DeepSeek's leaked investor call: inside the AI Playbook that erased $589 billion of Nvidia's value 📉
Liang Wenfeng walks investors through DeepSeek’s 6x pricing rule, its 20,000-GPU fleet, the Huawei plan to break CUDA, and an AGI roadmap with dates attached.
👋 Hey, Linas here! Every day, I break down 3 stories shaping the future of FinTech & Artificial Intelligence - plus the money movements and trends worth tracking. First time here? 397k+ FinTech and AI leaders get this daily. Join them:
Liang Wenfeng has given roughly two interviews in his life. So when a 3-hour-44-minute recording of DeepSeek’s founder walking his new shareholders through the entire playbook (pricing math, GPU counts, the Huawei bet, a step-by-step AGI roadmap) leaked out of a May 20, 2026 closed-door meeting, it instantly became the most complete picture available of the AI lab that erased $589 billion of Nvidia’s NVDA 0.00%↑ market value in a single trading day in January 2025.
The meeting was Liang’s debrief for the incoming investors in DeepSeek’s first external financing ever: roughly ¥50 billion (~$7.4 billion) at a post-money valuation of about $52 billion, per a regulatory filing surfaced by Caixin, from Tencent, CATL, JD.com and NetEase, with Liang personally putting in around ¥20 billion (~$3 billion) of his own money. Within weeks, a follow-on round was reportedly in early talks at around $70 billion 😳
Before we dive in, a word on sourcing. Everything below comes from a reconciled transcript, an English transcription of the audio, checked against the full Chinese text. The figures are Liang’s own claims, and DeepSeek itself has (unsurprisingly) not confirmed the document’s authenticity. But the details keep checking out. The most telling cross-check: his pricing rule of a ten-month hardware payback, roughly sixfold profit on compute cost, lines up with the 545% theoretical inference margin DeepSeek itself disclosed publicly in March 2025. Names of some competitor models are garbled by speech recognition and are flagged where relevant.
The call breaks down into six stories: the sixfold pricing rule that turns open weights into a moat, the first on-the-record GPU numbers from any Chinese frontier lab, a one-year deadline for breaking NVIDIA’s CUDA lock-in, the org design that has half of DeepSeek’s core researchers labeling data, a dated watchlist of falsifiable claims, and a ranked map of where the opportunity moves next. More importantly, the stakes reach well past DeepSeek, since the same pricing doctrine sets the terms of competition for GLM-5.2, Kimi K3, and the wider Chinese open-weight field.
If you build, operate, or invest anywhere near AI, the rest of this piece will pay itself back faster than Liang’s GPUs.