What Jensen Huang's Downfall Means For Public Affairs
A Brief Programming Note
Until now, this newsletter has been signed off only as "Neptune Ops," the company publishing Artificially Intelligent. We published the first edition in April of 2024, in what now feels like the Stone Age of AI. Our thinking at the time was that the pace of AI was accelerating so rapidly that Neptune Ops' clients and friends might benefit from a regular forum to pull up and make sense of how AI is changing the work of public affairs teams. In retrospect, the idea that AI was merely "accelerating" looks adorable.
In the ensuing years, our readership has grown to include thousands of public affairs professionals across the globe. We have received generous feedback from so many of you who have told us that you make space for Artificially Intelligent in your inbox each week for this never-ending conversation. Our focus here has always been, and will continue to be, one specific question: "What do the latest AI developments mean for the actual work of public affairs teams?" Our role is not to cover AI news in the abstract or AI political developments generally. There are plenty of great places for such news and we encourage you to read them. Our specific corner of the world is synthesizing those developments, and asking what they mean for the real-world work for each of us: how we can capitalize on those changes, how we can avoid the risks, and ultimately how we can evolve our profession into wherever the AI Age takes us.
In her 1976 essay “Why I Write,” the author Joan Didion wrote:
“I write entirely to find out what I’m thinking, what I’m looking at, what I see and what it means.”
This is entirely the spirit and the privilege I have in writing this newsletter. Each weekend I get to pull up and ask those very questions, and hope that you come on the journey with me. And because many of you have asked us who is the author of this conversation, going forward this newsletter will be signed by me, your host Bryan Miller, Founder and CEO of Neptune Ops.
To be clear, Artificially Intelligent has always been a team project of several Neptune Ops teammates plus some helpful AI friends. But I'm making this change to answer your questions about the authorship, to let you know that the opinions expressed here are mine, and that the mistakes are as well.
And there is no better moment to make this change clear than today, as I take on the biggest AI story of the year, and maybe the most important one yet.
— Bryan Miller, Founder & CEO, Neptune Ops
What Jensen Huang's Downfall Means For Public Affairs

We normally keep this to a three-minute read. Not this week. The man who built the most valuable company on earth and served as the AI industry's chief ambassador to the White House resigned in disgrace last Thursday. This is one of the biggest and most complex AI stories ever, and the headlines about affairs and spies are burying the part that matters most to anyone who works in Washington or a state capitol. In today's longer than usual read, we cover what happened, what the story is really about, and why the AI chessboard you thought you understood a few weeks ago just got flipped over.
What Happened, In 30 Seconds
🔱 The relationship. In late October, the New York Times reported a close relationship between Nvidia CEO Jensen Huang and Alice Huang (no relation), an executive at Singapore data center company Megaspeed. U.S. officials reportedly suspect she worked for China's Ministry of State Security. Beijing calls that "groundless speculation," and no charges have been filed.
🔱 The company. Megaspeed reportedly bought more than $4.6 billion in Nvidia technology and is under federal investigation for allegedly helping Chinese firms get around U.S. chip restrictions. The company denies wrongdoing.
🔱 The collapse. On November 8, Huang admitted to an affair in a letter to employees. Reports of additional affairs followed, including one with a senior executive who reported directly to him. On November 12, he resigned. CFO Colette Kress is interim CEO.
Forget The Affair (Mostly). Follow The Chips.
There are a dozen angles here, and the facts change by the hour. We will keep having this conversation in the weeks ahead. But for now, the sex and the spying are the sideshow (albeit quite a show). The main event is a question Washington has fought over for three years: how does China get the chips it needs to compete with the United States on AI? This scandal answers that question in uncomfortable detail, and it hands the answer to Congress at the exact moment Congress is writing the rules.
Why Chips Are The Real Story
Think of AI as a three-layer cake. On top are the apps you use. In the middle are the models, like ChatGPT, Claude, Gemini, or China's DeepSeek and Qwen. On the bottom are the chips that train and run those models. Every layer rests on the one below it. Whoever controls the bottom layer controls how tall everyone else can build.
And the bottom layer isn't just about who gets good chatbots. The most advanced chips also power intelligence analysis, drone targeting, and modern weapons systems. That is why the national security community treats them less like consumer electronics and more like a strategic material.
