In a move that rewrites the rules of American capitalism, NVIDIA and AMD have agreed this week to pay the US government 15% of the revenues from chip sales in China, not as a tax or as a royalty, but as a price for permission to export recently restricted AI chips to China. This is not a tariff, it's not a licensing fee. It's not even a penalty. It's a cut a direct slice of corporate revenue handed to the federal government in exchange for the right to export.
The deal personally brokered by the president applies to Nvidia's Age 20 and AM DS equivalent AI chips, both of which were specifically designed to comply with earlier export restrictions. In return for export licenses, the companies will pay the US Treasury an estimated $2 billion per year. This is not how American trade policy is supposed to work. Traditionally, export controls are binary and based on national security decisions.
Either it's reasonable to sell a particular product abroad or it isn't. You would think that national security concerns are non negotiable, but this deal introduces 1/3 category safe enough if the price is right. It's a transactional model which sets a precedent which could reshape how the US government interacts with the business
community. Based on estimates from Bernstein Research, NVIDIA would likely sell 1.5 million H 20 chips in China this year, generating about $23 billion in revenue, with AMD contributing another $800 million in sales. A 15% cut of that total yields roughly $2.1 billion to the US Treasury. Now, maybe you think none of this matters, that these chips are just powering chat bots and
translation tools. But if you worry about the military applications of AIA, deal like this starts to look less like pragmatism and more like a strategic blunder. And the 15% premium that Trump negotiated? It's probably less than what Chinese AI firms were already paying to smuggle these chips in through third countries. In that light, this isn't a windfall, it's a discount. Now, some reporting has floated higher sales figures, 5 to 7 1/2
billion dollars annually. But these projections assume perfect conditions, full market access, no domestic Chinese competition, and maximum volumes being sold. That scenario would require NVIDIA and AMD to capture 100% of China's AI chip market. Which is both unlikely and unsustainable in the long term, especially when Chinese firms like Huawei are racing to develop homegrown alternatives. Even if the upper bound were reached, the numbers are still
trivial in the context of federal finances. the US government spends around $18.6 billion per day, meaning that this $2 billion windfall covers about 2 1/2 hours of government spending. It's not so much a windfall as a rounding error in the federal budget and an insanely low price to be paid for national security. If the risks are real, critics argue that Trump may have surrendered a key pillar of America's tech containment
strategy for spare change. Liza Tobin, a former National Security Council China expert, told the Financial Times that Beijing must be gloating to see Washington turn export licenses into revenue streams. What's next, she asked, letting Lockheed Martin sell F30 fives to China for a 15% Commission? Trump, for his part, is not hiding the transactional nature
of the deal. At a press conference, he said that Nvidia's CEO had asked for a release on export restrictions and that Trump asked for 20% and was haggled down to 15%. This isn't just a new deal, it's a new doctrine. As CNN put it, Trump has ushered in an era of shakedown capitalism. Aaron Bartnik, a former White House tech official, told the FT that this is essentially just a kickback to the federal government for allowing NVIDIA and AMD to undermine US National Security.
The legal basis for this deal is shaky at best. the US Constitution's export clause prohibits taxes or duties on exports, and the Export Controls Reform Act bars fees for export licenses. Whether this 15% cut qualifies as either is unclear, but it walks a fine line between deal making and constitutional overreach. This type of deal isn't A1 off either. It's part of a broader pattern of a presidency that governs by deal, not doctrine. In Trump's Washington policy
doesn't appear to be written. It's instead bargained for. And if you're NVIDIA, the cost of doing business in China is a 15% cut to the US Treasury. The chip deal wasn't conceived in a committee room or hashed out by trade lawyers and national security experts. It was entirely forged in the Oval Office, where Nvidia's CEO made his case directly to the president. The result?
A fast tracked approval and a deal that looks less like thoughtful governance and more like a hastily arranged toll collected at the gates of executive discretion. This is by no means an isolated incident. It's a pattern. Apple's Tim Cook gifted Trump a gold mounted plaque and pledged $600 billion in U.S. investment. Days later, Trump announced plans for new tariffs on chip imports, but said that Apple would be spared.
Coincidence. Maybe. But in Trump's economy, flattery will get you a long way, and the right meeting can rewrite policy. The editorial board of the FT describes this as pay to play. Trump, on the other hand, calls it deal making. Either way, the rules are clear. If you want to sell, invest or operate with limited interference, you need a presidential relationship. What we're seeing now is governance by negotiation, not legislation.
