Source: raw/Everything_You_Need_to_Know_About_the_Jensen_Huang_Spy_Story_Hard_Fork.md (Hard Fork, NYT, https://www.youtube.com/watch?v=jxCTX_AyZVc) · raw/Nvidia_CEO_Admits_Cheating_with_Suspected_Chinese_Spy.md (The AI Daily Brief, https://www.youtube.com/watch?v=TGDVsntuFd0) · raw/China_Won_t_Take_600_000_Nvidia_Chips._Did_It_Target_Jensen_Huang_For_Better_Ones.md (AI News & Strategy Daily, Nate B Jones, https://www.youtube.com/watch?v=OEY-vm7qRHc) · primary documents saved to ai-research/: Nvidia board statement (Nov 8), Jensen Huang’s employee email (Nov 8), independent directors’ statement (Nov 10), House committee chairs’ letter (Nov 4), Reuters H200 stockpile report (Nov 5), Tom’s Hardware relay of Bloomberg on Fleet Intelligence (Nov 6).
Creators / platform: three YouTube podcast episodes, all fetched 2026-11-11. Hard Fork is an interview with Stephen Witt, author of the Nvidia biography The Thinking Machine (the transcript names the host as Max Read). The AI Daily Brief is NLW’s solo news read. Nate B Jones is a solo analysis.
In early November 2026 the New York Times reported that US officials believe Alice Huang (also called Huang Le; no relation to Jensen Huang), an executive at Singapore-based Megaspeed International, a large Nvidia cloud customer, is a Chinese intelligence asset, and that Nvidia CEO Jensen Huang had a relationship with her; on November 8 he admitted an affair in an email to employees. The personal side is out of scope here. What matters for AI users and buyers is the export-control side: how advanced Nvidia chips reach Chinese AI labs (remote cloud rental abroad, alleged diversion, and the legal H200 channel), what Washington and Beijing are now doing about each, and what that means for compute supply. Each claim below is attributed to the show or document it came from; allegations are labelled as allegations.
Update 2026-11-13: Jensen Huang resigned as Nvidia’s president, CEO and director; the board accepted the resignation on November 11 and announced it on November 12, with CFO Colette Kress as interim CEO. The succession, the financing questions and what it means for buyers are in Jensen Huang Resigns as Nvidia CEO. Everything below this line up to the section dated 2026-11-13 was written on November 11, while he was still CEO, and is left as written; the export-control developments since then are in “After the resignation” near the end.
Key Takeaways
- The confirmed facts are few. Nvidia’s board opened an independent investigation on November 8 into “recent reporting that founder and CEO Jensen Huang has been the target of a foreign intelligence effort”, said it has “no reason to believe that either national security or NVIDIA secrets were compromised”, and on November 10 retained Joon H. Kim and Wachtell, Lipton, Rosen & Katz. Three House Republican committee chairs asked Huang on November 4 to a closed-door briefing on November 18, with a reply due November 11. China’s foreign ministry calls the spy allegations “groundless speculation and slander” (AI Daily Brief).
- Megaspeed sits on the export-control fault line. It buys Nvidia hardware and rents compute from Southeast Asian data centers to customers including Alibaba (Nate B Jones). The AI Daily Brief describes it as giving Chinese AI labs remote access to GPUs that cannot be exported to China, which it calls “not technically illegal” but a loophole China hawks have complained about for over a year. US officials also suspect physical smuggling into China; the AI Daily Brief says this “hasn’t been proved”, and Nate B Jones reports Nvidia disputes it and says its checks found “zero evidence of smuggling”.
- The scale was large. Nate B Jones cites a Bloomberg analysis of 2025 trade records showing more than 17,000 GB300 processors imported to Megaspeed in a single month. Tom’s Hardware cites a C4ADS report linking $4.6 billion of hardware purchases to Megaspeed.
- The legal channel is now the H200, and Beijing is the one blocking it. Reuters (Nov 5): about 600,000 H200s sit in Hong Kong awaiting shipment to mainland China because Beijing is withholding approval for large-scale imports, out of concern about dependence on US technology. Several big Chinese internet companies got about 10,000 each. An eight-card H200 server cost more than 6 million yuan (about $900,000) in China by end-October, roughly three times the price a year earlier; Blackwell servers went for more than 20 million yuan.
