
The sudden arrival of breakthrough A.I. models from DeepSeek, a Chinese A.I. startup, has drawn ire, envy and admiration from Silicon Valley’s elite. Demonstrating capabilities that rival A.I. leaders like OpenAI, Meta (META) and Anthropic, DeepSeek’s models were developed with a fraction of the costs of their U.S. counterparts—a feat that has caught much of the tech industry by surprise.
DeepSeek’s V3 model, released in December and followed by a new reasoning model earlier this month, was created with only 2,000 of Nvidia (NVDA)’s pricey graphics processing units (GPUs), according to the Hangzhou-based startup, which grew out of the hedge fund High-Flyer just last year. The company’s self-claimed ability to create advanced A.I. models on the cheap spooked investors betting on American A.I. giants. Nvidia shares tumbled 17 percent today (Jan. 27). Stocks of AMD, Alphabet (GOOGL) and Microsoft (MSFT) also fell.
As some tech leaders question whether DeepSeek has more A.I. chips than it’s letting on, others advocate for tougher GPU export controls to prevent China from gaining an edge in the new technology. The venture capitalist Marc Andreessen called DeepSeek’s breakthrough a modern-day “Sputnik moment.” Microsoft’s Satya Nadella praised the company’s efficiency, and Meta’s chief A.I. scientist Yann LeCun lauded its decision to open source.
One thing’s for sure: DeepSeek’s developments have got Silicon Valley talking. Here’s a look at what prominent tech figures are saying about the startup’s surprising success:
Marc Andreessen, co-founder of Andreessen Horowitz
Marc Andreessen, general partner of the venture capital firm Andreessen Horowitz, described DeepSeek’s R1 reasoning model, released on Jan. 20, as A.I.’s “Sputnik moment” in a post on X yesterday (Jan. 26), referencing how the unexpected success of the Soviet Union’s 1957 satellite launch spurred the U.S. to step up its space efforts. DeepSeek-R1 “is one of the most amazing and impressive breakthroughs I’ve ever seen,” added Andreessen, who praised the company’s decision to open source the model as “a profound gift to the world.”
Yann LeCun, chief A.I. scientist at Meta
DeepSeek’s performance isn’t necessarily telling of China’s A.I. capabilities surpassing those of the U.S., according to Yann LeCun, but instead highlights the power of open source. “The correct reading is: ‘Open source models are surpassing proprietary ones,'” the computer scientist said in a recent post on Meta’s Threads.
LeCun additionally called the A.I. stock selloff in reaction to DeepSeek “woefully unjustified,” noting that much of the money tech companies spend on A.I. isn’t necessarily used to train models but to run them.
Dario Amodei, CEO of Anthropic
For some tech leaders, DeepSeek’s breakthrough is a sign that the U.S. should continue cracking down on chip export controls. Dario Amodei recently told CNBC that he believes Chinese A.I. companies have more GPUs than expected, as many were able to procure vast amounts of the hardware before such restrictions were implemented. In the case of DeepSeek, “it’s been reported, at least, that they have a cluster of 50,000,” said Amodei.
According to the Anthropic CEO, export controls must focus on preventing chip stockpiles from growing into the millions. “I think if the United States can’t lead in this technology, we’re going to be in a very bad place geopolitically,” Amodei told CNBC.
Marc Benioff, CEO of Salesforce
To Marc Benioff, DeepSeek’s newfound popularity proves that data, not compute or models, will make the difference for tech companies going forward. “The real treasure of A.I. isn’t the [user interface] or the model—they’ve become commodities,” said Benioff in a post on X yesterday, where he noted that DeepSeek’s success was achieved without vast funds or Nvidia chips. “The true value lies in data and metadata, the oxygen fueling A.I.’s potential.”
Satya Nadella, CEO of Microsoft
While speaking with CNBC in Davos earlier this month, Satya Nadella lauded the Chinese startup’s progress. “I think we should take the development out of China very, very seriously,” said Nadella, who described the DeepSeek models as “super impressive.”
The CEO followed up on his comments today with an X post seemingly linking DeepSeek to “Jevons paradox,” an economic theory that posits heightened efficiency leads to resource consumption instead of reduction. “As A.I. gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of,” he said.