The invisible hand of the market has been replaced by the visible handshake of the president. CE OS are no longer just managing companies, They're managing their relationships with Trump and those who can't afford the price of admission. They're left to navigate a regulatory landscape that shifts with the president's mood. It's not good for any country to have a leader constantly inserting himself into private businesses, making decisions for them while having no stake in
their success or failure. Historically, voters worried about left-leaning parties interfering in businesses to their detriment, and now it appears we have to worry about both sides. The implications of this new
reality are profound. Long term planning becomes impossible, compliance becomes bargaining, and the line between public policy and personal favour blurs into something more familiar in emerging markets than in an advanced economy like the United States. Jensen Wong, who was notably absent from Trump's tech broligarchy at the inauguration, still managed to find a way in by leveraging Nvidia's strategic
importance. His tactic worked, but it also set a precedent that lobbying, not law, is the path to policy and business success. This isn't simply a new style of leadership. It's a new operating system for American capitalism, one where executive power is personalized,
discretionary and transactional. And while Trump supporters hail it as bold pragmatism, where the United States has ACEO, critics see something else A presidency that's rewriting the boundaries of executive power and possibly the Constitution. Some parts of the Constitution are open to interpretation, but Article 1, Section 9 is not one of them. It states that no tax or duty shall be laid on articles
exported from any state. And yet the Trump administration has struck a deal that looks suspiciously like a constitutionally prohibited export tax. If it's not a tax, it could then be seen as an export license. But that doesn't really work either. Under the Export Controls Reform Act, the government is also prohibited from charging fees for export licenses. That's not a technicality. It's a direct legal constraint. The administration insists that this isn't a tax of duty or a
licensing fee. It's just a deal, they say. But if it's not a tax or a fee, what is it? A donation? A tribute more troubling than the semantics is the process, or lack thereof. Congress had no role in this arrangement. No hearings were held. There were no votes. There was no oversight. Just a handshake in the Oval Office, a press release and a tiny revenue stream directed to the Treasury. As Gary Hofbauer of the Peterson Institute put it, that's not how historically we've done business
in this country. The legal ambiguity is compounded by enforcement paralysis. Who's going to sue? Not NVIDIA or AMD. They are beneficiaries. NVIDIA had to take a four and a half billion dollar write down on its unsold H20 inventory in May, and now it gets to sell all of those chips and more. Without a clear plaintiff, the courts may never weigh in. And that leaves the deal floating in a constitutional limbo, contested, unchallenged, and potentially precedent
setting. Lawmakers from both parties have sounded the alarm about this. Republican Representative John Moulinar warned that the deal incentivizes the government to trade strategic restraint for short term revenue. Democrat Raja Krishnamurti called it a dangerous signal to allies and adversaries alike that American principles are now negotiable. The precedent that this deal sets is quite chilling. If export controls can be
monetized, what's next? Could a fee be paid so companies could sell weapons to enemy countries? Maybe a high duty on fentanyl would make that OK. The logic is slippery and the slope is steep. Stacy Rykson at Bernstein Research wrote that while the deal makes sense for NVIDIA and AMD, can you now just buy your way out of National security? Apparently the answer is yes for 15% of revenues, or 2 1/2 hours of the government budget.
If the legal footing is shaky, the strategic rationale is even more contested, especially when national security is the supposed justification for export bonds. Export controls were once the firewall of American strategy, a binary gate that separated what could be sold from what couldn't. If a technology posed a national security risk, it simply couldn't leave the country. No exceptions, no negotiations, no revenue sharing agreements.
But this deal with NVIDIA and AMD has introduced 1/3 category safe enough if the price is right. The result is a transactional model that turns national security into a negotiable asset.
The chip at the center of this debate, Nvidia's H20, is not obsolete, despite Trump's claims it's not powerful enough to train large language AI models, but it is one of the best chips in the market for inference, the operational phase of AI, where trained models are deployed to generate answers, make predictions, and do calculations. The the H20 is in fact 20% faster than Nvidia's Band H100 at inference, and importantly, is far more capable than anything available from Chinese firms.