- US vetting of H200 buyers is under question. The AI Daily Brief relays a NYT report that the Commerce Department approved dozens of Chinese companies to handle H200s, some with ties to sanctioned firms and some apparently empty offices (one, New Matrix, had a Beijing headquarters with no light fixtures).
- Nvidia is tightening its own controls. Per Bloomberg via Tom’s Hardware (Nov 6), Southeast Asian clouds must now enroll in Nvidia’s Fleet Intelligence system, which can report accelerator locations, to hold Nvidia Cloud Partner status. Tom’s Hardware also reports Nvidia cut about half of its pre-approved Asian buyers and told cloud providers to keep at least 51% of compute with customers outside China. Witt (Hard Fork) has heard Nvidia “cracked down on this a lot” this year.
- All three shows expect China policy to harden. Witt predicts Nvidia will “not be able to sell anything to China ever again”. The AI Daily Brief relays an analyst argument that the leak aims to reverse the H200 approvals and put Nvidia under congressional oversight. Nate B Jones frames it as “a compute constraint arriving at precisely the worst possible time”.
What was reported, and by whom
Primary documents (fetched 2026-11-11)
- Nvidia board, Nov 8: independent investigation run by a committee of independent directors with outside counsel; “The Board has confidence in Jensen and his leadership team.”
- Jensen Huang email, Nov 8: apologises for infidelity without naming anyone; on national security he says only that the board “has initiated an independent investigation. I support the investigation and will cooperate fully.”
- Independent directors, Nov 10: retained former acting SDNY US Attorney Joon H. Kim and Wachtell, Lipton, Rosen & Katz.
- House letter, Nov 4: signed by Brian Mast (Foreign Affairs), John Moolenaar (Select Committee on the CCP) and Rick Crawford (Intelligence). It asks for a joint closed-door briefing on “NVIDIA’s security protocols against foreign adversary espionage” and Huang’s interactions with “a publicly reported People’s Republic of China (PRC) intelligence asset: Megaspeed executive Alice Huang”. Room 2172 Rayburn, November 18; reply by November 11.
The AI Daily Brief (NLW)
- The NYT first reported the relationship at the end of October and called it close; it said Alice Huang was suspected of being a Chinese intelligence agent and that Trump took no action after being briefed. A Friday follow-up reported a memo, circulated in the White House “last spring”, identifying her as working for China’s Ministry of State Security. The FT then reported that US intelligence believed the relationship was romantic.
- The NYT reported that Secretary of State Marco Rubio oversaw drafting of the classified memo, which circulated before Trump’s May trip to China.
- In a Bloomberg interview the week before, Jensen Huang called his Megaspeed interactions “completely normal and not unusual” and said “We’ve studied all their deployments of our GPUs and counted every one.”
- Market: S&P 500 futures fell about $300 billion in market cap at Sunday’s open and recovered within about five minutes. Nvidia was down as much as about 8.8% early Monday before buyers recovered roughly half the loss; the episode saw no spillover to tech broadly at recording time.
- Senator Chris Coons, incoming Senate Foreign Relations chair, said “Congress will investigate and get answers.”
- A Taiwanese semiconductor analyst (name garbled in captions) argued the leak’s purpose is to end Huang’s lobbying against a total chip ban, to reverse the H200 approvals and cut the shell-company link to Southeast Asian smuggling networks, and to put Nvidia “on a national security leash” via congressional oversight. NLW notes this rests on the assumption that the leak was intentional.
- Commentator Ryan (surname garbled in captions) framed the issue as a conflict of interest: whether the relationship weighed on decisions to keep exporting to Megaspeed after Commerce began investigating it for diversion.
Nate B Jones
- Timeline he attributes to the NYT and Reuters: Megaspeed investigation underway since 2024; July 2025 Nvidia said it had been assured H20 licenses would be granted; August 2025 Nvidia’s top lawyer contacted Commerce investigators and invoked permission from their boss, Jeff Kessler (caption “Kesler”); investigators declined to meet, and the case went to the Justice Department, where the NYT reports a criminal inquiry is ongoing. Nvidia calls those contacts compliance cooperation; Megaspeed says it offered to cooperate; Commerce says it followed proper procedures.
- October 2025: Trump briefed on the suspected intelligence connection (NYT). December 2025: Trump announced approved Chinese customers could buy the H200 (not the GB300). January 2026: Commerce implemented the policy with security requirements and case-by-case review.
