Chinese startup DeepSeek saw its AI assistant climb to the No. 1 spot on Apple’s iPhone app store Monday, fueled by global curiosity about the emerging ChatGPT rival. The app’s rapid success has alarmed some U.S. tech analysts, who note the company appears to have matched American leaders in generative AI at a fraction of the cost.
If true, that raises questions about the massive spending commitments U.S. firms are making on data centers and advanced chips to maintain their lead in AI.
Still, experts caution against exaggeration. “The models they built are impressive, but not miraculous,” said Bernstein semiconductor analyst Stacy Rasgon, adding that Wall Street’s reaction was overblown. “They aren’t using any secret breakthroughs—these are tools everyone is testing.”
What is DeepSeek?
U.S.-China AI Rivalry
What Sets DeepSeek Apart
Founded in Hangzhou in 2023, DeepSeek launched its first large language model the same year under CEO Liang Wenfeng, a former hedge fund co-founder. Before U.S. restrictions tightened, the company’s hedge fund arm amassed thousands of Nvidia’s A100 chips. Today, DeepSeek trains its models on Nvidia’s lower-tier H800 processors, signaling that cutting-edge AI may not require the most advanced hardware.
Last month, DeepSeek gained global attention by unveiling a new model said to rival OpenAI’s ChatGPT while being cheaper to operate. The release of a follow-up paper last week, coinciding with Donald Trump’s inauguration, added fuel to the debate by showcasing its R1 model, which demonstrated strong reasoning skills—at a lower cost than OpenAI’s o1.
“The price points rattled people,” Rasgon noted.
The development has sparked geopolitical comparisons, with venture capitalist Marc Andreessen calling it AI’s “Sputnik moment.” He warned that U.S. overregulation could let China seize the advantage.
Other analysts see political timing in DeepSeek’s announcement, comparing it to Huawei’s strategic phone launch during U.S.-China trade tensions. “China wants to show that export controls are ineffective,” said Gregory Allen of CSIS.
Trump himself framed the news positively, arguing it proves the U.S. can achieve results without overspending, though he also urged American firms to stay competitive.
Meanwhile, Nvidia shares fell 17% Monday, even as the company praised DeepSeek’s work as compliant and innovative.
Unlike OpenAI, DeepSeek’s models are open source, making them accessible for developers worldwide—though the training data remains undisclosed. Its R1 model also excels at a technique known as Test Time Scaling, essentially “thinking out loud” and refining its reasoning without new data.
Researchers note OpenAI, Google, and Anthropic have similar capabilities, but DeepSeek’s rapid progress shows how quickly China has narrowed the gap.
“I thought OpenAI was untouchable,” said RAND researcher Lennart Heim. “Now it’s clear the race is much closer than expected.”
Post a Comment