Getting access to these chips at a 15% markup is a huge win for China, especially when Trump says that he'll put a 100% tariff on American chip imports. Getting access to inference chips matters a lot for Chinese AI companies, as inference has recently become the strategic bottleneck for AI companies. Training gets all the headlines, billion dollar data centers,
massive compute runs. But inferences where most of the energy cost and chips are consumed over time and as reasoning models become the norm, inference demands have skyrocketed. These models don't just spew out an answer to a question, they pause, simulate, and select. And that pause that you see is compute. And that compute requires a lot of chips and consumes a lot of
power. China's AI labs have managed to train frontier models despite hardware restrictions, but serving those models to users has proven a lot more difficult. Moon Shot's Kimi K2 model, for example, launched a great fanfare both in China and the West, but within days it was having to apologize online to users because it was so slow at inference, likely driven by a of inference chips like the H20. China does have a hidden advantage on this front, cheap
and abundant energy. Unlike in the West where data center expansion is often constrained by grid capacity and permitting delays, China's state backed infrastructure allows AI firms to plug in and scale up with minimal friction. Inference is energy intensive, and China is structurally better positioned to absorb more of that cost than almost any other
country in the world. Trump's reversal on H20 exports may unlock China's inference bottleneck, but it also unlocks something else strategic ambiguity. The H20 chip, while commercially valuable, may also carry telemetry features, data on temperature, power usage, and utilization. This could theoretically be repurposed for geolocation tracking. The proposed Chip Security Act would mandate this type of tracking to prevent smuggling and unauthorized use.
In an era of hyperconnectivity, the line between tracking and surveillance is thin. It's not just Washington that's uneasy about this deal. Beijing has launched its own investigation into American design chips, demanding the tech giants like Alibaba and Byte Dance justify any purchases and explain why they aren't using Chinese made alternatives. The concern that Nvidia's chips might contain back doors, latent vulnerabilities, or remote access features that could
compromise Chinese security. NVIDIA denies that anything like this exists, but the suspicion reflects a broader truth. In the modern world, even commercial hardware has spine capabilities. The Chinese governments concern isn't that far fetched. The US has long worried about similar risks from Chinese technology, especially in 5G networks, and it wouldn't be unprecedented for Washington to embed surveillance tools in foreign targeted hardware.
In 2021, the FBI ran a global sting operation using a norm A supposedly secure encrypted phone sold to criminal networks around the world. What users didn't know was that the supposedly secure devices were transmitting their messages directly to U.S. law enforcement. The platform was a Trojan horse, marketed as privacy enhanced but
engineered for surveillance. So when Beijing worries about using a chip that was designed specifically for the Chinese market, it's not paranoia, it's a reasonable fear. Trump's defenders argue that it makes sense to follow and engage to constrained strategy with China and AI chips if the US sells downgraded chips that are better than the locally produced ones. This, they argue, will keep China dependent on US technology.
But history suggests that this type of strategy energy doesn't work very well, as China has a history of refurbishing and optimizing even limited hardware for military grade applications. And the H20 is not that limited. Trump's deal sends a troubling signal to allies who now see the US putting a price tag on national security. If Washington can waive its own restrictions for $2 billion, why shouldn't they do the same thing? This isn't just a policy shift.
It's a credibility crisis. And while the chips may be flowing, the process that enabled them raises deeper questions about how decisions are made and who's making them. In Trump's Washington, the broader government no longer governs. It reacts. The NVIDIA deal wasn't vetted through the usual channels of export control or national security review. It was brokered in the Oval Office, bypassing the Bureau of Industry and Security and sidelining Congress entirely.
This isn't governance by process. It's governance by handshake. And it's not the first time that this has happened. The Nippon Steel acquisition was approved earlier this year only after Trump personally negotiated a golden share arrangement granting the US president veto power over corporate decisions. Intel CEO was publicly rebuked by Trump and then privately redeemed. Apple dodged taxes with an investment pledged and a gold plated plaque.
The pattern is quite clear. Policy outcomes hinge not on analysis and deliberation, but on executive discretion. Trump isn't just bypassing the system, he's replacing it with a loyalty test. This style of governance may work for huge companies like NVIDIA, but what about the rest
of the economy? Trump's hastily arranged and constantly adjusted tariff regime has created A2 tier system, one for mega cap companies with billion dollar war chests and direct access to the Oval Office, and another for the rest of American businesses. While the biggest businesses can secure meetings at the White House and lobby for favorable outcomes, small businesses are left absorbing the full cost of the trade war with no exemptions and no recourse.
