Key Takeaways
- Despite the success of a Chinese start-up’s $6 million AI model, tech giants Microsoft, Alphabet, Amazon and Meta are unlikely to change their plans to spend hundreds of billions of dollars on AI infrastructure this year, experts say.
- Major cloud providers are expected to spend a quarter of a trillion dollars on capital goods in 2025, with much of that spending going toward data centers and related infrastructure.
- More efficient models could lower the cost of running AI and encourage the development of more applications, ultimately supporting the need for more data center capacity.
The meteoric rise of Chinese start-up DeepSeek may have shaken Wall Street’s confidence in some favorite trades, but it’s unlikely to change the immediate outlook for the spending that has fueled the AI rally, experts say.
“We do not expect companies to present significant shifts in their capital allocation priorities around AI on the back of recent events,” wrote Goldman Sachs analysts in a note on Tuesday.
Goldman estimates U.S. cloud providers will spend about $270 billion this year on capital expenditures, with much of that going toward data centers and related infrastructure. Just this month, Meta (META) projected capex of $60-$65 billion this year, Microsoft (MSFT) reaffirmed its plans to spend $80 billion, and the White House hosted the announcement of a joint venture between OpenAI, Oracle (ORCL), and Japanese tech investor SoftBank that could spend up to $500 billion in the next four years.
Is All the Spending Really Necessary?
Yet the surprising success of DeepSeek’s open-source AI model, reportedly developed in less than 2 months at a cost of about $6 million, has prompted Wall Street and Silicon Valley alike to wonder whether all that spending was really necessary, and if it needs to continue.
The doubt arises at an inopportune time for America’s tech titans. The companies spending big on AI have faced questions over the last year about the sustainability of their spending and when they’ll see a return on their investment. To make matters worse, most of the Magnificent Seven stocks were trading at or near record highs before the DeepSeek shock roiled markets on Monday.
The question of return on investment will likely take center stage on Wednesday afternoon when Meta, Microsoft and Tesla (TSLA) report quarterly results.
Angelo Zino, Senior Vice President and tech analyst at CFRA, agreed with Goldman that Meta and Microsoft were unlikely to change their spending outlook, but noted that executives’ commentary could give markets another jolt. If they nod toward slowing spending or indicate that they’re looking into being more efficient, he said, “I think that has an impact in terms of how these chipmakers react.”
How Much Will AI Demand Grow?
The degree to which hyperscalers sustain current spending could to a large degree depend on how much AI demand grows. “If all of a sudden we see a massive uptick in demand here in 2025 related to agentic AI and maybe even cheaper large language models, then these companies will continue to be aggressive with the capex,” Zino said.
Analysts expect that more efficient inferencing inspired by DeepSeek could dramatically reduce the cost of AI, lowering the barrier to entry for developers and encouraging the development of more consumer and business applications. The economics of AI could ultimately express the Jevons paradox, when technological advancement makes a resource more efficient to use and subsequently increases consumption of that resource.
Lower AI pricing could spell trouble for large language model developers like OpenAI or Amazon-backed Anthropic, but greater demand would likely benefit cloud providers. “Those companies will benefit from having the ability to rent more GPU capacity to whoever wants to rent that space,” said Zino.
