A new player has entered the artificial intelligence arena, and it's causing quite a stir. DeepSeek, a Chinese AI company, has developed a model that rivals, and in some cases surpasses, leading AI models from OpenAI and Meta. This development has sparked significant interest and concern within the American AI industry. Let's delve into what DeepSeek is, what its emergence means for the global AI race, and whether the hype is justified.
Founded just under two years ago by Chinese hedge fund High Flyer, DeepSeek is an AI research lab with ambitions of achieving Artificial General Intelligence (AGI). Based in Hangzhou, the company quickly gained recognition in late 2024 with its open-source releases, notably the large language model "v3". This model outperformed Meta's open-source LLMs and rivaled OpenAI's closed-source GPT4-o.
The release of DeepSeek's R1 model has been particularly noteworthy. This model reportedly matched the performance of OpenAI's o1 model in several math and reasoning metrics, despite a significantly lower training cost of approximately $6 million. DeepSeek R1 is also offered to users at a dramatically lower cost than its American counterparts, with access costing about 95% less than OpenAI.
Marc Andreessen, a prominent American venture capitalist, described DeepSeek R1 as "AI's Sputnik moment," drawing a parallel to the Soviet Union's early lead in the space race. This comparison underscores the potential significance of DeepSeek's achievements for U.S. tech supremacy.
The success of DeepSeek raises a critical question: Can Chinese tech firms match or even surpass the technical prowess of their American counterparts while spending considerably less? The implications are far-reaching, particularly as companies like Google, Meta, Microsoft, and Amazon invest heavily in AI infrastructure.
DeepSeek's claims that it trained its models using only 2,000 second-tier Nvidia chips, especially given U.S. sanctions on exporting advanced semiconductors to China, have been met with some skepticism. Scale AI CEO Alexandr Wang suggested that DeepSeek may have access to a much larger number of high-powered Nvidia GPUs (H100s) but cannot disclose this due to export controls.
Despite current restrictions, experts agree that computing power remains a strategic advantage, and the U.S. continues to lead in this respect for the time being.
DeepSeek's progress suggests a reordering of AI power, potentially leading to a future with multiple competing centers of AI excellence. While OpenAI's "o1 Pro" model is still considered the most advanced, DeepSeek's R1 demonstrates that China is closer to the AI frontier than previously believed.
Miles Brundage, a former OpenAI policy staffer, suggests that the U.S. needs a vision for navigating multipolar AI outcomes now that the landscape is quickly changing.
As AI continues to evolve, it's clear that DeepSeek represents a significant development, one that has the potential to reshape the industry and the balance of power in the global AI race.
