DeepSeek Opens Second Round at $71B — One Month After $50B, Money Goes to Data Centers and In-House Chips
TL;DR
DeepSeek starts new funding round at $71B pre-money valuation one month after $50B round; capital targets self-built data centers and in-house chips.
Only one month after closing its first funding round, DeepSeek has begun early talks this week with potential investors for a new round — pre-money valuation $71 billion. The first round that closed in late May: $7.4 billion, $50B pre-money valuation, led by Tencent with CATL following.
A $21B valuation jump in one month — a pace never before seen in China's primary AI market. The only event in the past month: Liang Wenfeng's personal net worth hit $36B (first time topping the global AI-founder rich list), with no new model release, no major contract disclosure. The valuation jump is pure FOMO — investors afraid of missing allocation in the next round.
Where the money goes, per FT: self-built data centers + in-house AI chips. Reuters earlier this month reported DeepSeek is developing its own AI chips, reducing dependence on Nvidia and Huawei silicon — an enormous capex commitment. Training an R3-tier model costs $550M-800M, self-building a data center from scratch is a $1B+ order of magnitude, and chip R&D starts at $1.5-2B.
Post #207 covered the Nvidia Asia-buyer whitelist halving — Chinese frontier AI companies like DeepSeek face shrinking access under whitelist review. The $7.4B already raised plus the round now in talks maps precisely to the "H200 unavailable → in-house chips + Huawei Ascend backup" transition budget.
If it works, DeepSeek uses the combined $15B+ war chest to hit self-designed silicon and self-built GPU clusters by 2027, escaping US-China chip crossfire. If it doesn't, in-house chips slip, data centers half-built when cash runs out, and $71B becomes the down-round floor for the next raise.
via Bloomberg / Capacity
A $21B valuation jump in one month — a pace never before seen in China's primary AI market. The only event in the past month: Liang Wenfeng's personal net worth hit $36B (first time topping the global AI-founder rich list), with no new model release, no major contract disclosure. The valuation jump is pure FOMO — investors afraid of missing allocation in the next round.
Where the money goes, per FT: self-built data centers + in-house AI chips. Reuters earlier this month reported DeepSeek is developing its own AI chips, reducing dependence on Nvidia and Huawei silicon — an enormous capex commitment. Training an R3-tier model costs $550M-800M, self-building a data center from scratch is a $1B+ order of magnitude, and chip R&D starts at $1.5-2B.
Post #207 covered the Nvidia Asia-buyer whitelist halving — Chinese frontier AI companies like DeepSeek face shrinking access under whitelist review. The $7.4B already raised plus the round now in talks maps precisely to the "H200 unavailable → in-house chips + Huawei Ascend backup" transition budget.
If it works, DeepSeek uses the combined $15B+ war chest to hit self-designed silicon and self-built GPU clusters by 2027, escaping US-China chip crossfire. If it doesn't, in-house chips slip, data centers half-built when cash runs out, and $71B becomes the down-round floor for the next raise.
via Bloomberg / Capacity