- The NYT reported officials initially believed Jensen Huang did not know of the alleged intelligence role and had not received a defensive briefing.
- Nvidia says Megaspeed’s overseas cloud business is permitted under the export rules.
- His question for investigators: whether the relationship “inappropriately influenced who got access to those chips”, which bears both on how fast Chinese firms deploy AI and on how many American companies could get compute over the past year.
- Nvidia earnings fall on November 17, the day before the requested briefing.
Hard Fork (Stephen Witt)
- The Malaysian data-center channel was “pretty well known” and reported by the NYT more than a year ago; Witt reasons that if 10-15% of Nvidia revenue came from Malaysia, which has no large AI program, those chips were serving China one way or another.
- Witt’s description of the alleged supply chain: Nvidia sells to the US arm of a Chinese company, which sells on to Megaspeed; Alibaba appears to be Megaspeed’s main customer. He says Megaspeed appears to have been formed in 2023, when Nvidia named it a preferred cloud partner, and that it is a real operation with real data centers whose users are “open questions”.
- One method he says “seems to have possibly happened”: installing dummy servers so a site looks operational while the Nvidia equipment leaves for China. He presents this as unconfirmed. He also cites an earlier case where people at Supermicro swapped packaging labels.
- US Commerce has “very limited authority and insight” once hardware reaches Malaysia. Singapore authorities have been busting smuggling rings since 2024.
- His reading of the leak: China hawks in the administration, Rubio in particular, wanted to reduce Huang’s influence over the president.
- Opinion, labelled as his: the main risk to Huang is congressional hearings, not the board; Nvidia could face “a massive fine”; Huang’s credibility on China and export restrictions “is zero”; AMD, Broadcom and Google (now one of Nvidia’s top rivals) gain from Nvidia’s trouble; AI infrastructure stocks already trade at “very cheap valuations”, which may explain the muted market reaction.
- He thinks Chinese intelligence targets Anthropic and OpenAI more than Nvidia, because China “can’t actually manufacture what NVIDIA manufactures”; for Nvidia, he thinks it was “probably more about an influence campaign”.
What this changes for someone using or buying AI
- Nothing changes overnight for Western API users. No source reports a change to model availability, API pricing or Nvidia shipments to US customers.
- Chinese open-weight models are compute-squeezed from both sides. Washington’s H200 vetting is under scrutiny, Beijing is rationing H200 imports, Nvidia is tightening Southeast Asian rentals, and grey-market H200 servers in China cost three times what they did a year ago (Reuters). If you route cost-sensitive work to DeepSeek, Qwen, Kimi or GLM-class models, expect their release cadence and hosted prices to depend on this. ^[inferred] See the Chinese-models decision framework.
- If you rent GPUs from Asian neoclouds, check their partner status. NCP status in Southeast Asia now requires Fleet Intelligence enrolment, and Nvidia reportedly asks cloud providers to keep at least 51% of compute with non-Chinese customers (Tom’s Hardware). A provider that loses NCP status may lose preferred access to new chips.
- Plan 2027 capacity early (Nate B Jones’s view). He argues compute for non-lab companies will be “existential for many of them” over the next 6 to 12 months, especially for cyber defense, that the most capable models will stay gated inside labs for longer, and that “2027 budget planning is going to be very, very tricky”. His advice: plan over longer horizons so you can get the models you need when you need them.
- Dates to watch: Nvidia earnings on November 17; the requested House briefing on November 18; the scope and findings of the board investigation; any Commerce action on H200 licenses.
- Updated 2026-11-13: earnings are still on November 17, now with Kress as interim CEO (Nvidia press release). Huang has agreed to testify behind closed doors before Thanksgiving; no date is given and it is not stated whether November 18 stands (TechCrunch, citing Bloomberg and NBC). Add to the list: the AI Overwatch Act in the defense bill, and any Commerce rule on remote access from Southeast Asian data centers.
Additional reporting relayed by The Artificial Intelligence Show (recorded 2026-11-09)
Roetzer and Kaput read from news reports on air; none of the underlying articles were fetched.
- Earlier reporting: an earlier New York Times piece, read on air, described Megaspeed as “poised to buy $2 billion of Nvidia AI technology over the next year”.
