Pinned Tweet
Some personal news.
Today, I’m leaving The Atlantic after almost 17 years and moving my writing to Substack.
It would be convenient, for the purposes of crafting an exciting departure announcement, to have a dramatic exit story: a fight, a grievance, a shouting match with an editor that ended with me hurling a bunch of leather-backed Thoreau volumes across the open-plan office. That is not the case here.
I love The Atlantic, and I'll remain a contributing writer there. But after almost two decades at one publication, I wanted to write for myself. The things I've published that I'm most proud of—whether it was the original abundance agenda essay, or my piece on workism—emerged from a very personal expression of frustration, or confusion, or curiosity. I want to know what my thinking and writing is like if I lean into a more independent and personal writing life.
That's brought me to Substack, which is already home to an astonishing share of my overall reading. I'm excited to join their community and excited to build my own. The name of the newsletter should be easy to remember: Derek Thompson.
The newsletter will have three main pillars
1. Abundance
2. The frontier of science and technology—GLP1s, AI, biotech, energy breakthroughs—covered in a way that’s both curious and skeptical
3. The anti-social century & the social crises of anxiety and aloneness
Thanks to The Atlantic for 16.8 incredible years and thanks to everybody who follows me across the river.
- dt
Nasdaq is up 5% YTD. Philly semiconductor index is up 30%.
The explosion of "well, that's it, the AI bubble's just popped" analyses out there are, for now, quantitatively disconnected from any definition of the word "bubble" with which I am familiar.
Derek Thompson retweeted
I think @benthompson cuts through the conspiracy theory/Twitter noise quite well in his article today
AI is not “just another technology.” AI, chips, the supply chain etc are matters of critical national security, at least in the eyes of the government
It would really raise some questions about the modern relationship between boards and founders—and, by extension, between corp governance vs corp monarchy—if both of the two most famous CEO/founders in AI were fired and reinstated within the same three years.
“.. Huang has discussed with confidantes the possibility of returning, according to multiple people familiar with the matter — even as an independent probe by Nvidia’s board investigates the conduct that led to a spectacular leadership crisis ..”
@FT
ft.com/content/45383d86-33c4…
This is EXCELLENT. Two points of emphasis:
1. Focus is a virtue. We’ve had three cost-of-living elections in a row, and the Trump administration has demonstrated what happens when you win a COL election and then immediately pivot to doing a bunch of stuff—ICE roundups, tariffs, Iran War—that either fail to address prices or explicitly make them higher.
2. Speed is a virtue. In the last two admins, Democrats have passed laws that don’t go into effect for several years or authorized spending they couldn’t actually move for several years. In a swingy electorate, that’s a recipe for losing power quickly.
Derek Thompson retweeted
I've made it into the Ringerverse talking chips and China (and kanye, spygate, ethical hoops, and Indecent Proposal) with @DKThomp
millenial nirvana achieved
youtube.com/watch?v=x3rnhS7b…
Derek Thompson retweeted
For those like me trying to wrap their minds around the enormous national security implications of the NVIDIA / Huang / Trump / Lutnick scandal, the latest episode of @DKThomp's Plain English podcast with @jordanschneider and @BuchananBen is a must-listen.
youtu.be/x3rnhS7bmc8?is=XiBc…
really bad. all major Chinese models are being trained on Blackwell class chips
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.
The big question right now has to be: What did the White House know, and what did it know it?
When Jensen Huang joined the president on his trip to China in May and attended the White House state dinner with Xi Jinping in September, senior White House officials already suspected that Huang’s lover was a Chinese spy and Commerce had already been contacted by Nvidia lawyers over their investigation into her company's possibly illicit relationship with China.
There's negligence, there's stupidity, there's corruption, and then there's ... trotting your honeypotted AI mascot around Chinese leaders who know they're illegally siphoning his chips via a Singaporean intermediary overseen by their own spy asset.
Like, what was HAPPENING, here?!
derekthompson.org/p/this-is-…
New newsletter: THE JENSEN HUANG FAQ: WINNERS, LOSERS, WORST CASE SCENARIOS
My basic summary for people who aren't deep AI nerds: It’s sketchy for chief executives to have affairs with their major clients; sketchier still for them to have affairs with clients whose business is suspected of violating federal law; even sketchier for that paramour to also be a suspected secret agent of a foreign government; maximally sketchy if that foreign government is America’s top geopolitical adversary; and beyond all reckoning of sketchiness if the US government knew of these allegations and revelations and did approximately f*ck-all about it for months.
The biggest losers:
- Nvidia, obviously: Huang was the founder, rockstar mascot, and chief spokesperson, with 60 direct reports and no succession plan. The plane is entering turbulent air and the pilot just got thrown out the back window.
