Meta CEO, Mark Zuckerberg and Tesla and SpaceX CEO, Elon Musk
The news surrounding the power and efficiency of China’s DeepSeak roiled tech markets on Monday and landed just in time to give U.S. tech investors a reason to pay particularly close attention to earnings season, which kicks off after the bell on Wednesday.
DeepSeek, an artificial intelligence lab based in Hangzhou, released a reasoning model called R1 last week that outperformed OpenAI’s latest model in many third-party tests. In late December, DeepSeek launched a free, open-source large language model that it said took only two months and less than $6 million to build. The startup was founded in 2023.
While there’s plenty of skepticism surrounding DeepSeek’s claims, the latest pronouncements caused a panic on Wall Street due to the potential implications of significantly cheaper AI models coming sooner rather than later.
Nvidia’s stock price plummeted 17% on Monday, its steepest drop since the early days of the Covid pandemic in March 2020. The selloff wiped out almost $600 billion in market cap, the biggest one-day loss ever for a U.S. company. Shares of chipmaker Broadcom, which has also benefited from the AI boom, also dropped 17%. And Oracle, Dell, Super Micro Computer and Hewlett-Packard Enterprise all saw steep declines because they sell servers packed with Nvidia graphics processing units.
Meta, Microsoft and Tesla all report quarterly results on Wednesday and are among the megacap tech companies spending billions of dollars on AI infrastructure, with the biggest checks going to pay for Nvidia’s GPUs. Apple reports results on Thursday, followed by Alphabet and Amazon next week.
The big immediate issues for those companies with respect to DeepSeek are less about revenue and more about costs. It’s a topic they tend to spend a lot time talking about on earnings calls.
“The sheer efficiency of DeepSeek’s pre and post training framework (if true) raises the question as to whether or not global hyperscalers and governments, that have and continue to invest significant capex dollars into AI infrastructure, may pause to consider the innovative methodologies that have come to light with DeepSeek’s research,” wrote analysts at Stifel in a report on Monday.
Investors hope to get some initial answers in earnings reports this week and next. Nvidia doesn’t report until late February, and Broadcom is expected to announce results in March.
Here’s how DeepSeek could come up on next week’s earnings calls.
Mark Zuckerberg, chief executive officer of Meta Platforms Inc., wears Orion augmented reality glasses during the Meta Connect event in Menlo Park, California, Sept. 25, 2024.
Meta CEO Mark Zuckerberg said last week that his company plans to invest around $60 billion to $65 billion in capital expenditures in 2025 as it bulks up its AI infrastructure. Zuckerberg talks extensively about AI on his company’s quarterly updates, and likes to tout advancements in its open-source Llama family of LLMs.
Given how much Meta spends on Nvidia GPUs, any change in the market dynamics is of deep interest to shareholders. Some tech investors have expressed concern about the hefty amount of spending on AI relative to the immediate returns companies are seeing.
Ralph Schackart, an analyst at William Blair, told CNBC that DeepSink’s recent claims about its allegedly cheaper and quicker-to-train open source AI model “caught everybody by surprise, but it’s hard to know whether that investment is real or not.”
Meta’s stock price rose almost 2% on Monday, garnering a very different response from Wall Street compared to Nvidia and the other hardware companies. The potential of lower costs could be a boon for Meta. But, like Llama, DeepSeek is open source, so there’s also the possible competitive threat if developers gravitate to another model.
The one thing Schackart said investors can be sure of is that Meta will continue investing in AI.
“There’s too much to gain or lose by not investing,” he said.
Satya Nadella, chief executive officer of Microsoft Corp., speaks during the company event on AI technologies in Jakarta, Indonesia, on Tuesday, April 30, 2024.
Microsoft slid 2.1% on Monday, a drop that was narrower than the overall tech market’s decline.
Among the large software and internet companies, Microsoft is closest to the AI boom due to its intimate relationship with ChatGPT creator OpenAI. The roughly $14 billion Microsoft has pumped into OpenAI has been critical for the startup, which needs boatloads of GPUs to build and run its models.
DeepSeek is going up directly against OpenAI. In a post on X late Monday, OpenAI CEO Sam Altman called R1 “an impressive model, particularly around what they’re able to deliver for the price.” He added that OpenAI will deliver better models, but that it’s “legit invigorating to have a new competitor!”
In addition to funneling money to OpenAI, Microsoft is a big spender on its own when it comes to GPUs for its AI products and its Azure cloud-computing service.
Analysts at Morgan Stanley said the rise of DeepSeek is a positive development for the company because of the possibility of lower capital expenditures.
However, Oppenheimer analyst Timothy Horan sees DeepSeek as a negative sign for Microsoft, which “has bet more on proprietary models (OpenAI) and higher-cost hardware,” he wrote in a report.
Microsoft said earlier this month that it plans to spend $80 billion in fiscal 2025 on the construction of data centers that can handle AI workloads.
Elon Musk speaks during the Tesla Annual Shareholder’s meeting in Austin, Texas, on June 13, 2024.
On his social network, X, Tesla CEO Elon Musk, who also runs SpaceX, expressed doubt that DeepSeek’s models were created on the reportedly low budget of under $10 million.
But Musk showed appreciation for DeepSeek’s technical achievements. Musk has a history in the space. He was a co-founder of OpenAI, when it was launched as a nonprofit in 2015, but he’s now involved in a heated legal battle with the company and Altman and is suing to stop OpenAI from transforming into a for-profit entity.
Musk is also founder of AI startup and OpenAI competitor xAI, which is raising billions of dollars, largely to buy Nvidia GPUs.
At Tesla, Musk has been ramping up AI spending as the company works toward building a fleet of self-driving cars. Musk has talked about his company’s sizable AI investments as a key advantage. He said in a post on X in April that Tesla would spend $10 billion for the year “in combined training and inference AI.”
UBS analysts wrote in a note on Monday that, “greater efficiency to both model training and inference can help advance, and eventually proliferate,” autonomous vehicles. However, they added that affordable open-sourced models “could break the narrative that the more capital put into compute and infrastructure, the more likely the probability of success, thereby lessening any ‘moat'” Tesla has established.
Tesla shares fell 2.3% on Monday.
Chief Executive of Apple, Tim Cook gives a thumb’s up during a tour the Apple Headquarters on December 12, 2024 in London, England.
Investors didn’t express any concern about Apple on Monday. The stock rose 3.2% for the day, outperforming all of its closest peers.
Apple is less profligate than its rivals in terms of capital expenditures. In the company’s latest fiscal year, it spent $9.5 billion on capex, 14% less than in 2023.
Unlike some of its tech competitors, Apple doesn’t run a cloud service that rents out computing capacity. The company’s big AI bet so far has been around Apple Intelligence, its suite of software for iPhones and Macs that can summarize text and generate images.
Apple’s suite runs on its own chips, either on the user’s device or on the company’s servers. For training its models, Apple has published research saying that its initial models for Apple Intelligence used Google’s proprietary Tensor Processing Units (TPUs), a chip that’s more limited than a GPU but can handle some AI tasks.
Apple contracts with OpenAI to integrate ChatGPT into its Siri voice assistant, which is powered by pricey Nvidia GPUs. That’s a topic that could be of interest to analysts after the DeepSeek developments. However, while neither company has publicly confirmed the deal underpinning the partnership, analysts have expressed doubt that Apple is paying for the ChatGPT access.
WATCH: How DeepSeek achieved its AI breakthrough