- The investigation: “Nvidia says independent directors and outside lawyers will oversee its investigation”, and the board says it currently has no reason to believe national security or Nvidia secrets were compromised.
- Governance (Semafor, as read on air): Huang’s “near 4% stake gave him a net worth of $155 billion at Friday’s close” and “does not protect him should the board or investors try to” remove him.
- Stock: Nvidia “started down 8%” on Monday November 9 while other large tech stocks rose; it had recovered part of that by 10:15 a.m. Eastern.
- Roetzer’s view that the leak was coordinated from inside the US government is his opinion.
Additional discussion on Moonshots ep. 310 (added 2026-11-11)
Source: Moonshots with Peter H. Diamandis, ep. 310 (raw/Jensen_s_Spy_Probe_Anthropic_s_Debut_Week_and_Elon_s_30M_Robot_Workforce_310.md, https://www.youtube.com/watch?v=ILQwMvCA2_o), hosts Peter Diamandis, Alex Wissner-Gross, Dave Blundin, Salim Ismail and Emad Mostaque. Recorded during the US trading day after the board statement; exact date not stated. The auto-transcript has no speaker labels, so attributions below are inferred from context (who is addressed, who replies). Hosts’ figures are relayed, not checked.
- Market move at recording: a host says Nvidia was “down 6% as of this moment” while the market was up, and AMD was up “10, 11%” the same day. The AI Daily Brief’s “as much as about 8.8%” above is an intraday low; the two were probably read at different times.
- A larger diversion figure. Wissner-Gross relays reporting “in the past 24 hours” that the alleged channel may go back to 2023, with “something like **4.6B and the NYT’s $2B figures above.
- The Pax Silica framing (Wissner-Gross): the US Pax Silica initiative was meant to keep Western compute “within our sphere of influence”; his reading is that one person at the top of one company was enough to pierce it. Opinion.
- How export rules apply to lab staff (Wissner-Gross, “my understanding”): a US person (citizen or permanent resident) physically in the US does not trigger a deemed export; a non-US person, or a US person overseas, does. He expects any restriction on frontier-lab staff to turn on nationality, not ethnicity. Not legal advice and not checked against EAR text.
- Expected follow-on (a host, probably Mostaque; Ismail makes the same prediction later, opinion): a push against Chinese open-weight models in government and regulated industries (banking, insurance, cars, drones), which he links to an analysis captioned “Coher’s” (probably Cohere) showing some Chinese open models automatically generating backdoors. The analysis was not located. See the Chinese-models decision framework.
- Robot telemetry anecdote (Blundin, by context): he says MIT investigated Chinese-made robots internally and found them “transmitting video and audio back to China”, and now keeps them in a separate room. Unpublished, by his account; unverified.
- Why now (several hosts): the Malaysia GPU flow was long known (one host: Malaysia is “the biggest consumer per capita of Nvidia GPUs”) and read the timing as political, tied to the new Congress after the midterms. Opinion.
After the resignation: export-control developments (added 2026-11-13)
Sources: Hard Fork’s November 12 panel with NYT reporters Anna Swanson and Ryan Mac and Stephen Witt (raw/Inside_the_Downfall_of_the_Most_Powerful_Man_in_A.I._Hard_Fork.md, https://www.youtube.com/watch?v=W79ML7zFyZg; no speaker labels, attributions from context); The AI Daily Brief of the same day (raw/Jensen_Huang_Is_Out._Everything_in_AI_Is_Now_Up_for_Grabs..md, https://www.youtube.com/watch?v=Dt7Q_VQxGHc); AP, TechCrunch and the Senate Banking Committee statement saved to ai-research/ on 2026-11-13. The resignation itself is covered in the CEO-transition article.
- How the prediction above turned out. Witt’s view in the earlier Hard Fork episode was that the main risk to Huang was congressional hearings, not the board. The board accepted his resignation on November 11 (Form 8-K). Witt says on the later episode that he had expected Huang to survive the first story.
- Huang will testify. After subpoena threats he agreed to appear before the three House committees behind closed doors before Thanksgiving; Chairman Mast, who announced it on November 12, expects the session to be recorded and transcribed (TechCrunch, whose own pieces differ on whether he agreed on November 11 or 12).
