DeepSeek’s AI assistant shot to the top of the free app charts on Apple’s iPhone store, driven by curiosity about this competitor to ChatGPT. Some folks in the U.S. tech scene are worried that this Chinese startup might be catching up to American leaders in generative AI without the hefty price tag.
If that’s the case, it raises questions about the massive investments U.S. tech firms claim they need for data centers and chips to push AI forward.
However, the excitement and misunderstandings surrounding DeepSeek’s tech have also created some confusion.
“The models they built are fantastic, but they aren’t miracles either,” said Bernstein analyst Stacy Rasgon, who follows the semiconductor industry and was one of several stock analysts describing Wall Street’s reaction as overblown.
“They’re not using any innovations that are unknown or secret or anything like that,” Rasgon said. “These are things that everybody’s experimenting with.”
What is DeepSeek?
DeepSeek is a startup that kicked off in 2023 in Hangzhou, China, and launched its first AI large language model that same year. The CEO, Liang Wenfeng, is a co-founder of High-Flyer, one of China’s leading hedge funds that specializes in AI-driven quantitative trading. By 2022, this fund had gathered a bunch of 10,000 high-performance A100 graphics processor chips from Nvidia, which are essential for building and running AI systems. However, shortly after, the U.S. imposed restrictions on selling those chips to China.
DeepSeek has mentioned that its latest models were developed using Nvidia’s H800 chips, which aren’t banned in China, suggesting that top-tier hardware isn’t always necessary for cutting-edge AI research.
The company started gaining traction in the AI scene last month when it unveiled a new AI model that it claimed rivaled those from U.S. firms like OpenAI, the creators of ChatGPT, while being more economical in its use of pricey Nvidia chips for training on large datasets. The chatbot became more accessible when it launched on Apple and Google app stores earlier this year.
However, it was a research paper released last week—coinciding with President Donald Trump’s inauguration—that really stirred things up. This paper discussed another DeepSeek AI model named R1, which demonstrated impressive “reasoning” abilities, like re-evaluating its approach to math problems, and was much cheaper than a comparable model from OpenAI called o1.
“What their economics look like, I have no idea,” Rasgon commented. “But I think the price points freaked people out.”
The ‘Sputnik’ Backdrop
At the heart of the buzz around DeepSeek’s tech prowess is a bigger conversation happening in the U.S. about how to effectively compete with China in the AI arena.
Venture capitalist Marc Andreessen described the DeepSeek R1 as AI’s own “Sputnik moment” in a post on X, drawing a parallel to the 1957 satellite launch that ignited a fierce space race between the U.S. and the Soviet Union.
Andreessen, who has provided tech policy advice to Trump, cautioned that excessive regulation of the AI sector by the U.S. government could stifle American innovation and give China an edge.
However, the spotlight on DeepSeek could also jeopardize a crucial aspect of U.S. foreign policy aimed at limiting the export of American-designed AI chips to China. Some analysts believe this timing is no coincidence.
“The tech advancements are genuine, but the timing of this launch is politically motivated,” noted Gregory Allen, who heads the Wadhwani AI Center at the Center for Strategic and International Studies. He likened DeepSeek’s recent announcement to Huawei’s launch of a new phone during sensitive diplomatic talks about Biden’s export controls in 2023.
“Trying to show that the export controls are futile or counterproductive is a really important goal of Chinese foreign policy right now,” Allen added.
On Monday, Trump remarked that DeepSeek’s innovation was “great because it cuts costs.” While addressing House Republicans in Miami, he labeled the DeepSeek news as “positive” if it holds true, suggesting it means less spending for the same outcomes. He emphasized that this development serves as a “wake-up call for our industries to stay focused on winning the competition.”
Trump also signed an order on his first day in office last week, stating that his administration would “identify and eliminate loopholes in existing export controls,” indicating he plans to continue and strengthen Biden’s approach.
DeepSeek's advancements in AI, achieved with less spending, might challenge the massive $500 billion investment in AI that Trump highlighted at the White House, backed by OpenAI, Oracle, and SoftBank.
Nvidia's stock took a hit, dropping 17% on Monday, but the company praised DeepSeek's efforts, calling it “an excellent AI advancement” that utilizes “widely-available models and compute that is fully export control compliant.”
So, what sets DeepSeek apart?
Unlike competitors like OpenAI, DeepSeek offers “open source” models, meaning essential parts are accessible and modifiable for anyone, although the company hasn’t shared the training data it used.
The standout feature of DeepSeek’s R1 model is what Nvidia refers to as a “perfect example of Test Time Scaling.” This means the AI models can articulate their thought process and use that for further training without needing new data inputs.
“It’s just thinking out loud, basically,” explained Lennart Heim, a researcher at Rand Corp.
OpenAI’s reasoning models, starting with o1, operate similarly, and Heim suggests that other U.S. competitors like Anthropic and Google likely have comparable capabilities that haven’t been made public yet.
However, Heim noted, “this is the first time we see a Chinese company making such strides in a relatively short time. That’s why it’s grabbing so much attention. I used to think OpenAI was unbeatable, but it seems that’s not entirely true.”