Here is the fact that gets lost in most coverage: no Chinese frontier model has been shown to be trained start to finish on Chinese chips. DeepSeek's breakthrough model was trained on Nvidia hardware. Its newest model was partly fine-tuned on Huawei chips, but experts reportedly believe Nvidia still did the heavy lifting. China's top tech companies have even moved training to data centers abroad to reach Nvidia chips. Without American silicon, Chinese models would not be competitive with American ones. Period.
Two Camps, One Cheerleader
The fight over chip sales to China has two sides. Here are the arguments they make:
🔱 The Jensen camp. Sell China American chips, and China stays hooked on American technology instead of building its own. Cut China off, and you hand the market to Huawei and speed up the very competitor you fear. Huang argued for years that export controls had failed and that China was barely behind the U.S. anyway.
🔱 The mainstream camp. You do not hand your greatest geopolitical adversary the most important strategic input of the century. To AI's biggest boosters, selling top chips to China is the equivalent of handing the Russians enriched uranium during the Cold War. This view spans China hawks in both parties, most of the national security establishment, and, notably, much of the rest of the AI industry.
Huang's position was never neutral analysis. Nvidia controls an estimated 80 to 90 percent of the market for chips that run frontier AI, and China was a market Huang himself pegged at tens of billions of dollars a year. As Huang well knew, the idea that China would not do everything in its power to build its own chips regardless of American policy is at best naive, and at worst bad faith.
Yet no one made that argument louder or more effectively than Huang. He lobbied Trump directly, repeatedly, and successfully. After Huang's Oval Office pitch in July 2025, the administration reversed its ban on Nvidia's China-specific H20 chip, with Nvidia agreeing to pay the U.S. government 15 percent of those sales. In December 2025, Trump approved sales of the more powerful H200, with the government reportedly taking 25 percent. No other executive came close to his influence on this issue.
How The Most Advanced Chips Actually Get To China
Recent reporting lays out two main routes, and neither requires a trench coat.
🔱 The back door. Chips are sold legally to a company in Singapore, Malaysia, or Thailand, then quietly re-exported to China. Epoch AI estimates that smuggled chips equal to roughly 1.6 million of Nvidia's H100s reached China through the end of 2025.
🔱 The rental car. Chinese firms don't need to own the chips at all. They rent computing power from data centers in Southeast Asia and use the chips remotely. Megaspeed reportedly rented to Alibaba, and under current U.S. rules, that is largely legal.
Now look at the numbers. In the quarter that ended July 2025, customers billed in Singapore accounted for $10.2 billion, or 22 percent, of Nvidia's total revenue. Nvidia said billing location reflected where customers process invoices, not where chips ship. The next quarter, it changed how it reports geography, and the Singapore line disappeared. Separately, a Chinese company linked to Alice Huang's family reportedly became an Nvidia preferred cloud partner and facilitated about $7.5 billion in chip purchases. Nvidia reportedly kept selling to Megaspeed even after the investigation was first reported. And Nvidia's top lawyer reportedly intervened in the 2025 federal probe of the company.
Nvidia denies any diversion of its products, and Huang has not been charged with anything. But ask yourself how plausible it is that the CEO of a company this obsessed with knowing its customers had no idea where a fifth of a quarter's revenue was headed, while personally close to an executive at one of the companies under investigation. Congress is going to ask that exact question.
Washington Looked The Other Way
The Biden administration built an aggressive export control system and, in January 2025, issued a rule that would have required licenses for advanced chip sales across much of Southeast Asia. The Trump administration scrapped that rule in May 2025. Its replacement rules were never issued or were withdrawn. Analysts say enforcement became a lower priority across the board.
The warnings were there. Trump was reportedly briefed on intelligence about Alice Huang in late 2025 and did not act. Treasury, State, the Pentagon, and the U.S. Trade Representative all reportedly backed blacklisting Megaspeed. Commerce never did. When Treasury Secretary Scott Bessent pressed Commerce Secretary Howard Lutnick on why in a Situation Room meeting, Lutnick reportedly answered: "I don't know. I'll look into it."
Now the circular firing squad has formed. Bessent is publicly blaming Commerce. Trump advisers are reportedly discussing pushing Lutnick out, while the White House says the president has "complete confidence" in him. Expect this blame game to be a central plot line of the hearings.
The Trump-Huang Problem
All of this is harder to untangle because of how close the president and Huang became. Huang attended a $1 million-per-plate MAGA Inc. dinner at Mar-a-Lago, where he reportedly made an appeal against chip restrictions. Nvidia reportedly gave $25 million to the White House ballroom project. Trump brought Huang to the Beijing summit in May. And on November 8, the day Huang admitted the affair, Trump called him "a fine person" who should stay on as CEO.
