A Disruptive Force in the AI Market
Chinese AI startup DeepSeek sent shockwaves through global equity markets last month, triggering a $1 trillion sell-off with the launch of its cost-effective reasoning model, R1. The model, despite being built on less-powerful Nvidia chips, outperformed many Western competitors, sparking concern among U.S. tech giants.
Now, the Hangzhou-based company is accelerating the release of its successor, the R2 model. Originally planned for early May, sources indicate that DeepSeek is pushing for an even earlier launch. While specific details remain undisclosed, the new model is expected to improve coding capabilities and expand reasoning abilities across multiple languages beyond English.
DeepSeek has yet to respond to requests for comment regarding its timeline or strategy.
A Challenge to U.S. AI Dominance
DeepSeek’s rapid advancements are forcing global competitors to reassess their strategies. The R1 model, built at a fraction of the cost of U.S. tech giants’ AI projects, has disrupted the market.
Vijayasimha Alilughatta, Chief Operating Officer at Indian tech services provider Zensar, sees the release of R2 as a transformative event. “DeepSeek’s success at creating cost-effective AI models could spur companies worldwide to accelerate their own efforts, breaking the stranglehold of a few dominant players,” he said.
Beyond commercial implications, R2’s launch is likely to raise alarms in Washington. The U.S. government considers AI leadership a national priority, and DeepSeek’s rapid progress could intensify geopolitical tensions over AI technology.
A Mysterious Founder with an Unconventional Approach
DeepSeek’s rise is closely tied to its enigmatic founder, Liang Wenfeng, a billionaire who made his fortune through the quantitative hedge fund High-Flyer. Known for his introverted and low-profile nature, Liang has avoided media appearances since mid-2024.
Interviews with former employees and industry experts paint a picture of a company that operates more like a research lab than a traditional profit-driven enterprise. Unlike many Chinese tech firms that emphasize rigid hierarchies and grueling work schedules, DeepSeek fosters a collaborative and flexible work environment.
Liang’s approach contrasts sharply with China’s high-pressure tech industry. Instead of adhering to the notorious “996” culture (9 a.m. to 9 p.m., six days a week), DeepSeek employees typically work standard eight-hour days. Liang himself is deeply involved in technical aspects and frequently collaborates with Gen-Z engineers and interns.
“Liang gave us control and treated us as experts,” said former researcher Benjamin Liu. “DeepSeek allowed me to take ownership of critical parts of the pipeline, which was very exciting.”
Betting on Research Over Consumer Apps
While rivals like Baidu focused on developing consumer-facing AI applications to capitalize on the ChatGPT boom in 2023, DeepSeek took a different path. Liang prioritized refining AI models over investing heavily in application development.
This strategic focus has paid off. DeepSeek has become one of China’s highest-paying AI employers, with senior data scientists earning up to 1.5 million yuan annually—almost double the industry standard. The company’s financial backing comes from High-Flyer, which, despite regulatory crackdowns on quant trading, still manages tens of billions of yuan.
The Secret to DeepSeek’s Success: Computing Power
DeepSeek’s rapid advancements stem from High-Flyer’s decade-long investment in AI research and computing infrastructure.
As early as 2020, High-Flyer executives signaled a full commitment to AI, reinvesting 70% of the firm’s revenue into research. The firm spent 1.2 billion yuan on supercomputing clusters, including Fire-Flyer II, which houses 10,000 Nvidia A100 chips—a crucial asset for AI training.
At the time, DeepSeek had not yet been established, leading Chinese regulators to question why High-Flyer was acquiring such vast computing resources. However, authorities ultimately allowed the firm to proceed—a decision that proved crucial when the U.S. banned the export of A100 chips to China in 2022.
With Fire-Flyer II already operational, DeepSeek was able to continue advancing its models despite U.S. restrictions.
Innovation Through Efficient AI Architecture
Some Western AI leaders, including Scale AI CEO Alexandr Wang, have alleged that DeepSeek possesses as many as 50,000 high-end Nvidia chips, despite export bans. However, no evidence has been presented to support these claims.
Former employees credit DeepSeek’s success to its innovative approach rather than sheer computing power. The startup has leveraged techniques like Mixture-of-Experts (MoE) and Multihead Latent Attention (MLA), allowing it to achieve high performance at lower costs.
MoE divides AI models into specialized regions that activate only when needed, optimizing efficiency. Meanwhile, MLA enables simultaneous processing of different information aspects, improving reasoning accuracy.
While companies like France’s Mistral have experimented with MoE, DeepSeek is the first to fully implement the architecture while achieving performance parity with more resource-intensive models.
As a result, DeepSeek offers AI services at a fraction of the price of Western competitors. Analysts at Bernstein estimate that its pricing is 20 to 40 times cheaper than OpenAI’s equivalent models.
A Shifting AI Landscape
DeepSeek’s rise is forcing major players to adapt. OpenAI has slashed prices, Google’s Gemini has introduced discounted tiers, and OpenAI’s latest O3-Mini model emphasizes efficiency over sheer computational power.
Adnan Masood, a lead AI researcher at U.S. tech firm UST, noted that benchmarks indicate R1 processes three times as many data tokens for reasoning tasks compared to OpenAI’s scaled-down models.
DeepSeek’s Growing Political Significance
Even before R1’s global impact, Beijing had taken notice of DeepSeek’s potential. In early 2024, Liang was invited to a high-profile meeting with Premier Li Qiang, where he was positioned as the face of China’s AI sector.
Since then, Chinese government bodies and major corporations have rapidly integrated DeepSeek models. At least 13 city governments, 10 state-owned energy firms, and tech giants such as Lenovo, Baidu, and Tencent have adopted DeepSeek’s AI.
“Xi Jinping and Li Qiang have signaled their endorsement of DeepSeek,” said Alfred Wu, a policy expert at Singapore’s Lee Kuan Yew School of Public Policy. “Now, everyone in China follows suit.”
Global Concerns and Regulatory Challenges
As China embraces DeepSeek, Western regulators are growing wary. Several governments, including South Korea and Italy, have removed DeepSeek from national app stores, citing privacy risks.
“If DeepSeek becomes the default AI model for Chinese state entities, Western regulators may escalate restrictions on AI chips or software collaborations,” warned Stephen Wu, founder of hedge fund Carthage Capital.
Liang has acknowledged that chip restrictions pose a major challenge. “Our problem has never been funding,” he said in a July interview. “It’s the embargo on high-end chips.”
With R2’s imminent release and growing geopolitical tensions over AI dominance, DeepSeek’s impact is only beginning to unfold.