The immediate catalyst for DeepSeek to embark on a fundraising journey was its founder, Liang Wenfeng's, realization of the remarkable prowess exhibited by Anthropic's Claude Mythos, honed through extensive computing power and data training. This revelation underscored the critical need for financial reserves to stay competitive in the rapidly evolving AI landscape. The timeline of fundraising rumors closely aligned with the release of Claude Mythos' preview version. Following the fundraising, DeepSeek announced ambitious plans to at least double the headcount across all departments. With the current team size hovering around 300 individuals, the core department, Harness, is now fast-tracking its interview process to meet this goal.

Meanwhile, DeepSeek is also focusing on adapting its technology to Huawei's chips, a task that necessitates rewriting the underlying software. This endeavor has led to a 15-month hiatus in new model releases, causing the company to miss out on the recent programming tool craze. However, Liang Wenfeng remains steadfast in his belief that short-term product gains should not be overemphasized. Instead, he advocates for a laser focus on achieving the ultimate objective of Artificial General Intelligence (AGI). AGI refers to the capability of machines to perform tasks at a human level across a diverse range of domains. Liang Wenfeng firmly believes that AI technology should not be monopolized by a select few.

In 2023, DeepSeek faced setbacks in its fundraising efforts due to the absence of a clear commercialization roadmap. Undeterred, Liang Wenfeng personally financed the lab's operations for three years. In this latest $7.4 billion fundraising round, he made a substantial personal contribution of approximately $3 billion, accounting for two-fifths of the total amount. Post-fundraising, the company instituted an employee stock ownership plan to foster a sense of ownership and commitment among its workforce. Liang Wenfeng's strategy remains unwavering: continue with open-sourcing initiatives, maintain competitive pricing, and keep the spotlight on AGI.

As the sole major AI lab that fully discloses the underlying code of all its models, DeepSeek has witnessed its flagship model, V4, and its lightweight counterpart, V4 Flash, swiftly gain traction among U.S. developers. Within a month, the share of token usage for V4 on Vercel's AI Gateway platform surged to 17%, making it the third-largest model in terms of usage. Notably, V4 Flash is priced 20 to 50 times lower than Anthropic's models, and its growth trajectory continued unabated in June, further solidifying DeepSeek's position in the AI market.