- A second, Democratic inquiry targets the administration. Rep. Richard Neal and Sen. Ron Wyden wrote to Commerce Secretary Howard Lutnick on November 12 asking the department to preserve records on the H200 arrangement and on why Megaspeed kept receiving chips after the intelligence concerns were reportedly known. As the minority they cannot compel answers (TechCrunch; the letter was not fetched).
- Legislation. Senators Warren and Banks urged on November 9 that Congress pass “key bills like the AI Overwatch Act in the NDAA” and that Commerce “close export control loopholes” (Senate Banking Committee statement). Swanson expects more hearings and chip bills.
- Commerce is where the delay is being pinned. Swanson says Lutnick is seen inside the administration as the one pushing back on an entity-list designation for Megaspeed and on wider rules to stop Chinese companies “remotely dialing into data centers in Southeast Asia”. NLW relays an Axios report that Trump approved action against suspected diversion three times this year (May, August and November 5) and that Commerce had not carried it out; the Axios body was paywalled, headline and date confirmed.
- The remote-access rule is under discussion, not written. Swanson: officials have “struggled with that question for over a year” and she is unsure how quickly a rule would come, “if at all”. NLW relays reports that Chinese firms are rushing to rent Southeast Asian capacity before the loophole closes; AP quotes an Asia Society analyst expecting the same.
- Investigations of Megaspeed. Swanson says the Justice Department is running a criminal investigation of Megaspeed now. NLW says Commerce opened its investigation in late 2024 on an allegation of physical diversion, not only remote access. AP adds that Singapore questioned Alice Huang and took her passport last year without finding enough evidence of a crime, that Malaysia is checking whether chips brought in by Megaspeed or its affiliate Speedmatrix were re-exported, and that Megaspeed says it takes US export-control compliance extremely seriously.
- What the first Commerce inspection found (TechCrunch timeline, citing the NYT): an official visiting Megaspeed’s Malaysian site in 2024 found Nvidia equipment still in boxes, which raised the concern it was being shipped onward.
- A wider corporate network (The Information, via TechCrunch): Alice Huang and her family are reported to have built companies in China and Singapore to acquire Nvidia chips, including Shanghai Century Litong, founded in 2023 and an Nvidia preferred cloud partner, which facilitated 7.5 billion as Megaspeed’s own purchases since its founding in 2023.
- Nvidia’s lawyer and the Commerce probe. TechCrunch, citing Bloomberg, names general counsel Tim Teter and says he contacted Commerce Under Secretary Jeffrey Kessler in 2025. NLW’s reading is that the contact made the Commerce investigation “go away”; Nate B Jones’s earlier account above is that investigators declined to meet and the case moved to Justice.
- Revenue disclosure (Swanson): late last year Nvidia changed how it reports revenue so that the share from Singapore and Malaysia became much harder to work out. Not checked against the filings.
- Is the offshore cloud business legal? Swanson describes renting Southeast Asian capacity to Chinese clients as “what is now a legal business”, which supports the reading in Open Questions below that the channel is permitted and the dispute is over physical diversion and over whether to change the rule.
- Size of the China business at stake. TechCrunch puts Chinese customers at about $7.9 billion of Nvidia’s second-quarter revenue, roughly 8%.
Export-control discussion on All-In and Moonshots ep. 311 (added 2026-11-13)
Sources: All-In Podcast, recorded Thursday November 12 (raw/Jensen_OUT_as_Nvidia_CEO_-_Spy_Scandal_Export_Restrictions_Market_Impact.md, https://www.youtube.com/watch?v=Uv2w4VTU-iQ), hosts Jason Calacanis, David Sacks, Chamath Palihapitiya and David Friedberg; Moonshots with Peter H. Diamandis ep. 311, a live episode recorded November 13 (raw/Jensen_s_Departure_NVIDIA_s_succession_plan_AI_doomer_response_and_Demis_talks_RSI_311_Live.md, https://www.youtube.com/watch?v=GZrDapFXk0I), with Diamandis, Alex Wissner-Gross, Emad Mostaque and Salim Ismail. Neither transcript has speaker labels; attributions are from context (who is addressed, who replies). Sacks says on air that he first met Huang at the White House while serving in the administration, so his account of chip policy is that of an administration participant. Most of both segments is opinion about who leaked what; only the policy-relevant items are kept here.
