticipate that capital expenditures will keep increasing as they procure chips from companies like Nvidia (NASDAQ:NVDA) Corp to enhance the performance of these data centers.

The Windows maker’s shares were down more than 3% on Wednesday, while Alphabet gained nearly 7%.

As analysts explain, Microsoft is spending money on AI in two main ways. First, they are using AI to improve their own products, like the upcoming Copilot AI assistant, which will cost $30 a month. Second, they are also investing in AI to help other companies that want to use Microsoft’s Azure cloud computing services to create their own AI products.

Microsoft executives said the service will start generating the bulk of its revenue in the second half of its fiscal 2024 ending June 30.

“They’re buying a bunch of H100s,” said Ben Bajarin, chief executive and principal analyst of Creative Strategies, referring to Nvidia’s flagship chips for AI.

Amazon (NASDAQ:AMZN) might experience a similar pattern this quarter or the next. Like Microsoft, both companies are top picks for businesses using cloud services to train their AI systems.

Alphabet, however, kept down costs, though not for long. Chief Financial Officer Ruth Porat, who will become president and chief investment officer, said delays in data center construction are why second-quarter capex was lower than expected.

“As far as AI is concerned, while Google may have spent upwards of $200 billion on AI investments over the past decade, much of that isn’t necessarily appreciated by users and investors,” said Scott Kessler, global sector lead for technology media and telecommunications at Third Bridge.

One advantage Google has, analysts said, is its own custom chip for handling AI work called the Tensor Processor Unit (TPU), which helps lower costs.

Microsoft may be “aggressively buying Nvidia chips, given Microsoft does not have its own silicon as an alternative,” said Atlantic Equities analyst James Cordwell.

But Google conceded that it will buy chips from other companies and use its own, and Porat said that spending could drag on profit and growth.

“The message on inflection point was the same,” from Microsoft and Google, said Gene Munster, managing partner at Deepwater Asset Management, “but the difference was Microsoft investors wanted to see more.”