Edward Lee retweeted
Jensen Huang will appear for a transcribed interview with the House Foreign Relations, Intel, and China committees on December 1st, and NVIDIA General Counsel Tim Teter on December 2nd, the committees announce.
👀
Nvidia has hired Heidrick & Struggles to lead its CEO search following Jensen Huang's resignation.
The search is led by Jeff Sanders, and the firm is in talks with potential candidates, including $ARM CEO Rene Haas.
Nvidia is considering both internal and external candidates.
1
89
👀
Both Nvidia and the Commerce Department have now said that Nvidia placed Megaspeed on a "no sell" list in June 2025.
But trade data the Times obtained from @importgenius show that, between July 2025 and February 2026, Megaspeed's Malaysian subsidiary received more than $800 million of shipments containing controlled Nvidia chips.
nytimes.com/2026/11/17/busin…
1
166
Eyes👀
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me.
I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA.
I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation.
Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO.
Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours.
Jensen
69
Bessent wins.
The Trump administration just dropped new remote access rules, no doubt spurred by the Nvidia scandal.
In the first paragraph, they made Commerce publish a line effectively indicating the Administration’s lack of faith in Commerce.
Remarkable.
bis.gov/press-release/bis-is…
3
293
Edward Lee retweeted
The Commerce Department has released a major new export control rule targeting China’s access to advanced AI compute.
It’s pretty good! Here’s what it does:
(1) Restricts Chinese access to overseas AI clusters (seeks to address one of China’s largest and fastest-growing sources of AI compute)
This is a new worldwide end-use control, which we’ve previously recommended (third option in image below).
A license is required if providing cloud AI compute to Chinese or D:5-headquartered countries, or if acting as an intermediary training models for these actors.
Technical requirements are:
- Applies to any data center or group of “linked data centers” or providing more than 12 million aggregate TPP of compute (the equivalent of around 760 H100s) of AI chips meeting the current controlled thresholds.
- “Linked data centers” are defined as sharing physical or virtual computing resources, jointly balancing workloads, or replicating data as a unified whole. So this is intended to capture multi-data center training runs.
This applies everywhere except the US. It also covers existing data centers, which have a 90-day deadline to get a license.
(2) New controls for AI chips going to Malaysia, Singapore, Indonesia, Thailand, and Brazil (seeks to address smuggling of AI chips to China via third countries)
The evidence for large-scale smuggling of AI chips to China via Malaysia and Singapore is strong, and there’s also evidence this is happening via Indonesia and Thailand. Brazil, on the other hand, is a bit of a head-scratcher.
This is a case-by-case license review policy, rather than a presumption of denial (that's the policy for chips going directly to China/D:5 countries).
BIS will evaluate each license on national security grounds.
There’s an exclusion for qualifying US-headquartered recipients, subject to reporting requirements.
(3) Formally repeals the AI Diffusion Rule
This has been a long time coming. This eliminates the country-tier system, the Validated End User authorizations (VEU) for data centers, and export controls on advanced AI model weights.
However, the replacement measures above address many of the gaps created by this, and, of course, the AI Diffusion rule wasn’t actually being enforced.
(4) Tightens controls on Chinese chip designs manufactured abroad
This modifies and strengthens the Biden-era “Foundry Due Diligence” rule, which applies to foundries and OSATs.
This seeks to prevent Chinese chip designers from circumventing export controls through overseas foundries and packaging facilities. It’s an important complement to the measures above.
—
So what are the gaps? More on this to come, but from an initial read:
(1) To determine remote-access eligibility, the rule requires due diligence on direct or indirect Chinese ownership or control. However, its licensing trigger is based around company headquarters and ultimate-parent headquarters, which could allow for some shenanigans around corporate structures.
(2) The rule doesn’t prescribe how AI compute providers must verify customers' identities, or require them to identify and verify every user accessing their compute. The rigor of these checks could vary widely across providers.
(3) The end-use restrictions don’t apply when compute is used exclusively for commercial recommender systems. So this opens a path for large-scale Chinese AI compute usage. This could be fine if verifying this kind of use is possible, but this is likely outside the current technical capabilities of many AI compute providers.
(4) The end-use restrictions don’t apply to data centers in the US. So a Chinese company can still access large amounts of AI compute via US data centers.
This is an interim final rule, meaning it would take effect upon publication in the Federal Register, while allowing 60 days for public comments.
Read the full interim final rule here: bis.gov/media/documents/rin-…
Looks like someone called a code red.
The return of Pax Silica, haven't heard from them in a while. The allied nations have signed the San Francisco Declaration on the Secure Foundations of Enduring Prosperity. The name sounds a little kantselyarit to me, but I think most people probably won't notice.
71
Edward Lee retweeted
Commerce is closing the loopholes.
