On January 27, the tech-heavy Nasdaq Composite closed down 3.1%, wiping over $1 trillion off the index. This dramatic drop was mainly attributed to what is being called the largest single-day value loss in history for any company—NVIDIA. Once the leading maker of chips powering AI models and the most valuable listed company in the US, NVIDIA’s shares tumbled 17%, eventually being overtaken by Apple. But, at the same time, it just returns it to its valuation in October.
One investor described this moment as the “Sputnik moment” of the 21st Century Cold War between the USA and China—a reference not to the original space-age shock but to the breakthrough R1 model developed by Chinese AI startup Deepseek AI. Founded in Hangzhou by Liang Wenfeng, a “geeky tech millionaire” known for his ambition to build the world’s leading AGI model, Deepseek represents a new paradigm in rapid innovation. After a successful first startup, Wenfeng reinvested his energy and capital—backed by his hedge fund High-Flyer, which manages $8 billion using AI-guided trading strategies—to launch Deepseek in 2023.
Wenfeng’s determination was evident when he acquired large quantities of the reduced-capability H800 chips. This move came after US sanctions banned NVIDIA’s most advanced H100 chips from export to China since September 2022. Although NVIDIA later developed the less powerful H800 chips for the Chinese market, they too were banned in October 2025, forcing Chinese innovators like Deepseek to adapt rapidly.
Deepseek claims that developing its R1 model costs just $5.6 million—an astonishing contrast to the $100 million to $1 billion typically spent on AI training, as highlighted by Dario Amodei, co-founder of leading AI firm Anthropic, last year. Moreover, Deepseek’s decision to open-source its model was strategic. By democratizing AI and allowing greater scrutiny, the company not only built trust among users and stakeholders but also set a new benchmark for performance. Independent research by Artificial Analysis suggests that the R1 model surpasses OpenAI’s o1-mini model across various credible benchmarks and even exceeds the overall quality of models developed by Google, Meta, and Anthropic.
This breakthrough quickly translated into market success. When Deepseek’s AI app launched on the Apple Store on January 10, 2025, it rapidly ascended to become the top free iOS app in the United States by January 27, even surpassing ChatGPT.
The unfolding AI war escalated further on January 29 with Alibaba’s introduction of its Qwen 2.5 Max model—a new contender claimed by many to outperform even Deepseek’s R1 across several benchmarks. As this turbulent future unfolds, these events offer profound lessons for business leaders –
- Execution Over First-Mover Status: Deepseek was neither the first LLM AI globally nor in China. Instead, its rapid, efficient execution as a fast follower allowed it to learn from early movers, avoid their missteps, and iterate faster.
- Strategic Differentiation: Success need not rely solely on feature additions. Instead, Deepseek’s focus on efficient capital allocation and computing efficiency—rather than mere hardware expansion—underscores the value of strategic resource management.
As we witness this dramatic realignment in the tech landscape, the lessons from Deepseek’s journey—of agile execution, strategic differentiation, and efficient capital management—resonate not only in the AI sector but across all industries.