This matters, as innovation doesn't typically come from the biggest businesses. It bubbles up from startups and firms that are challenging the status quo. When policy rewards incumbents and punishes newcomers, the result isn't just inefficiency, it's economic stagnation. the US economy has long thrived on competition, creativity, and the ability of small firms to disrupt giants. But in a pay to play economy, relationships start to matter
more than productivity. The winners are those who can afford to lobby, and the losers are those who can't afford a seat at the table. This cost isn't just borne by the small businesses, it's paid for by every citizen who suffers lower economic growth in an economy where old connections matter more than new ideas. China has been manipulating trade for decades. Not with tweets or tariffs. However, it's strategy is long term, structural and quietly coercive. China's goal isn't to win every
battle, it's to win the war. As an export driven economy, China is in a much weaker position than the United States. In any trade conflict, it needs America to buy Chinese production. Americans, on the other hand, don't need to overconsume. Yet despite this asymmetry, China has developed an arsenal of economic weapons with none more potent than its grip on rare earth materials.
Rare earths aren't actually very rare, but China dominates their production by selling them so cheaply that no other country even bothers to compete. It's a quiet monopoly built not on scarcity but on scale, refinement, efficiency and environmental tolerance. When China choked off exports in April, Ford had to halt a factory line in Chicago. Car makers in India and Japan curtailed production, too, and the EU scrambled to negotiate. The panic was real, and China's
leverage worked. The contrast with Trump's approach is striking. Instead of undercutting rivals like China does, he marked up chip exports, charging a premium for access. The result? A market signal to China to build their own chips. Where China uses price to entrench dominance, Trump uses price to provoke competition. Donald Trump is often seen by his supporters as a probusiness Republican, a traditional conservative and a champion of deregulation.
But his second term economic interventions tell a very different story. From the golden share in US deal to the revenue sharing deal with NVIDIA, Trump has repeatedly inserted executive branch into private enterprise not to deregulate but to direct action. These moves diverge sharply from the Reaganite model of limited government and free market autonomy. And while they don't amount to nationalization, they do echo the style of leaders who governed with a much heavier hand.
That's why so many people are comparing Trump to figures like Hugo Chavez, not as a moral judgment, but as a structural observation. Like Chavez, Trump has shown a preference for personal negotiation over institutional process and for loyalty over market logic. Chavez used socialism to justify executive control. Trump uses nationalism. Both cultivated direct relationships with their support base, bypassed traditional media channels and cast themselves as the sole interpreters of the
people's will. Trump's interventions may be lighter and his context far more constrained, but the pattern is familiar. A populist president using executive power to shape the private sector in service of his own goals. This is a huge departure from conservative orthodoxy, from institutional norms, and from the American idea that governments should set the rules but not pick the winners. The NVIDIA deal is by no means
an outlier. It's part of a long list of presidential interventions in private businesses, from Trump's golden share in US Steel, to his call for Intel CEO to step down, to his bizarre announcement that Coca-Cola would be switching from using high fructose corn syrup to cane sugar in it's US production after his encouragement. Each of these examples shares a common thread executive interference, not in the form of regulation, but in the form of negotiation. Trump isn't setting rules, he's
cutting deals. And the deals aren't based on any grand strategy. They're based on the whims of the president. The golden share in US deal is especially telling. It mirrors the very mechanisms the US has long criticized in China, state control through minority stakes and veto rights under trade law. the US treats foreign firms with golden shares as public bodies. Now, by its own definition, it's created one of its own. The NVIDIA deal may be without precedent, but it may not be the
last of its kind. Treasury Secretary Scott Bescend has already floated the idea of rolling out similar revenue sharing arrangements to other industries. What began as a one off chip deal could become a template, a new model of executive capitalism where export access is monetized and national security becomes negotiable. If this model expands to biotech, aerospace or energy,
the implications are profound. America's global leadership has long rested on principles, rule of law, institutional process, and strategic coherence. This new model risks replacing all three with a presidentially negotiated price tag. Trump's defenders hailed a deal as a masterstroke, a way to monetize, leverage and outmanoeuvre China. But even they must ask at what cost? And what happens when the next president, who Trump supporters may not like, wants a bigger cut?
Or when foreign governments retaliate with their own executive tolls. Thanks for tuning into this week's podcast. If you found it interesting, I'd really appreciate it if you could write a short review on whatever podcasting app you use to help the channel grow. Thanks as always to my supporters on Patreon whose contributions make the podcast happen. Have a great week and talk to you again soon. Bye.