The Trillion-Dollar Cover-Up: What Washington Knew About Jensen Huang

NVIDIA

Across the shows: 114 speakers in other episodes, 98 bullish and 69 bearish calls, past 30 days

• Chris McGuire · Council on Foreign RelationsBullish

  Export controls have cost Nvidia no sales because all its controlled chips still sell out.

  McGuire argues demand for AI chips and compute is voracious. Musk and Jassy say they cannot get enough chips, and Jensen says products are sold out. Every chip Nvidia has made since controls began has still been sold, and it has no excess supply. If Nvidia did not sell to Megaspeed, the chips would go to SpaceX, Meta, Google, Amazon or Microsoft. He is less concerned about Nvidia's longer-term sales, though he expects big fallout for the stock from the investigation.

  “The export controls so far, I would argue, have cost NVIDIA exactly zero sales. Every single chip they have made since their chips started to be controlled has actually still been bought.”at 18:56

• Chris McGuire · Council on Foreign RelationsUncertain

  Nvidia's export-control advocacy has no credibility after Jensen's poor judgment put security at risk.

  McGuire says Nvidia has been very vocal about export controls and about sending chips to China, and that this now carries no credibility. He says Jensen Huang built the best-products company in the world but executed extremely poor judgment in this case and put US national security at risk. He says it is up to investigators to decide what that means for the company's future, and that the policy approach has very low credibility too.

  “I think it's also very clear that he executed extremely poor judgment in this case and has put The US National Security at risk.”at 20:29

Megaspeed

Across the shows: 1 speaker in other episodes, 0 bullish and 1 bearish calls, past 30 days

• Chris McGuire · Council on Foreign RelationsBearish

  Megaspeed is a Chinese company led by Chinese spies, yet it still receives AI chips.

  Megaspeed, a Singapore-based Nvidia customer, is in McGuire's view clearly a Chinese company being run by Chinese spies. Those spies are trying to buy AI chips and infiltrate Nvidia, and the chips are still flowing to the company. He argues it should have been cut off under US policy, which says AI chips do not go to Chinese companies. He also says Megaspeed is still listed as a cloud partner on Nvidia's website. He is careful to say he is not accusing anyone of knowing about it.

  “[Megaspeed] very clearly was and is a Chinese company. It's being led by Chinese spies.”at 5:24

US–China AI & chip race

• Chris McGuire · Council on Foreign Relations

  Commerce is refusing to use entity listings and not enforcing chip rules, so untrusted buyers keep getting chips.

  McGuire argues the Commerce Department has barely used its entity-listing tool. The current period between blacklistings is the longest in decades. A Biden-era rule from January 2025, which the Trump administration said in May 2025 it would not enforce, would stop untrusted companies in Malaysia, Singapore and other non-allied countries from receiving billions of dollars of AI chips without a license. He calls that a rational policy and says the lack of enforcement leaves the government with no visibility into who is buying the chips.

  “We are in the longest period between entity listings, between blacklistings of companies in the last many, many years for commerce, certainly going back decades.”at 12:14

• Chris McGuire · Council on Foreign Relations

  Huge AI chip shipments to Singapore, Thailand and Malaysia suggest large-scale diversion to China.

  Trade data shows about $80 billion of AI chips went to Singapore, Thailand and Malaysia this year, making them the second, fourth and fifth largest importers of AI chips after the US. Roughly half of AI chips that do not go to the US end up in those three countries, which lack robust AI industries of their own. Many of the data centres there are run by cloud providers supplying compute to China. McGuire says Megaspeed is 'undoubtedly the tip of the iceberg' and that a slew of similar companies are also receiving chips. He calls this export-control circumvention by China.

  “There have been $80,000,000,000 of AI chips that this year that went to Singapore, Thailand, and Malaysia.”at 17:32

Frontier AI labs & model race

• Ara Kharazian · Ramp

  Compute-rich labs like SpaceX AI and Meta will soon turn the OpenAI–Anthropic duopoly into a four-way race.

  SpaceX AI has a highly performing model that is popular with engineering teams and is gaining market share. It also has a compute advantage, since it owns many chips and rents them out to Anthropic. If OpenAI and Anthropic hit a funding crunch and cannot invest in compute, companies with compute already in hand will have the advantage. Meta is not a big player in business AI spend yet, but Ara expects it to follow the same path, which will intensify price competition.

  “They're gonna expand their market share there. And what is today a two way race between [OpenAI] and [Anthropic] is very quickly gonna become a four way race between them and SpaceX AI, Meta, maybe some other players.”at 28:47

• Ara Kharazian · Ramp

  Model companies' reported ARR is not truly recurring because enterprises can easily switch providers.

