Claude Mythos made Liang Wenfeng decide to raise funds.

What was the direct trigger for DeepSeek to start fundraising?

According to the latest news from The Information: Liang Wenfeng saw that Claude Mythos had developed superhuman capabilities through massive computing power and data training, and realized that DeepSeek couldn't continue without stockpiling ammunition.

Anthropic released its first preview in April this year, and rumors of DeepSeek's fundraising first emerged in mid-April.

The timeline matches perfectly.

DeepSeek's 300-person team is set to double

After the fundraising was completed, DeepSeek issued a rare public statement this Thursday:

All departments are hiring "at least double," covering AI system development, infrastructure, product development, and deep learning research.

DeepSeek currently has about 300 employees company-wide.

Among them, the Harness team, the core department responsible for transforming models into autonomous AI agents, has already entered a state of daily interviews.

The team's leader, Cui Tianyi, joined from Jane Street this March and revealed this pace on X earlier this month.

Meanwhile, DeepSeek is accelerating its adaptation to domestic chips. Liang Wenfeng believes that Huawei chips will catch up to Nvidia's level within a few years, and DeepSeek should be the first to complete the adaptation.

Huawei only learned that DeepSeek was privately testing its chips last year and then began direct collaboration.

But the cost of adaptation is significant.

DeepSeek's training and deployment systems are built around Nvidia's CUDA software, and engineers must rewrite the underlying software for models to run efficiently on Huawei chips.

This directly led to DeepSeek not releasing any new-generation models for 15 months—a glaring gap in an era where other top labs release new models every two or three months.

It also caused DeepSeek to miss the programming tool boom sparked by Anthropic's Claude Code in the second half of last year.

But Liang Wenfeng told investors during the roadshow that programming tools and AI chatbots are both temporary products in AI evolution. Heavier bets on these short-term products would deviate from the ultimate goal of achieving AGI.

"AI should not be controlled by a few"

He did meet with investors in 2023, but when he told them that DeepSeek only focuses on deep research and scientific exploration without any commercialization or product roadmap, investors backed out.

In the end, Liang Wenfeng funded the lab himself for three years.

In this $7.4 billion fundraising round, Liang Wenfeng wrote the largest check himself: 20 billion RMB (about $3 billion), accounting for two-fifths of the total.

After the fundraising was completed, DeepSeek established an employee stock ownership plan, distributing shares based on the actual valuation.

Liang Wenfeng told those around him that his strategy would not change: continue open-sourcing, maintain low prices, and focus on AGI. He defines AGI as machines reaching human-level performance on a wide range of tasks including understanding, reasoning, learning, planning, and adaptation.

He also told those around him that AI should not be controlled by a few.

DeepSeek is currently the only major AI lab that fully open-sources the underlying code of all its models.

Among American developers, DeepSeek's presence is rapidly expanding. Its flagship model V4, released in April this year, saw its token usage share on Vercel's AI Gateway platform surge from less than 1% to 17% in May, becoming the third-largest model on the platform after Anthropic and Google within a month.

The lightweight version, V4 Flash, is 20 to 50 times cheaper than Anthropic's models.

This growth continued into June.

Source: QuantumBit

Risk Warning and Disclaimer

        Market risk exists, and investment should be made with caution. This article does not constitute personal investment advice, nor does it consider the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this article is at your own risk.