Here’s my map on the NVIDIA Chip - Megaspeed controversy. Includes reporting by news media on estimated quantities of chips sold. Also Commerce Dept just issued an interim final rule attempting to close the loophole via renting of data center use.
chatgptiseatingtheworld.com/…
2
90
Edward Lee retweeted
NEW: The Department of Commerce just published new AI and semiconductor export control regulations that close some major loopholes in existing export controls. The regulation includes new rules to: (1) counter Chinese smuggling of AI chips; (2) restrict Chinese companies from renting access to AI chips; and (3) block Chinese chip designers from making AI chips at TSMC.
These rules are a major step forward. They still contain some key gaps that need to be filled urgently in a follow-up rule, namely, they do not block Chinese companies from buying AI chips inside the United States or renting access to AI chips at US-based data centers. However, this rule will help and is the first good-faith effort to actually close loopholes in export controls in a year and a half. Hopefully more will follow.
1. Countering Chinese smuggling of AI chips. The rule expands export controls on AI chips to five new countries: Singapore, Malaysia, Thailand, Indonesia, and Brazil. It does not block AI chip orders to these countries, but requires a US government review (unless it is a US company that meets certain requirements). This is a massively needed step. Singapore, Malaysia, Thailand, and Indonesia have been the second, fourth, fifth, and eighth biggest importers of AI chips this year.
The rule doesn’t cover Taiwan itself, where chip smuggling is a massive issue (exports of AI chips to China are still not illegal under Taiwanese law), and still doesn’t prevent AI chip sales to Chinese companies in the United States, which somehow remain legal. Ultimately, large orders of AI chips by untrusted companies in any country should require a license to fully combat smuggling, as Chinese companies will move to buy AI chips anywhere there isn’t a license requirement. But this is a start.
2. Restricting Chinese companies from renting access to AI chips. The rule creates a new remote access control that restricts companies from providing Chinese companies with remote access to AI chips. It requires a US export license to provide Chinese companies with access to 12 million TPP or more (equivalent to ~200 of the most advanced Blackwell chips, or ~800 H100 chips). This applies to chips already shipped, as well as all future shipments. Some of these requirements even apply retroactively; Commerce says existing rules already barred data supplying a majority of a data center’s compute to Chinese companies, and tells companies that violated this to self-disclose. This effectively kills the business of neoclouds like Megaspeed that exist to provide remote access to large amounts of AI compute to China.
The biggest gap is the remote access controls do not apply to data centers in the United States. So Chinese companies can legally train on large numbers of Blackwells at data centers in Northern Virginia. This gap should be closed immediately with a new rule that uses Commerce’s US persons authority to capture this activity. The rule also exempts recommendation algorithms, so impact will depend on enforcement. Last, new authorities included in the text of the AI Overwatch Act would strengthen Commerce's ability to enforce these controls--that bill should pass immediately.
3. Blocking Chinese chip designers from making AI chips at TSMC. The rule closes loopholes in existing controls that Chinese companies are likely exploiting to illicitly make AI chips at TSMC or Samsung. Specifically, it now requires any Chinese chip designer who is seeking to make a 16/14nm or more advanced chip at TSMC/Samsung to get a U.S. export license, and restores old rules that weren’t being enforced until now that require any untrusted company anywhere in the world to get a license to make an AI chip at TSMC or Samsung. This is a significant upgrade.
Overall, this is a major step forward to countering Chinese smuggling of and access to advanced AI chips. The work here is not done, as there are some critical loopholes that remain open and need to be closed in a follow-up rule (namely those related to activities inside the United States). Commerce should move extremely quickly to close these. But this is the first time in over 18 months that the administration is starting to take export controls seriously, and does represent a meaningful step forward.
Last, the fact that the Commerce press release says this rule was “written by and at the direction of the Department of Treasury and the National Security Council” is remarkable and unprecedented. I have never seen a BIS rule that was not written by BIS. That implies that Commerce was dragged kicking and screaming over the finish line by the White House and Treasury, and instructed to publish this rule rather than the rumored weaker one that it preferred. Remarkable.
👀
JUST IN: Nvidia ($NVDA) lacks liability insurance protecting the company itself against shareholder lawsuits leaving the chipmaker exposed to potentially massive legal costs following Jensen Huang’s ouster, per FT.
1
2
381
Edward Lee retweeted
The Denny's where Nvidia was born has removed the sign honoring the company
nbcnews.com/tech/tech-news/d…
What a week in the AI sector. Classes in business schools will study the fallout. Corporate governance a big question mark on many levels. 📌
Jensen Huang is gone. Will the AI industry pay the price? ft.trib.al/dHDbZOQ
2
107
Edward Lee retweeted
Anthropic CEO Dario Amodei on why now:
"The biggest one ... goes back to these compute bets. The more cash we have on hand, the more buffer we have."
"Never let your company be worth a single dollar more than it's really worth."
"We want to underpromise and overdeliver."
Edward Lee retweeted
New bipartisan chip export controls to China are on the way, and they will probably speedrun this with support from both Republicans and Democrats. President Trump will not veto this. The old ways are gone with Jensen.
👀
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.
122
