  Model companies report ARR, which is recurring revenue annualised. But businesses can move their spend from one provider to another easily, and they are already finding ways to cut and control it. Market share can shift between companies within weeks, not months, as new frontier models pull demand from rivals. So Ara thinks it is hard to treat any of this spend as recurring, which will be a big challenge for model companies.

  “It's also one of the funny things about how the model companies have been reporting ARR... But it's hard to think of any of the spend as recurring when businesses can so easily move it from one provider to another.”at 30:21

Enterprise AI adoption & AI software

• Ara Kharazian · Ramp

  Engineering-led enterprise AI spend, the largest segment, is plateauing as teams optimise costs.

  Engineering use is the largest segment of enterprise AI spend, and it is starting to plateau. Engineering teams are the ones now incentivised to spend on AI more efficiently, which is what Ramp hears from its customers. Whether AI spend as a whole flatlines depends on whether model companies can build commercially advanced versions that drive productivity gains for teams outside engineering.

  “The largest segment of spend in terms of enterprise spend on AI, which is engineering use, is starting to plateau. Because those are precisely the teams that are incentivized now to figure out how to spend on AI more efficiently.”at 31:47

• Ara Kharazian · Ramp

  AI for customer support and cybersecurity could drive enterprise spend beyond engineering.

  Customer support operations is the second-largest department for AI spend on Ramp's platform after engineering. Ara expects some spend to shift there and to accelerate growth. He also thinks AI used to defend against cyberattacks will be another accelerant. These are the most likely takers of growth outside engineering, though whether they can match the hyper-growth of coding remains an open question.

  “AI for customer support operations, that's the second largest department in terms of spend on our platform that we can observe after engineering. You can imagine some spend is gonna be taken up by that.”at 32:29

AI pricing & unit economics

• Ara Kharazian · Ramp

  Flat enterprise AI spend reflects a token price war, not weakening demand.

  Ramp data shows AI spend has been roughly flat since August, even though firms use more models, spread spending across providers and use more tokens. Token prices have fallen sharply because of a price war between model providers. So spending is flat not because demand is weak but because each token now costs much less. He says that opens new questions for OpenAI and Anthropic about where growth comes from next.

  “It's not a demand issue and firms are using more AI by pretty much any measure. They're using more models... but it's that those tokens have come down so significantly in price because of this price war that their actual spend has remained relatively flat.”at 26:14

Open-source AI models

• Ara Kharazian · Ramp

  Chinese models hold under 5% of business AI spend, so the price war is not China-driven.

  Ramp's dataset shows Chinese models have less than 5% of business AI spend. Ara says that even if all Chinese models were banned tomorrow, OpenAI and Anthropic would still face price competition from each other and from new American labs. The price fall is therefore driven by domestic competition and compute access, not by open-weight models flooding the market.

  “Chinese models have less than 5% of the market as far as business spend goes.”at 27:36

Venture capital & startups

• Ara Kharazian · Ramp

  OpenAI and Anthropic revenue depends on VC-backed firms, so a VC pullback would hit them disproportionately.

  Ramp data shows the top 1% of companies drive 80% of spend on OpenAI and Anthropic, and pretty much all of them are VC-backed, mainly in tech, finance and technical services. Concentration is fine in itself. It becomes a problem when the customers are highly correlated, as VC-backed firms are because their dollars move together regardless of company or use case. If spending slows, the effects would cascade through the stack and hit OpenAI and Anthropic's revenues disproportionately.

  “Concentration in industry is fine. It's a problem when it's highly correlated, especially when it's something like VC where these dollars tend to move with each other no matter the company or its use case.”at 36:54

AI: overall outlook

• Ara Kharazian · Ramp

  AI trade's big risk is that OpenAI and Anthropic still depend on tech-sector customers; not yet proven otherwise.

  Kharazian says one of the big risks to the AI trade today is that AI spending is concentrated among tech companies. If AI labs' spending has a disproportionate effect on markets, the damage can reach parts of the economy that don't touch AI. He says he would change his view if OpenAI and Anthropic develop a commercially advanced product that generates enough ROI outside the tech sector and becomes less reliant on tech companies. He treats that as an unproven 'if' that the labs have not yet demonstrated.

  “if [OpenAI] and Anthropic are able to develop a commercially advanced product that drives sufficient ROI outside of the tech sector... then I would start to change my tune. I wouldn't worry about this at all necessarily. But that is an if, and it's something that they have not yet necessarily demonstrated.”at 37:36