Two Theories Behind Allowing Sales

This week we talk about NVIDIA, AI companies, and the US economy.

We also discuss the US-China chip-gap, mixed-use technologies, and export bans.

Recommended Book: Enshittification by Cory Doctorow

Transcript

I’ve spoken about this a few times in recent months, but it’s worth rehashing real quick because this collection of stories and entities are so central to what’s happening across a lot of the global economy, and is also fundamental, in a very load-bearing way, to the US economy right now.

As of November of 2025, around the same time that Nvidia, the maker of the world’s best AI-optimized chips at the moment became the world’s first company to achieve a $5 trillion market cap, the top seven highest-valued tech companies, including Nvidia, accounted for about 32% of the total value of the US stock market.

That’s an absolutely astonishing figure, as while Nvidia, Apple, Microsoft, Alphabet, Amazon, Broadcom, and Meta all have a fairly diverse footprint even beyond their AI efforts, a lot of that value for all of them is predicated on expected future income; which is to say, their market caps, their value according to that measure, is determined not by their current assets and revenue, but by what investors think or hope they’ll pull in and be worth in the future.

That’s important to note because historically the sorts of companies that have market caps that are many multiples of their current, more concrete values are startups; companies in their hatchling phase that have a good idea and some kind of big potential, a big moat around what they’re offering or a blue ocean sub-industry with little competition in which they can flourish, and investment is thus expected to help them grow fast.

These top seven tech companies, in contrast, are all very mature, have been around for a while and have a lot of infrastructure, employees, expenses, and all the other things we typically associated with mature businesses, not flashy startups with their best days hopefully ahead of them.

Some analysts have posited that part of why these companies are pushing the AI thing so hard, and in particular pushing the idea that they’re headed toward some kind of generally useful AI, or AGI, or superhuman AI that can do everyone’s jobs better and cheaper than humans can do them, is that in doing so, they’re imagining a world in which they, and they alone, because of the costs associated with building the data centers required to train and run the best-quality AI right now, are capable of producing basically an economy’s-worth of AI systems and bots and machines operated by those AI systems.

In other words, they’re creating, from whole cloth, an imagined scenario in which they’re not just worthy of startup-like valuations, worthy of market caps that are tens or hundreds of times their actual concrete value, because of those possible futures they’re imagining in public, but they’re the only companies worthy of those valuation multiples; the only companies that matter anymore.

It’s likely that even if this is the case, that the folks in charge of these companies, and the investors who have money in them who are likely to profit when the companies grow and grow, actually do believe what they’re telling everyone about the possibilities inherent in building these sorts of systems.

But there also seems to be a purely economic motive for exaggerating a lot and clearing out as much of the competition as possible as they grow bigger and bigger. Because maybe they’ll actually make what they’re saying they can make as a result of all that investment, that exuberance, but maybe, failing that, they’ll just be the last companies standing after the bubble bursts and an economic wildfire clears out all the smaller companies that couldn’t get the political relationships and sustaining cash they needed to survive the clear-out, if and when reality strikes and everyone realizes that sci-fi outcome isn’t gonna happen, or isn’t gonna happen any time soon.

What I’d like to talk about today is a recent decision by the US government to allow Nvidia to sell some of its high-powered chips to China, and why that decision is being near-universally derided by those in the know.

—

In early December 2025, after a lot of back-and-forthing on the matter, President Trump announced that the US government will allow Nvidia, which is a US-based company, to export its H200 processors to China. He also said that the US government will collect a 25% fee on these sales.

The H200 is Nvidia’s second-best chip for AI purposes, and it’s about six-times as powerful as the H20, which is currently the most advanced Nvidia chip that’s been cleared for sale to China. The Blackwell chip that is currently Nvidia’s most powerful AI offering is about 1.5-times faster than the H200 for training purposes, and five-times faster for AI inferencing, which is what they’re used for after a model is trained, and then it’s used for predictions, decisions, and so on.

The logic of keeping the highest-end chips from would-be competitors, especially military competitors like China, isn’t new—this is something the US and other governments have pretty much always done, and historically even higher-end gaming systems like Playstation consoles have been banned for export in some cases because the chips they contained could be repurposed for military things, like plucking them out and using them to guide missiles—Sony was initially unable to sell the Playstation 2 outside of Japan because it needed special permits to sell something so militarily capable outside the country, and it remained unsellable in countries like Iraq, Iran, and North Korea throughout its production period.

