Chinese AI startup DeepSeek is racing to release its next-generation model, R2, ahead of its planned May debut, aiming to capitalize on the success of its R1 model that sparked a $1 trillion global stock sell-off last month.

The Hangzhou-based firm’s R1, built with less-powerful Nvidia chips, outshone pricier Western competitors, and R2 promises enhanced coding and multilingual reasoning, sources say.

READ MORE: China claims DeekSeek under cyberattacks

Nvidia plummets 14%

DeepSeek overtakes ChatGPT in Apple Store

Meet Liang Wenfeng, the visionary behind DeepSeek

Founded by billionaire Liang Wenfeng of quant fund High-Flyer, DeepSeek has disrupted the AI race with cost-effective models, leveraging a decade of computing investments like the Fire-Flyer II cluster of 10,000 Nvidia A100 chips.

Its innovative approach—using techniques like Mixture-of-Experts (MoE) and multihead latent attention (MLA)—delivers high performance at a fraction of rivals’ costs, pricing 20-40 times cheaper than OpenAI, per Bernstein analysts.

The rushed R2 launch could intensify U.S. concerns over AI leadership, while China doubles down, with dozens of firms and cities integrating DeepSeek’s tech. As Western giants like OpenAI and Google adjust pricing to compete, DeepSeek’s rise—backed by a flat, research-driven culture—may reshape the global AI landscape.

‍