DeepSeek, a Chinese AI startup founded in late 2023 by hedge fund manager Liang Wenfeng, launched its R1 reasoning model on January 20, 2025, quickly topping Apple’s App Store charts ahead of ChatGPT and challenging the dominance of leading AI models by offering competitive performance at a fraction of the cost.
DeepSeek R1’s unique approach combining supervised fine-tuning and reinforcement learning allowed it to excel in logical inference, mathematical problem-solving, and chain-of-thought reasoning. The open-source R1, built on a 671B-parameter Mixture-of-Experts (MoE) architecture with only 37B active parameters per pass, excelled in math, coding, and reasoning benchmarks via reinforcement learning without initial supervised fine-tuning, plus cold-start data for coherence.
“AI Sputnik Moment” Hype
DeepSeek’s technical report claimed V3 base training (leading to R1) cost $5.6 million using 2.8 million Nvidia H800 GPU hours versus US rivals’ billions despite US chip export curbs, sparking a $600 billion+ Nvidia stock plunge and “AI Sputnik moment” hype from Marc Andreessen, who declared: “DeepSeek R1 is AI’s Sputnik moment.”
Deepseek R1 is AI’s Sputnik moment.
Marc Andreessen
The launch challenged the prevailing belief that state-of-the-art AI required massive capital expenditure and the most advanced chips, which had primarily benefited US companies. Nearly $1 trillion of market capitalization was wiped from US tech stocks in a single day, with Nvidia’s stock plunging 17%.
Praise for efficiency, skepticism over claims
OpenAI CEO Sam Altman described China’s DeepSeek as an “impressive model, particularly around what they’re able to deliver for the price”, and stated that it was “invigorating to have another competitor.”
Researchers and startups largely welcomed the news, as the open-weight, cost-efficient model lowered the barrier to entry for developing AI solutions. Former Intel CEO Pat Gelsinger praised the launch, noting that “Open Wins” and that the market would expand faster with cheaper AI.
However, some experts and tech figures, including Elon Musk, expressed skepticism about DeepSeek’s claims regarding development costs and suggested the company might have used output from other models like OpenAI’s GPT-4 in violation of terms of service. Microsoft launched an investigation into the training data claims.
US authorities investigated whether DeepSeek accessed restricted Nvidia chips via Singapore intermediaries amid export controls; Singapore’s MTI denied evidence of violations, Nvidia confirmed compliance, but three men were later charged in February 2025 for related fraud involving chip shipments to China.
US President Trump said the release should be a wakeup call for America.
The release of deep seek AI from a Chinese company should be a wakeup call for our industries that we need to be laser focused on competing to win.
US President Trump
R2 stalled, V3.1 update released instead
R1’s launch faced a cyberattack outage amid millions of users, but the May 2025 R2 successor stalled due to Huawei Ascend chip failures, data delays, and quality issues, leading to a V3.1 update instead.
The R1-0528 upgrade in May 2025 improved benchmarks, reduced hallucinations, added JSON/function calling, and enhanced chat interfaces, signaling China’s AI push despite biases in sensitive topics. However, the planned follow-up R2 launch faced significant delays originally slated for May 2025.
China’s high-flyer hedge fund funded DeepSeek
Liang Wenfeng is viewed as China’s equivalent to Sam Altman, championing AI advancement and substantial investments in cutting-edge research. His quantitative hedge fund, High-Flyer, heavily incorporates AI into its trading strategies and provided initial funding for DeepSeek after venture capital hesitated due to long-term horizons.
In 2019, High-Flyer became China’s pioneering quant hedge fund to surpass 100billion yuan ($13.8billion USD) in assets under management.
Read more:
Google launches Gemini 3 and Ironwood
OpenAI launches GPT-5 for all users
DeepSeek launches V3.2 rivaling OpenAI