- The bill behind the remote-access debate is the Remote Access Security Act (RASA). Wissner-Gross (Moonshots) names it as the legislation that would make it harder by statute for China to reach GPUs through cloud services, including clouds in third countries, and argues the midterm result makes it easier to pass; Ismail notes it already has bipartisan support. Checked on 2026-11-13 via search results: H.R. 2683 passed the House on January 12, 2026 by 369-22 and has not passed the Senate (congress.gov; CNBC, August 19, 2026). A CNBC-quoted analyst adds that the bill would give Commerce the authority but Commerce “would still need to create a rule”. This is the legislative route to the remote-access rule Swanson describes above as “not written”.
- A second bill: the Chip Security Act. Calacanis describes it as bipartisan and focused on tracking chips after sale, with chips needing re-initialisation if resold; the open question he raises is whether enforcement would sit with Nvidia or with a government body. Bill text and status were not checked.
- What exporters must provide today, per Palihapitiya. He says current export approvals already require the exporter (he names Amazon, Microsoft and Google) to keep telemetry on how chips are used and a remote kill switch, as a private obligation. His concern is a shift to a kill switch held by the government. Not checked against licence conditions; compare Nvidia’s own Fleet Intelligence requirement for Southeast Asian cloud partners above.
- Signal from the House China committee. Palihapitiya reads a post from Rep. John Moolenaar (captioned “Molinar”; he chairs the Select Committee on the CCP, per the House letter above) saying Nvidia must change its practices and appoint leadership that puts national security first. He reads it as the terms for ending the dispute. The post was not fetched.
- Sacks’s history of the rules (his account). The Biden administration’s 2022 controls banned leading-edge chips for China while letting “lagging” chips such as the H20 be sold in unlimited quantities without a licence; the Trump administration kept the framework but moved lagging chips to limited quantities under licence. The stated reason for allowing lagging chips was to take market share and R&D money from Huawei. He now says that if Congress wants no chip sales to China at all, “fine”, because “the juice isn’t worth the squeeze”.
- Beijing is keeping the chips out (Sacks). He says China banned the H20 and has kept approved H200s from entering, consistent with Reuters’ 600,000-unit Hong Kong stockpile above, and relays “recent press” that the latest DeepSeek model is trained only on Huawei chips. The DeepSeek claim was not checked; the wiki’s DeepSeek V4.1 Flash article records serving on both Nvidia and Huawei Ascend, not training.
- How Huawei closes the gap (Sacks). Huawei’s CloudMatrix system networks 384 Ascend chips where Nvidia’s NVL72 networks 72; the Nvidia rack was still better, but Huawei uses more chips and more energy to get closer at the rack level. His forecast, opinion: if Nvidia loses “a tempo or two” in its yearly chip cadence, Huawei gains most. Palihapitiya disagrees that anyone gains in the short term, because rivals still need TSMC and HBM capacity. Epoch’s measurement of the same trade-off is in Epoch’s Horinger analysis (US data centers get about 2.5 times more compute per unit of power).
- Doubt about the spy finding. Sacks relays a Politico report that the incoming Democratic chair of the House Intelligence Committee, after a classified briefing, said he was not sure Alice Huang is a spy and that she “might be a useful idiot”. Mostaque says Trump and others have said there is no evidence. Neither claim was checked.
- Earlier public warnings. Calacanis says the NYT first mentioned the relationship in October 2025 and that the short seller Culper Research named Megaspeed about six months ago in a thesis that Nvidia understated its reliance on China. Wissner-Gross also cites activist short reports and Nvidia’s change to Singapore revenue reporting (the same point Swanson makes above).
- Who told the press. Wissner-Gross relays Bloomberg reporting that US officials informed the Financial Times. Not fetched.
- Date to watch. Diamandis says Trump is due in Shenzhen to meet Xi Jinping about a week after the resignation. Not confirmed from a primary source.
Washington after the resignation: The AI Daily Brief, November 13 (added 2026-11-14)
Source: The AI Daily Brief, “Washington Fired Jensen Huang”, NLW’s solo news read recorded Friday November 13 (raw/Washington_Fired_Jensen_Huang.md, https://www.youtube.com/watch?v=1kf1QWG_YKQ). NLW’s thesis is in the title: the resignation reads as the opening of a period in which Democrats assert power over large private companies and China hawks gain ground. Everything below is relayed from named outlets or posts; none of it was fetched. Caption fixes: “Muinar” to Moolenaar, “John Thun” to John Thune, “Demitri Al Perovich” to Dmitri Alperovitch, “Alice Hang/Hong/Wang” to Alice Huang, “Lutnik/Letic” to Lutnick, “David Saxs” to David Sacks.
