Good morning. Astonishing amounts of market value went poof! yesterday after investors came to terms with the broader implications of DeepSeek, the low-cost Chinese AI company we wrote about yesterday (and last week, and in Fortune’s Eye on AI newsletter, and…).
Nvidia shed so much market capitalization that it was the largest such haircut for any company on a single day in U.S. history. Its chipmaking peers recorded double-digit drops in their stock prices, too.
What’s the deal, as Jerry Seinfeld might ask? Strap in, then read on. It’s an (almost) all DeepSeek edition today, ‘cause that’s all anyone is talking about…again. —Andrew Nusca
Want to send thoughts or suggestions to Data Sheet? Drop a line here.
DeepSeek spurs brutal rout of tech stocks, crisis for U.S. AI companies
The surprise success of the Chinese AI startup DeepSeek is hammering U.S. tech stocks and undermining the perceived dominance of U.S. firms in emerging AI technology.
Nvidia, the California chip behemoth that has added nearly $3 trillion in market capitalization over the past three years, shed roughly $600 billion in market cap while its stock dropped 17% as part of a broader rout of the tech sector.
Broadcom saw its stock plunge over 19%, while U.S.-listed shares of TSMC dropped more than 15%. The stock of TSMC supplier ASML fell nearly 8%. Shares of Oracle, one of the main investors in the newly announced $500 billion Stargate Project, plunged roughly 15%.
DeepSeek says its new large-language model was developed in just two months for under $6 million, or about 3% to 5% of what it reportedly cost OpenAI to develop its next-generation o1 counterpart. Last week, DeepSeek released a report that showed its model matching or exceeding o1’s performance on several critical benchmarks.
Wedbush Securities’ Dan Ives, one of Wall Street’s most noted tech bulls, says that while some of DeepSeek’s technological innovations are impressive, the company will not be able to replicate the infrastructure and ecosystem of America’s tech giants. No major U.S. company, he said, will use DeepSeek to launch their AI architecture and use cases.
“Launching a competitive LLM model for consumer use cases is one thing,” he wrote in a note Monday morning, but “launching broader AI infrastructure is a whole other ball game.” —Greg McKenna
DeepSeek flips the AI script on open source
What does the DeepSeek fallout mean for the most deep-pocketed AI startups, OpenAI and Anthropic, as well as highly funded competitors like Mistral and Cohere?
There are indications that DeepSeek may have used outputs from OpenAI’s o1 to kick-start the training of its R1 model’s reasoning abilities. This process of analyzing and learning from another model’s outputs is sometimes referred to as “reverse engineering.”
Open-source developers have been reverse-engineering closed OpenAI models like o1 for months, AI developer and consultant Reuven Cohen told Fortune.
DeepSeek’s efforts make it clear that models can self-improve by learning from other models released by OpenAI, Anthropic, and others—which puts those companies’ existing business models, cost structures, and technological assumptions at risk.
“The problem is that the companies have momentary advantages but haven’t built durable moats,” said Patrick Moorhead, founder of Moor Insights & Strategy.
Proponents of open-source AI have long predicted the commoditization of AI models. “If these models turn out to be pretty capable, which they really are looking like, and they’re very cheap, then there’s a world where companies stop using OpenAI at scale,” said William Falcon, CEO of Lightning AI.
But Vaibhav Srivastav, a researcher at open-source platform Hugging Face, does not think OpenAI, Anthropic, and other model companies are in deep trouble.
“I think the real moat is in the application layer,” he said, meaning that the value for these companies lies not just in building models but in how those models are integrated into applications. But “I do think DeepSeek must be a humbling moment for them.” —Sharon Goldman
DeepSeek may not be the danger to Nvidia and U.S. export controls many assume
The prognostications of Nvidia’s doom at the hands of DeepSeek may be premature. So, too, may be claims that DeepSeek’s success means the U.S. should abandon policies aimed at curtailing China’s access to the most advanced computer chips used in AI.
DeepSeek’s impact could, counterintuitively, increase demand for advanced AI chips—both from Nvidia as well as its competitors. The reason is partly due to a phenomenon known as the Jevons Paradox, which dictates that when technological progress made the use of a resource more efficient, overall consumption of that resource tended to increase.
One of the things that has slowed AI adoption within big organizations has been how expensive these models are to run, making it hard to realize a positive ROI. But DeepSeek’s models are so inexpensive to run that companies can now afford to deploy them for many more use cases. This may cause overall demand for computing power to skyrocket, even as each individual computation requires far less power.
What of Nvidia, then? While its top-of-the-line GPUs are optimized for training the largest large language models, the company has less of an edge when it comes to what AI researchers and developers call inference— using a fully trained AI model to perform a task.
Here, Nvidia rivals including AMD and Groq appear to have the upper hand. Alphabet’s Google and Amazon’s AWS also build their own AI chips, some of which are optimized for inference. These companies could begin to erode Nvidia’s dominant market position, though it’s unlikely to lose its dominance quickly.
If overall demand for AI chips increases due to Jevons Paradox, Nvidia’s overall revenues could still continue to climb, even if its market share drops, as it will own a smaller percentage of a larger, and growing, pie.
For these reasons, it probably still makes sense—if the U.S. sees it as a national security priority to make it more difficult for China to compete on AI—to continue to restrict the country’s access to the most cutting-edge computer chips.
As Miles Brundage, an AI policy expert who recently left OpenAI, put it on a recent podcast: “There are all sorts of ways of turning compute into better performance, and American companies are currently in a better position to do that because of their greater volume and quantity of [advanced] chips.” —Jeremy Kahn
More data
—Expect Trump tariffs on foreign chips in the “near future,” the prez says.
—Pebble is dead; long live Pebble. The Google-owned smartwatch pioneer is reborn as an open source operating system..
—Apple OS updates arrive. New releases for iOS, iPadOS, macOS enable Apple Intelligence by default, disable news AI summaries.
—Google Maps will make Trump’s changes—“Gulf of America,” “Mount McKinley”—once the U.S. government does.
—Reid Hoffman reveals Manas AI, a drug discovery startup founded with researcher Siddhartha Mukherjee and focused on select cancers.
—Banks eager to unload debt from X, the service f/k/a Twitter, may find it easier given the company’s stake in Elon Musk’s xAI.
—Samsung’s Project Moohan headset gets the hands-on treatment from Marques Brownlee.
Endstop triggered
