Pinned Tweet
in late 2025 (Q3 of FY2026), NVIDIA moved from counting revenue by billing or invoicing location to counting revenue by headquarters (HQ) location, a much less geographically informative methodology
Cassandra Unchained retweeted
This, NVIDIA’s changing of methodology, and Huang’s desire to get chips to China, as alleged this morning in the WSJ shows both incentives and means were in place for unsavory actions.
Cassandra Unchained retweeted
Medicinal Whiskey Charity is the Cassandra Unchained November 2026 Charity of the Month
New Trading Post November 16, 2026
michaeljburry.substack.com/p…
Unicus Research @unicusresearch out with "A Gold Rush Gone South" detailed "state of the data center" run down with pics, on the ground reporting across seven major data center sites . MUST read. Excerpts below and there is so much more - you will not read or hear about about data centers the same way again.
"One final sleight of hand worth mentioning is Project Jupiter’s claims about its local water usage....$ORCL obfuscates and states that they will not be using “potable” water. The problem is that the well draws from the same aquifer as that being drawn upon for drinking water... Project Jupiter appears to be Ground Zero for developer shenanigans... Recent reports have circulated that Oracle has considered trucking in natural gas to Project Jupiter as an emergency measure – “the same technique Oracle has used for over a year to fuel a data center in Salt Lake City.”"
🚨Our boots on the ground deep dive research on datacenter will be published this upcoming week.
🚩What we found on the ground about $ORCL's Project Jupiter goes beyond what the national media is reporting.
If you are an investor or an allocator in the AI space, this is a must read report.
Access below to pre-order👇
Another week, another Anthem
Compression, The End of Moats youtu.be/5aCqEOwIUSg?si=NhtV… via @YouTube
Nvidia’s interim CEO is expected to rein in deals after a stretch in which the company aggressively financed computing deals and startup investments wsj.com/tech/ai/nvidias-days… via @WSJ
In light of everything, it is worth a re-read of the Culper report to understand seriously worrisome the whole $NVDA scenario is.
culperresearch.com/wp-conten…
This, NVIDIA’s changing of methodology, and Huang’s desire to get chips to China, as alleged this morning in the WSJ shows both incentives and means were in place for unsavory actions.
The press is now, finally, doing its job decently well. Sad that it took a lurid affair to snap everyone out of it.
How Jensen Huang built the backdoor route that keeps Beijing in the AI race. wsj.com/tech/jensen-huang-ba… via @WSJ
This is no surprise to analysts who years ago questioned SE Asia/Singapore’s need for so many advanced $NVDA chips.
The real reveal would be evidence that NVDA booked more revenue and profit from Asia than it had let on.
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo….
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
Trading Post November 12, 2026: Enough is Enough, $NVDA, SMH vs SOXX, 100% Human
michaeljburry.substack.com/p…
The $NVDA Board would not have chosen Colette Kress during this crisis if it believed she were complicit. I imagine those that work with her know her to be beyond reproach. At the same time, the Board's inability to elevate, from inside, someone with sufficient operating experience is in itself concerning.
According to reports out of TechCrunch and Axios, the Don had as many as 60 direct reports and delegated little.
Heaven's on Fire
Kiss - Heaven's On Fire youtu.be/EZjevnnkA20?si=j71q… via @YouTube
My take on Anthropic's amended S-1/A Prospectus michaeljburry.substack.com/p…
Considering the Payments Processing stocks in the wake of a potential deal for Global Payments, Retail Stocks amidst the worst year for consumer sentiment ever, and Hong Kong's Calling Again
Compression, The End of Moats youtu.be/5aCqEOwIUSg?is=-azt… via @YouTube