- Congress says the bills now have the votes. NLW reads a statement from House China committee chair John Moolenaar (the post Palihapitiya read above) that Huang’s “firing” could have been avoided and that Congress must “act quickly to pass bipartisan export control legislation”. Speaker Mike Johnson speaks of “a developing consensus”; Senate Majority Leader John Thune calls passing chip controls “a high probability”; Brian Mast, sponsor of one of three chip-control bills, says the week “100% boosted the chances of all three being passed” and that sponsors are “getting direct assistance from the White House”. The controls were already expected to ride on the must-pass defense bill (NDAA) in the lame-duck session; a congressional aide told Politico Democrats will push for tougher limits.
- OpenAI now backs all three bills, per Politico: it had been non-committal and has privately told congressional offices it supports them. Anthropic was already a supporter. Senator Jim Banks, sponsor of one bill, calls it “a generational opportunity”. (The Neuron’s version of OpenAI’s position is in the CEO-transition article.)
- The two positions, as NLW frames them. Gavin Baker (captioned “A trades”, probably Atreides Management ^[inferred]) argues that selling China older GPUs kept its labs permanently behind under scaling laws, and that a full ban will push China onto its own, less power-efficient hardware path and let it catch up. Dmitri Alperovitch replies that the claim is now testable: years of the leaky approach produced Chinese models under nine months behind, so try giving them no compute, though smuggled Nvidia chips already in China spoil the test. NLW’s view: Alperovitch’s version is the more politically resonant because it needs no knowledge of scaling laws, and Baker’s side never made its case in those terms in public.
- Testimony and hearings will run into 2027. Bloomberg (relayed): Huang agreed to testify under oath at a closed-door House hearing before Thanksgiving, which Mast says will be recorded and transcribed. The House China committee was still threatening a subpoena with an 8 p.m. deadline. In the Senate, Banking chair Tim Scott asked Huang to appear on export controls and invited Lutnick to testify alongside him. Warren and Chris Coons asked the administration to preserve documents on Nvidia exports and Alice Huang, citing their roles as incoming committee chairs when Democrats take control in January.
- Lutnick is the official most exposed. Axios and others (relayed) report Commerce slow-walked export-control enforcement all year, which rival officials call defiance of presidential orders. Politico’s sources split on whether he is fired over it.
- The spy finding is in doubt; the smuggling question is not going away. CNN (relayed) reports multiple sources say the intelligence does not prove Alice Huang spied for Beijing, and that officials at ODNI and the CIA are not convinced; Attorney General Todd Blanche (as captioned) said no conclusions have been reached. David Sacks called the spy narrative “collapsing”. NLW expects the hearings to focus instead on what Huang knew about Megaspeed giving Chinese companies access to Nvidia chips.
- New smuggling reports (all relayed). Bloomberg says it saw corporate documents of chips moving from Singapore to a data center in China. The Information reports companies connected to Alice Huang invested in DeepSeek and promised it a steady supply of Blackwell chips. The Justice Department is investigating links between Alice Huang and Supermicro, whose co-founder was charged with chip smuggling in March.
- A claimed GPU fingerprint in open weights. Former OpenAI researcher Will DePue (captioned “Will Depu”; ^[inferred] spelling) posted that a model’s initial weight layout depends on the exact GPU and seed, that traces survive training, and that by searching more than six trillion chip-and-seed combinations he matched six Chinese open-weight models, including DeepSeek V4 Pro, Kimi K3 and MiniMax M3, to Nvidia Blackwell systems, four of them to GB200/GB300 racks. He says it was independently verified by senior researchers at OpenAI and Anthropic. Not checked; no paper or code was fetched. It conflicts with the “recent press” Sacks relayed above that DeepSeek’s latest model was trained only on Huawei chips. ^[ambiguous] If it holds, it would undercut the view that Chinese labs reached near-frontier models on a fraction of the compute ^[inferred], which matters for anyone pricing Chinese open-weight models on a “cheap because efficient” assumption; see the decision framework.
