The Information reports, citing sources, that DeepSeek has decided to abandon its founder-funded model and launch its first funding round of over 500 billion yuan (about $7.4 billion). The direct trigger was Anthropic's security model Mythos, demonstrated in April.
Mythos is capable of finding and exploiting software vulnerabilities. DeepSeek's R&D team assessed that against models requiring massive compute and data stacking — even with national-security-level destructive power — relying solely on Liang Wenfeng's personal wealth would not keep up in the compute arms race. They had to bring in external capital to expand their compute pool.
Last year, to cope with US export controls, DeepSeek tried migrating its training and deployment systems from Nvidia's CUDA ecosystem to Huawei chips. Rewriting the underlying software delayed the next-generation model by 15 months.
DeepSeek previously operated with an extremely flat research management structure. There were no HR or PR departments — all researchers reported directly to Liang, and were encouraged to maintain only six to eight hours of efficient research daily. With the first massive funding round closing, the team is now starting to build professional departments for legal, finance, procurement, and HR.
As DeepSeek's R&D approaches the frontier, regulatory scrutiny has increased significantly. Liang must report before leaving the country, domestic travel includes security personnel, and some core researchers are restricted from free travel and required to hand over their passports.