The article profiles Liang Wenfeng, founder of DeepSeek, and explains how his lean, efficiency-focused AI approach disrupted the industry by delivering competitive models at a fraction of the cost, drawing significant market attention and funding by 2026.

DeepSeek’s technical strategy centers on a mixture of experts that activates only a subset of parameters per token, resulting in 671 billion total parameters but roughly 37 billion active per token, enabling cheaper inference and training; the V3 model reportedly cost about USD 5.6 million in GPU time to train, and the company has collaborated on sparse attention techniques to handle long text with less memory.

DeepSeek released R1 in January 2025, an open-source reasoning model that matched top rivals on many tests at a lower running cost, helping drive a decline in NVIDIA’s market value amid concerns about the cost of scaling expensive hardware.

Looking ahead, DeepSeek’s open models and efficiency-first approach challenge the notion that massive budgets are required for strong AI, with strategic moves into robotics and ongoing competition and talent retention as key hurdles.

Context and reception: DeepSeek’s model design, pricing strategy, and openness have influenced industry players and sparked broader questions about the traditional reliance on large compute budgets for AI progress.

Lean operations and pricing: DeepSeek emphasizes a non-overtime work culture, a focus on a few core tasks, and low pricing with aims for reasonable profits; Liang Wenfeng reportedly owns about 81.7% of the company after the first outside round.

Funding milestones: In 2026, DeepSeek raised about USD 7.4 billion in its first outside round (with Liang investing around USD 3 billion) and a second round reportedly nearing USD 15 billion, with investors including CATL and Tencent.

Origin of DeepSeek: Liang Wenfeng used profits from his hedge fund High-Flyer to launch DeepSeek in 2023 with the goal of pursuing artificial general intelligence through low-cost, efficient models.