That could describe the reaction to DeepSeek, the China-based chatbot that stunned the tech world and briefly sunk the stock market early last year. It applies equally well to what is happening Monday following the launch of Alphabet’s new Gemini 3.
The company, whose stock is by far the best-performing member of the so-called Magnificent Seven so far this year, unveiled the latest version of its chatbot last week. And investors are just now starting to see the kind of impact it is likely to have on the AI trade over the coming months.
Gemini 3 is reportedly faster, sharper, and performs with deeper reasoning than OpenAI’s ChatGPT, Elon Musk’s Grok, and Perplexity, backed by Jeff Bezos. It fits neatly into the group’s broader family of apps and its market-leading search business. It is also priced at similar levels, or at a discount, to rival AI models.
But the fact that it was trained primarily on Google’s Tensor Processing Units, or TPUs, as opposed to its rivals’ reliance on Nvidia chips, is more important. TPUs aren’t as flexible as Nvidia’s graphics processing units, which can make them less valuable in a market where hyperscalers are demanding the ability to reprogram the systems that they are buying for several billions of dollars.
But they are both cheaper to develop and use less power to run at full capacity. And that is spooking Wall Street.
“Some investors are petrified that Alphabet will win the AI war due tohuge improvements in its Gemini model and ongoing benefits from its custom TPU chip,” said Ben Reitzes, tech strategist at Melius Research.
“It is a bit early to declare that Alphabet’s recent advances make it the AI winner long-term,” he added. “That being said, semis and hyperscaler companies (especially Oracle), need to wake up to the fact that the ‘Alphabet issue’ is a concern.”
Oracle has bought billions of dollars of Nvidia chips to rent out via the cloud. The advent of less costly TPUs means it could be undercut on price if other companies were to establish an AI cloud competitor.
D.A. Davidson analyst Gil Luria estimates a stand-alone division comprised of Google’s DeepMind AI research lab and selling its TPUs would be worth nearly $1 trillion, making it “arguably one of Alphabet’s most valuable businesses.”
On the opposite side, even a modest narrowing of Nvidia’s AI competitive advantages could have ripple effects that could shock markets over the coming months. Companies that had spent big, to say the least, on Nvidia semiconductors could have buyer’s remorse if it becomes clear that less expensive chips can perform equally well.
Valuations are already sky high, from the publicly traded hyperscalers to OpenAI, and doubt remains about how the new technology will benefit the real economy.
CEO Sam Altman, in fact, conceded in an internal memo published by The Information last week that Google’s AI progress is likely to create “some temporary economic headwinds” for the company. “I expect the vibes out there to be rough for a bit,” he said.
Google shares are reflecting a portion of that concern, rising 6% and hitting a record high of $318.57 each earlier in the session. The stock is up 68% this year, compared with a 22% gain for an index of the Magnificent Seven and the 18% advance for the Nasdaq Composite.
Broadcom, its TPU-making partner, was 10% higher in early afternoon, taking its 2025 advance to just under 62%.
Nvidia shares, meanwhile, are moving modestly higher, but remain nearly 9% lower since the start of the month. The gap between its $4.35 trillion in market value and Google’s has narrowed to around $526 billion, the smallest since April, according to Dow Jones Market Data.
Stacy Rasgon, a senior analyst at Bernstein, isn’t as focused on identifying a near-term victor in the current AI arms race as he is on watching it continue.
“We’re not at the point where we have to worry about who’s winning and who’s losing,” Rasgon told CNBC on Monday. “It’s more a case of ‘is the opportunity in front of AI sustainable or not?’.”
“If it is, they’re all fine, if it isn’t, they’re all screwed,” he added.
Write to Martin Baccardax at martin.baccardax@barrons.com
