As Microsoft Corp. and Meta Platforms Inc. prepare to announce their quarterly results this week, they’re likely to face some hard questions from analysts about their staggering investments in data centers, chips and training for artificial intelligence (AI) in the wake of DeepSeek’s claims of a lower-cost reasoning model.
Big Tech executives, too, may be questioning themselves after more than two years of rampant spending. The overnight emergence of DeepSeek has unleashed panic within the most-valuable tech companies and AI startups after the Chinese company said its R1 reasoning mode out-performed those of OpenAI and Meta but at a fraction of the cost.
A major sell-off of about $1 trillion in tech stocks roiled markets Monday that underscored concerns over whether big bets on AI will lead to success for companies and investors in Silicon Valley and other major tech hubs.
“The panic-talking to investors non-stop about this DeepSeek model over the last few days is driven less by what model they built and instead how they built it,” Wedbush Securities analyst Dan Ives said in a note Tuesday.
Large U.S. cloud companies will spend $250 billion this year on AI infrastructure, analysts estimate. Venture capital firms threw some $130 billion into AI startups in 2024, or about one in every three venture dollars spent, according to PitchBook.
Meta Platforms Inc., which is splurging $65 billion on AI this year, has constructed four war rooms of engineers to determine how DeepSeek produced a presumably breakthrough reasoning model for less than $6 million, The Information reported. Two teams will attempt to determine how DeepSeek lowered the cost of training; one will try to determine what data DeepSeek used to train its model; and the other will consider how Meta’s Llama can restructure its models based on attributes of the DeepSeek models, the news outlet said.
Microsoft, meanwhile, is on pace to invest $80 billion in fiscal 2025 to build AI-enabled data centers to train models and deploy AI and cloud-based applications around the world. The software giant has invested at least $13 billion in OpenAI. Amazon.com Inc., conversely, has injected $8 billion into Anthropic.
At the extreme bookend of DeepSeek is Stargate, a project announced last week by President Donald Trump and tech executives with the audacious goal of raising up to $500 billion for data center expansion in the U.S. over several years, though the project comes with major question marks.
“The market naturally will worry about demand growth in computing power,” Jefferies analysts wrote in a note.
With so much money being pumped into AI, as well as the boasts of the companies spending it, DeepSeek has created a temporary existential debate over the wisdom of their AI profligacy and evoked discussion of Jevons Paradox, a well-known economic principle that states increased efficiency can lead to increased consumption.
Indeed, the price of using AI models has been falling with rising competition and advances in the technology, Bernstein Research analyst Stacy Rasgon said in a note to investors Monday. DeepSeek has priced its models up to 40 times lower than OpenAI’s comparable models, but Rasgon added DeepSeek spent more money to build its system than it claims.
Trump said he considered the low-cost model to be “very much a positive development” for AI overall, because “instead of spending billions and billions, you’ll spend less, and you’ll come up with, hopefully, the same solution,” he remarked.
Tech executives said DeepSeek’s apparent overnight success story has only served to increase chatter over the large gobs of money going into AI products before they are widely adopted within enterprises and bring in revenue.
“It is an insufficient, weird thing to boast about how much you spend on AI training and inference, which is just a fraction of the innovation,” Phil Libin, tech serial entrepreneur and current CEO of mmhmm, said in an interview. “The advantage is not about more money, but efficiency and talent. The competitive advantage of massive investments this early is not a reliable defensive barrier. DeepSeek has burst the bubble of strictly having access to capital.”
Steve Wilson, chief product officer at Exabeam, added the narrative of DeepSeek “shows us that just buying more GPUs and power plants isn’t the only route to success. “We have the chance for open communities and academia to step up and find more efficient ways to ramp up AI. Ultimately, this is still a field where smarts and creativity are critical,” Wilson said in an email.
Still, big tech CEOs remain committed to spending aggressively, says Daniel Newman, CEO of The Futurum Group, who spoke to a few on Monday. The executives noted the necessity of “taking the best techniques from the AI research community and adding them to known model development,” Newman said on LinkedIn.
“The opportunity for the West is to keep our foot on the pedal and be absolutely sure we are outpacing, not slowing down to a lower common denominator because there is one and it may be cheaper,” Newman wrote. “Let’s embrace our powerful hardware architectures, our software innovation, and the craftiness of China and others to go even faster.”
