Mark Zuckerberg, founder and CEO of Meta, spoke at the Meta Connect event held at Meta's headquarters in Menlo Park, California on September 27, 2023.
Josh Edelson | AFP | Getty Images
“I think there's a good chance that a lot of companies right now are overbuilding and then looking back and thinking, Oh, we probably spent billions more than we needed to,” Zuckerberg said this week on a podcast with Bloomberg's Emily Chang.
He is not alone in expressing such sentiments.
During Alphabet's earnings call on Wednesday, CEO Sundar Pichai said the company may be spending too much on its AI infrastructure, which consists mostly of Nvidia graphics processing units, but he believes it has little choice.
“As we navigate this curve, the risk of us underinvesting is much greater than the risk of us overinvesting,” Pichai said.
In addition to Meta and Alphabet, NVIDIA has won business from Microsoft, Amazon, Oracle and Tesla, all of which have publicly said that AI investments are a top priority this year and for the foreseeable future. NVIDIA's revenue has more than tripled for three consecutive quarters and is expected to more than double this quarter.
Alphabet and Tesla highlighted the costs of building AI in their earnings calls this week, and investors will hear more next week when Microsoft, Amazon and Meta report earnings.
Meta announced its latest Llama AI model on Tuesday. Called Llama 3.1, it comes in three versions, including one that is Meta's largest and most performant AI model to date. Meta is committed to open source, meaning that outside developers can access the technology for free, even though the company is putting money into the underlying infrastructure.
On his podcast with Chan, Zuckerberg said companies are “making rational decisions” when it comes to investing in AI, despite the prohibitive costs.
“The downside of being late is that we stand to lose out on the most important technology of the next 10 to 15 years,” Zuckerberg said.
In Pichai's view, even though Alphabet is investing heavily, its infrastructure is “broadly useful to us.”
Nvidia shares are up 131% this year after soaring 239% in 2023. The company's market capitalization is now nearly $3 trillion, third largest after Apple and Microsoft, and briefly surpassed both companies' market caps in June.
Nvidia gets more than 40% of its revenue from Microsoft, Amazon, Google and Oracle, all of which need large amounts of GPUs for their public cloud services. These companies are some of the best-capitalized in the world, but investors are growing concerned about their large inventories.
The driving force behind spending is competition, which follows the dynamics of game theory and creates a “cycle of increased competition,” David Kahn, a partner at venture firm Sequoia, wrote in a blog post last week.
“The cloud giants see AI as both a threat and an opportunity, and they can't afford to wait to see how the technology evolves,” Khan wrote. “They must act now.”
Khan calculated that the tech industry would need $600 billion in annual AI-related revenue to justify all the money being spent on data centers and chips.
Khan added Wednesday that comments from Zuckerberg and Pichai about containing downside risks support his theory.
“Within the last 24 hours the CEOs of Google and Meta have both agreed with my view of the AI arms race: AI capex is driven by game theory and FOMO, not real revenue/use,” Cahn posted on LinkedIn.
Jensen Huang, co-founder and CEO of Nvidia Corp., showed off the new Blackwell GPU chip at the Nvidia GPU Technology Conference on March 18, 2024.
David Paul Morris/Bloomberg via Getty Images
Nvidia says demand remains strong through its latest generation of AI chips, Blackwell, which are due to start shipping later this year. But the company is beginning to answer investor questions about returns on investment as growth inevitably slows due to historically difficult comparisons.
Nvidia Chief Financial Officer Colette Kress told investors in May that the company calculated that for every $1 a cloud provider would pay for an Nvidia-based server, it could rent it for $5. Goldman Sachs analysts said in a recent note that Nvidia is seeking to share such data points to instill confidence among investors.
“Demand for NVIDIA hardware is extremely high and GPUs are often hard to come by,” Tesla CEO Elon Musk said during the company's earnings call on Tuesday. Tesla is investing in self-driving cars and humanoid robots, and said its AI capital spending reached $600 million in the quarter.
Musk said Tesla is focusing on building its own Dojo supercomputer because Nvidia chips are expensive and hard to obtain.
““I don't think they have a choice because the demand for Nvidia is so high,” Musk said, “and it's clear that it's essentially their obligation to drive the price of GPUs up to where the market will tolerate it, and the market is so high priced.”
clock: Investors should look to Tesla's earnings drivers