DeepSeek’s financing has effectively stalled this time—this is absolutely the biggest black swan in the AI space this year.

A leaked account of Liang Wenfeng’s internal remarks has exposed this matter; the nature of it is extremely outrageous.
This isn’t just offending a single founder. It has fully laid bare the “family treasures” of domestic compute power—and even the whole supply channel.
Just look at a few leaked hard data points and you’ll know how happy the Americans would be:
1. Total compute power:
DeepSeek has about 20 thousand H100-equivalent cards.
2. Hardware gap:
Huawei’s 950 performance lags Nvidia by 4 times, with a 2-year cycle lag.
3. Model scale:
Tens of billions of parameters in China versus the US’s 800 billion parameters—this gap is enough to make you despair.
What’s even more outrageous is that during the meeting they also discussed how to obtain those restricted chips.
Once this information leaks overseas, it’s almost like a message to the other side:
There are loopholes in the export controls here—fix them quickly.
Liang Wenfeng halting the financing is a mix of anger and self-protection.
For a technology-driven company like DeepSeek, data and compute channels are lifelines.
Now those lifelines have been turned into conversation topics by investment people they trusted and posted online—how could anyone stand that?
Those investment representatives who just wanted to post on Moments and show off that they had internal info this time have genuinely overplayed it.
When trust in business circles collapses completely, the losses may not be limited to this round of financing, but could also shrink the survival space for the entire domestic AI ecosystem.
This investigation will most likely target a few typical cases.