The concern with these Nvidia chips is that if China has access to the most powerful AI processors, it might be able to close the estimated 2-year gap between US companies and Chinese companies when it comes to the sophistication of their AI models and the power of their relevant chips. Beyond being potentially useful for productivity and other economic purposes, this hardware and software is broadly expected to shape the next generation of military hardware, and is already in use for all sorts of wartime and defense purposes, including sophisticated drones used by both sides in Ukraine. If the US loses this advantage, the thinking goes, China might step up its aggression in the South China Sea, potentially even moving up plans to invade Taiwan.

Thus, one approach, which has been in place since the Biden administration, has been to do everything possible to keep the best chips out of Chinese hands, because that would ostensibly slow them down, make them less capable of just splurging on the best hardware, which they could then use to further develop their local AI capabilities.

This approach, however, also incentivized the Chinese government to double-down on their own homegrown chip industry. Which again is still generally thought to be about 2-years behind the US industry, but it does seem to be closing the gap rapidly, mostly by copying designs and approaches used by companies around the world.

An alternative theory, the one that seems to be at least partly responsible for Trump’s about-face on this, is that if the US allows the sale of sufficiently powerful chips to China, the Chinese tech industry will become reliant on goods provided by US companies, and thus its own homegrown AI sector will shrivel and never fully close that gap. If necessary the US can then truncate or shut down those shipments, crippling the Chinese tech industry at a vital moment, and that would give the US the upper-hand in many future negotiations and scenarios.

Most analysts in this space no longer think this is a smart approach, because the Chinese government is wise to this tactic, using it itself all the time. And even in spaces where they have plenty of incoming resources from elsewhere, they still try to shore-up their own homegrown versions of the same, copying those international inputs rather than relying on them, so that someday they won’t need them anymore.

The same is generally thought to be true, here. Ever since the first Trump administration, when the US government started its trade war with China, the Chinese government has not been keen on ever relying on external governments and economies again, and it looks a lot more likely, based on what the Chinese government has said, and based on investments across the Chinese market on Chinese AI and chip companies following this announcement, that they’ll basically just scoop up as many Nvidia chips as they can, while they can, and primarily for the purpose of reverse-engineering those chips, speeding up their gap-closing with US companies, and then, as soon as possible, severing that tie, competing with Nvidia rather than relying on it.

This is an especially pressing matter right now, then, because the US economy, and basically all of its growth, is so completely reliant on AI tech and the chips that are allowing that tech to move forward.

If this plan by the US government doesn’t pan out and ends up being a short-term gain situation, a little bit of money earned from that 25% cut the government takes, and Ndvidia temporarily enriching itself further through Chinese sales, but in exchange both entities give up their advantage, long term, to Chinese AI companies and the Chinese government, that could be bad not just for AI companies around the world, which could be rapidly outcompeted by Chinese alternatives, but also all economies exposed to the US economy, which could be in for a long term correction, slump, or full-on depression.

Show Notes

https://www.nytimes.com/2025/12/09/us/politics/trump-nvidia-ai-chips-china.html

https://arstechnica.com/tech-policy/2025/12/us-taking-25-cut-of-nvidia-chip-sales-makes-no-sense-experts-say/

https://www.pcmag.com/news/20-years-later-how-concerns-about-weaponized-consoles-almost-sunk-the-ps2

https://archive.is/20251211090854/https://www.reuters.com/world/china/us-open-up-exports-nvidia-h200-chips-china-semafor-reports-2025-12-08/

https://theconversation.com/with-nvidias-second-best-ai-chips-headed-for-china-the-us-shifts-priorities-from-security-to-trade-271831

https://www.economist.com/business/2025/12/09/donald-trumps-flawed-plan-to-get-china-hooked-on-nvidia-chips

https://www.scmp.com/tech/tech-trends/article/3335900/chinas-moore-threads-unveil-ai-chip-road-map-rival-nvidias-cuda-system

https://www.investopedia.com/nvidia-just-became-the-first-usd5-trillion-company-monitor-these-crucial-stock-price-levels-11839114

https://aventis-advisors.com/ai-valuation-multiples/

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit letsknowthings.substack.com/subscribe