- Why a chip ban could suit a slower US build-out. A Natixis senior economist (relayed) argues slower US data-center construction after the midterms keeps US token prices high and gives China room to export cheap inference on abundant power, which makes denying China chips a coherent companion policy. See the data-center slowdown.
Open Questions
-
When was Trump first briefed? ^[ambiguous] Nate B Jones, citing the NYT, says October 2025. The AI Daily Brief, citing the NYT, describes a memo that circulated in the White House “last spring”, before the May China trip. Hard Fork says Trump was aware “for a year”. These may be two separate events (an initial briefing and a later memo); none of the shows reconciles them, and the NYT articles were not fetched (paywalled).
- 2026-11-13: TechCrunch’s timeline, citing the NYT, lists them as two events: Trump briefed in late 2025, and the Rubio-supervised memo circulating in spring 2026. That fits the two-event reading; the NYT articles are still unfetched.
-
Is renting Nvidia compute in Malaysia to Chinese customers legal? ^[ambiguous] The AI Daily Brief calls remote access “not technically illegal”; Nate B Jones reports Nvidia says Megaspeed’s overseas cloud business is permitted; Witt (Hard Fork) thinks selling to fronts whose main customer is Alibaba is “in violation of the export restriction”, while saying it “depends on how you read the law”. The wiki’s Epoch compute economics article also describes offshore cloud compute used by Chinese customers as “generally-permitted”. No regulation or ruling was located.
-
How much hardware went through the channel? ^[ambiguous] C4ADS links 2 billion” over a year (AIS); Wissner-Gross (Moonshots ep. 310) relays about $8B since 2023 without naming the outlet. The figures may measure different things (purchases linked, planned purchases, alleged diversion).
- 2026-11-13: two more figures. The Information reports **8B”. ^[inferred] TechCrunch separately says Megaspeed reportedly bought $4.8 billion over the last two years.
-
Alice Huang’s title. Hard Fork says “top executive”, the AI Daily Brief “executive”, the House letter “Megaspeed executive”; Reuters (search snippet only) calls her a “former executive”. Not material to the export-control question; noted for accuracy.
- 2026-11-13: still unsettled. Hard Fork’s host now calls her “the CEO of MegaSpeed”; AP calls her a “former Megaspeed executive” and reports Singapore suspected she gave inaccurate information to downplay her role.
-
Unverified figures and claims: the Bloomberg 17,000-GB300 single-month import figure (Nate B Jones), Megaspeed’s 2023 formation and preferred-partner designation (Witt), the dummy-server method (Witt, self-described as unconfirmed), and the NYT reports on New Matrix and the Commerce H200 approvals (AI Daily Brief). Searched with Tavily on 2026-11-11; found only secondary relays, no primary documents.
-
The H200 policy documents. The December 2025 announcement and January 2026 Commerce implementation (Nate B Jones) were not fetched. Reuters confirms only that the administration “gave Nvidia permission last year to sell the chips to approved Chinese customers”.
-
Whether Huang accepted the November 18 briefing. The reply deadline is today (November 11); MLQ.ai (search snippet) reports Chairman Mast said he would “absolutely” consider a subpoena if Huang did not respond. Not confirmed from a primary source.
- 2026-11-13: answered in part. He agreed to testify before Thanksgiving (TechCrunch, citing Bloomberg and NBC); the date is not public.
-
Host attribution. The Hard Fork transcript names the host as Max Read; Hard Fork’s regular hosts are Kevin Roose and Casey Newton. Not verified; the article attributes Hard Fork content to the show and to Witt.
-
The remote-access rule needs both a law and a rule. RASA passed the House in January 2026 and was pending in the Senate as of August (CNBC, via search). Whether it rides on the NDAA alongside the AI Overwatch Act, and how fast Commerce would write a rule under it, is not known.
Related
- Jensen Huang Resigns as Nvidia CEO (November 2026) — the next chapter: succession, financing and what changes for buyers
- The Compute Economics of the AI Buildout (Epoch AI, 2026) — the smuggling estimates and Malaysia trade gap behind this story
- Chinese Open-Weight Models: A Decision Framework
- Kimi K3 (Moonshot AI)
- The Data-Center Slowdown After the November 2026 Midterms
- AI Bubble Signals, Late October 2026
- The 2026 US Midterms and AI
- METR’s Assessment of AI R&D Acceleration at Anthropic