That leaves the White House in an awkward spot. Every policy decision that favored Nvidia now looks like a favor to a friend, and every move to tighten controls looks like an admission that the friend got it wrong.
A Rare Bipartisan Consensus
A bipartisan consensus is forming to restrict the entire lineup of Nvidia's AI chips to China, and to the countries where chips get smuggled or rented out, not just the most powerful chips. To understand why that matters, it's critical to understand the menu:
🔱 Blackwell. Nvidia's current flagship. Still banned for China. Roughly two and a half times as powerful as the H200.
🔱 H200. The previous-generation flagship. Approved for China last December.
🔱 H20. A deliberately weakened chip designed to slip under earlier U.S. limits. Approved for China in 2025.
Until now, the debate was about where to draw the line on that menu. The new consensus says draw it off the menu entirely. Speaker Mike Johnson says there is a "developing consensus in Congress." Incoming Speaker Hakeem Jeffries wants restrictions in the defense bill. Senators Elizabeth Warren and Jim Banks, who agree on almost nothing, are jointly pushing the AI Overwatch Act to give Congress power to review and block chip sales. The Chip Security Act, which would require tracking where chips end up, has a Democratic and Republican co-sponsor and cleared committee unanimously. A bill to close the remote access loophole already passed the House 369 to 22.
Versions of these measures are already in the Senate's draft of the annual defense bill, the National Defense Authorization Act, which is usually signed in December. The House version doesn't include them yet, and a similar provision got stripped last year. If they miss this train, a Democratic Congress arrives in January with chip restrictions near the top of its list.
Could Trump veto it? Sure. He vetoed a defense bill in 2020 and Congress overrode him. Doing it again to protect chip sales to China, in the middle of a spy scandal, with Republican China hawks like Brian Mast, John Moolenaar, and Tom Cotton leading the charge, would be politically perilous.
Most importantly, the biggest cheerleader for selling chips to China is off the chessboard. Nvidia has zero credibility making that argument now. Its trade group still opposes the new measures, while Amazon and Microsoft are backing tighter limits. The lobby that held the line for two years just lost its quarterback.
What This Means For Public Affairs
Start by assuming the chips get restricted. If they do, Chinese open-weight models, the free downloadable models like DeepSeek and Qwen that many organizations quietly use because they cost next to nothing, take a serious hit. They will get fewer updates, slower improvements, and will fall behind Western models.
American frontier labs like OpenAI, Anthropic, and Google get a near-term boost from weaker Chinese competition, but they are not standing on solid ground either. As we covered last week, the election results reshuffled the politics around AI generally, and a Democratic Congress will be far less friendly to the AI industry's light-touch wish list. Put the election and the Jensen story side by side, and the whole map of AI regulation gets scrambled. Washington is likely to act on chips and potentially do so this year. States (including several with new Democratic-controlled legislatures) are likely to act on model regulation, and the new Congress may join them. Everything we thought we knew about the future of AI policy a few weeks ago is up for grabs. For practitioners, that volatility is the lesson. The tools you rely on are now political footballs, and any one of them can be fumbled.
Two Ways To Operationalize This Email
🔱 Eliminate single points of failure. No organization should run any critical workflow without a backup, and that is particularly true in AI. Build workflows that can swap between vendors and models. If Claude is doing all of your public affairs work, ask yourself how you would adapt if Anthropic decides that, given the political heat around AI, it wants nothing to do with politics and shuts off your workflows tomorrow, completely or partially. Platforms have walked away from political customers before (Meta blocked new political and social-issue ads in the final week before this month's midterms, as it has before every U.S. election since 2020), and every public affairs practitioner should take this scenario seriously.
🔱 Watch Western open-weight models. Downloadable models built in the U.S. and Europe let you run your own workflows on your own terms, with no vendor able to pull the plug. Keep an eye on this corner of the market. And here is the irony: its biggest backer is Nvidia, which builds its own Nemotron open models, has invested in open-weight developers, and has agreed to buy Hugging Face, the main hub for the open-model community.
Follow that circle. For years, Jensen Huang was the loudest voice for keeping Chinese AI running on Nvidia chips. His downfall may now cut those models off, and leave Nvidia powering the very Western alternatives that replace them. The company that fueled China's AI boom may end up as the very bulwark against it.
To be continued. . .