- "Just sell to China, it's cool" loses its top spokesperson
- Open-weight models lose a top advocate
- Neoclouds and other less-capitalized AI players lose their bank, as Nvidia's circular financing likely pulls back with collapsing stock price
The biggest winners
- China hawks—ie, those who want to restrict chip sales to China—since their policy case gets a heavy serving of righteousness
- Alternative chipmakers, including AMD, Google, Amazon, and smaller chipmakers, who might find more buyers in a world where Nvidia innovation slows down and the company loses its best salesperson.
- Hyperscalers: If neoclouds struggle without the bank of Nvidia, established clouds like Microsoft and Google might gain at the expense of those upstarts
- uhh, Anthropic? Dario was always on the other side of the China debate with Trump/Jensen and the company is among the least dependent on Nvidia tech
The worst-case scenario: I mean, look ... If one of Nvidia’s largest overseas clients was an illegal backdoor to China, overseen by a spy with whom Jensen Huang was having an affair, and the US govt had information on all of the above and did nothing to arrest China's espionage/legal violations because they wanted to preserve an intimate state-business relationship? Many "if"s there, but sweet JESUS, that looks bad.
derekthompson.org/p/this-is-…
If the NYT/WSJ timelines are correct …
When Jensen Huang joined the president on Air Force One to go to China in 2025 and attended the White House state dinner with Chinese leader Xi Jinping in September, Trump had already been told that Huang was sleeping with a suspected Chinese spy, and his administration was already investigating an illegal relationship between the spy’s company and the Chinese Communist Party
If you're a China hawk, and you read the following revelations/allegations in a 48 hours period ...
1. Jensen Huang is sleeping with a Chinese spy.
2. Nvidia has been selling chips to Chinese AI companies for years through a Singapore intermediary overseen by that Chinese spy.
3. Nvidia’s lawyers pushed back on a federal investigation into this illegal scheme.
4. The White House knew all of the above and did absolutely nothing.
... I think it is understandable if what happens next is your head explodes
derekthompson.org/p/this-is-…
If you’re philosophically opposed to AI chip export restrictions, I think you should be the most mad about this timeline.
Like, I personally did not need the last 72 hours of news to make me doubt the wisdom of selling advanced chips thru to China.
But anybody on the fence or in a politically vulnerable position is going to look at this and go: oh shit, these anti-restriction ppl are corrupt and insane!
One of many problems with getting into bed with Trump is that he’s incredible at driving public opinion against everything he does, because everything he does is so sloppy and corrupt and terrible.
If the NYT/WSJ timelines are correct …
When Jensen Huang joined the president on Air Force One to go to China in 2025 and attended the White House state dinner with Chinese leader Xi Jinping in September, Trump had already been told that Huang was sleeping with a suspected Chinese spy, and his administration was already investigating an illegal relationship between the spy’s company and the Chinese Communist Party
What level of America First is knowing about a Chinese espionage honeypot scheme to steal your most valuable product and covering it up?
The year is 2029. Ten percent of the attendees at all AI business events are ludicrously attractive actors paid by the firm to seduce employees and executives to pretend to solicit corporate secrets. Also half of all corporate emails are phishing tests.
Imagine a historian in 2035 writing a political history of the 2020s, and he points out that every election, the out-of-power party won a cost-of-living election and then used their mandate to ignore or exacerbate affordability only to be destroyed in the next election, where voters again said “to be clear, I don’t want whatever bullshit you’re up to, I just want life to feel more affordable,” and then, once again, the winning politicians prioritized everything except affordability, and this story repeated itself over and over again.
One of my strongest held political opinions right now is: Democrats should do everything they can to avoid that history being written.
“We Were Always Blue” is the first AI art that was also a documentary.
Politico: The Tech Right Pivots Left
Nowhere is that hard pivot more on display than with Alex Karp, the CEO and co-founder of secretive tech giant Palantir, which bet massively on Trump 2.0 and became an administration darling in the process
Inside Palantir’s Manhattan lair, in an exclusive sitdown with Playbook, Karp — who once criticized the Democratic Party as “totally irresponsible” — deployed the word ‘progressive’ eight times to describe his philosophy while acknowledging Democrats ‘absolutely’ corrected in 2026 enough to win back his support in 2028.
Karp: “I’ve been a Democrat my whole life. I grew up in a far left environment. I know more about being a progressive than almost anybody on the planet, and I absolutely believe Palantir— all of our products are essentially progressive.”
Link to article: politico.com/newsletters/pla…
Derek Thompson retweeted
My statement on senate passage of the motion to proceed on the Bipartisan American Affordability and Jobs Act of 2026 (S. 5653).
Derek Thompson retweeted
Always a treat to be a guest on one of your favorite podcasts. Thanks @DKThomp for letting me talk youth vote trends and the future of the GOP on the show. open.spotify.com/episode/4Ze…






















