DeepSeek required prospective investors to promise they would not recruit its engineers or encourage them to start competing companies, a condition founder Liang Wenfeng laid out directly during a four-hour virtual meeting with backers in May. CNBC reported the term, citing a fundraising-focused outlet owned by 36Kr, and noted it could not independently verify the account. The clause was attached to a maiden round that valued the Hangzhou-based lab at more than $50 billion.
The condition is the story, not the capital. Investor agreements routinely restrict the sharing of proprietary information. They do not usually obligate backers to police their own hiring. DeepSeek, the lab whose V3 model shipped at a fraction of frontier training costs, treated its investor network as a leak in the dam rather than a source of dry powder. That framing tells you how acute the retention problem has become.
The fear is grounded in losses DeepSeek already absorbed. Luo Fuli, a core contributor to the V3 model, left late last year to lead Xiaomi’s MiMo team, which has since released models that beat DeepSeek’s own on several benchmarks. The lab had refused outside funding since its founding to keep its attention on research rather than commercialization. It reversed that stance only after watching key researchers walk to rivals, which makes the no-poaching clause less a flourish than a defensive reflex.
The competition for engineers extends well beyond DeepSeek. Tencent hired Yao Shunyu away from OpenAI last year to serve as its chief AI scientist. The Information reported Monday that Tencent had put $20 million into a new lab founded by Juyang Lin, formerly the lead researcher on Alibaba’s Qwen models, who stepped down from that role in March. 36Kr reported in March that ByteDance had lost two senior AI developers to Tencent. Alibaba separately replaced the head of its DingTalk collaboration app after an internal dispute over the unit’s place in its AI strategy, Bloomberg reported in June. Each move is a wager that a researcher trained at a rival can compress a competitor’s lead in months.
This is what the talent war looks like when capital and engineers become the same asset. A loyalty clause signals that DeepSeek holds real leverage on terms (a $50 billion valuation on a first raise is not weakness on money) while conceding weakness on retention. Asking investors to act as a containment perimeter is a heavy demand, the kind sophisticated backers usually refuse. That DeepSeek apparently got it suggests the opportunity was compelling enough to override the discomfort, or that Liang holds enough scarcity to dictate terms.
A precedent is worth watching here. Yao and Liang both frame China’s path as a full commitment to AGI, and a labor market that prices individual researchers this highly tends to produce more spinouts, not fewer, because the people being hoarded learn exactly what they are worth.
For any operator sourcing talent from China’s frontier labs, the takeaway is concrete: DeepSeek’s alumni pipeline will narrow as legal and social pressure on its backers tightens. Build recruiting plans that do not assume that channel stays open over the coming year.
Reporting by CNBC, published June 18, 2026, which cited a 36Kr-owned fundraising outlet describing the investor meeting and noted it could not independently verify the account.
