DeepSeek is working on the sequel to its R1 blockbuster.
The Chinese artificial intelligence (AI) startup is rushing development of its latest low-cost AI reasoning model after R1, released in January, matched or exceeded the performance of models from Western rivals OpenAI, Meta and Google, precipitating a $1 trillion stock sell-off.
DeepSeek planned an early May release of R2, its latest edition, but is moving up the date, according to a Reuters report. The new model boasts better coding and the ability to reason in languages besides English.
R2’s imminent release might not only roil concerns in the U.S. over AI leadership worldwide yet again, but is likely to galvanize scores of Chinese companies that intend to start integrating DeepSeek models into their products.
“The launch of DeepSeek’s R2 model could be a pivotal moment in the AI industry,” Vijayasimha Alilughatta, chief operating officer of tech services provider Zensar, told Reuters. DeepSeek’s knack for designing cut-rate AI models “would likely spur companies worldwide to accelerate their own efforts … breaking the stranglehold of the few dominant players in the field,” he said.
“DeepSeek represents engineering ingenuity and the natural evolution of science and technology. Throughout human civilization, innovation has built upon the work of those before us, and this marks another landmark moment in our legacy of progress,” Songyee Yoon, founder and managing partner at venture firm Principal Venture Partners, said in an email.
The AI race between the U.S. and China is being reshaped almost daily as companies from each country throw billions of dollars into product development and integration. Alibaba has pledged $52.4 billion over the next three years on AI and cloud. ByteDance, the Chinese owner of TikTok, has committed some $20.7 billion in capital expenditure for 2025. Meanwhile, Amazon.com Inc., Google, Meta, and Microsoft Corp. plan to outlay some $320 billion in 2025 on AI-related projects, up from $246 billion in 2024 and $151 billion in 2023.
When upstart DeepSeek exploded on the AI scene late last year with a high-performing model it claimed cost only $5.6 million to create, it spawned a Sputnik-like moment, spreading fear, uncertainty and doubt throughout Silicon Valley, where a handful of companies like OpenAI, Google, Anthropic and Meta assumed they had a significant edge.
Since then, those companies and others have doubled-down on funding and rushed models to market to assuage investors. Meanwhile, several research papers have called into question the actual budget to build R1. Market researcher SemiAnalysis estimated the tab was closer to $1.6 billion in hardware costs and a fleet of 50,000 NVIDIA Corp. Hopper GPUs. DeepSeek officials have been conspicuously silent and shared little information with the media for nearly a year.
R1 has also prompted worries over its collection of personal data and security flaws, leading to bans of the model in segments of the U.S. government, and by countries ranging from Australia to Italy.
