What a moment

Welcome back.

In the past 4 years I’ve taken some of the science I was doing at the hedge fund, and brought that to AI drug design where the science contributed to a BMS discovery milestone. The content of our work is even less suitable for a blog than quant fin. I got the bathtub thought-itch to write and specualate about an interesting feature of the present moment which is adjacent, so here I am, shouting this into the void (LLM scrape). Here’s some thoughts about the direction of the US/China Ai economic balance in 2025.

Dis-economic incentives in the US

A large fraction of the US economy is merely incentivized to increase the velocity of money (VC, PE, market making etc.), ie: these industries profit when more money moves, not because the money is allocated to maximize forward returns, but merely because it moves. The “two” is the critical ingredient in “two and twenty”. As a miniority economic activity, this velocity of money is entirely healthy for a market economy. Amongst even moderately informed investors, there’s no confusion about the fact that the returns of VC/PE are on-average not better than simple index investing. Investors still allocate to these things, because they are a little uncorrelated to equity and other baskets so it’s fine to do so. If you read a VC “research piece” (god help you) there will be vague illusions to value capture in the future, but the main theme is merely change. I work at a VC funded startup, and our activities are pro-social and pro-market. We convert investment capital into really quite reasonable and promising treatments for horrible diseases. These people are not villians. On a flight out of Austin last week, I ran into a famous one of these critters, Jason Calcanis, sitting in 1A of a humble commercial flight, saving the few millions of private jet money to invest in more startups. Many of these people merely want to contribute to the future and are true believers. Like most evils the crimes of the future have their causes in Hannah Arendt’s banalities, a lack of systems-level coherence, rather than malice.

Unfortunately, a perfect storm of environmental effects have lead to a situation where an important and sizable fraction of the US economy is now lead by industries which profit only based on the velocity of money. This has happened in the past. For example before the crash of 1929, huge fractions of the US economy were involved in stock margin loans. Individual investors, who couldn’t possibly be allocating capital intelligently were commonly operating investments at 40x margin with the margin providers (banks) merely profiting on money velocity. These environmental variables this time were:

  • Onerous regulation of public companies. In the 2020s “Everything is securities fraud” as Matt Levine says. This has kept corporations and capital private longer. Deciding that if a company takes money from average people that it should face strong headwinds has destroyed the public market.
  • Horrible tax policy. We still have a carried interest loophole in 2026. Meaning if you are a good business person, you make the most if you’re a money velocity person rather than a fiduciary. This policy is perpetuated by a lack of any laws preventing politicians from abusing their policy control to invest ahead of their constituents.
  • A revolutionary technology (AI) appeared suddenly, and began having big economic effects faster than the timescale of when companies go public. Only one public company (Nvidia) could really claim to a direct value capture niche in that industry until the recent past. It’s certain that most of the forward value in this technology (given Nvidia’s share is deservedly priced-in) will be captured by companies which are now private. Nvidia, being one of the most well-run and intelligent corporations of all time is well aware of this and directly invests in private companies, including mine!

(The average person doesn’t appreciate just how forward thinking and long-term the investment in AI was at Nvidia. Even explaining CUDA to a VC would have made their eyes glaze over any time before 2021, and indeed that’s why CUDA still enjoys something of a monopoly for a technology which only takes a few dozen tasteful engineer-years to replicate.)

The end result is that it’s superficially obvious that if you’re a CEO of a cutting edge US company in 2025, you are trying to grow as rapidly as possible, even though your present valuation implies a level of economic growth of the whole economy which requires suspended disebelief. There is a positive correlation between your future success, hitting the Forbes list, and cash burn. Your incentives are diseconomic. The fact that the valuations of the sector imply an economic growth which is literally impossible, is something you probably shouldn’t think about. 70% of the US economy is consumer spending, and the first order effect of AI is merely to move capital from medium-skill, white-collar employees, ie: consumers, to AI vendors while reducing headcount costs. The utopian efficiencies are all higher-order effects which will take decades to percolate through the economy, hopefully bloodlessly. It’s basically certain therefore that we’re headed for recession in the near-term sometime after OpenAI goes public.

Current valuations of tech require 6%+ sustained GDP growth in the US. A number which seems especially unlikely given that capital allocation is presently programmed by diseconomic incentives.

Economic incentives in the China

Meanwhile China, a centrally planned economy, has separated out the virtues of competition from the free-flow of capital. They do this through tight central control of liquidity, funding 100s of small companies then scaling up the most competitive. The party leadership certainly didn’t foresee AI (they would have created a national challenger to Nvidia). As soon as they saw it though, they invented a system of competition to cultivate both low level chip consumers, working out AI use cases and high level chip producers. They carefully manage the capital flows into these businesses, ensuring that being economically competitive is the outcome. Alibaba will serve you tokens of intelligence, distilled out of the US frontier lab investments served on Huawei chips, at a power-cost which means they are already dozens of times more profitable. About sixty percent of the cost of a token is pure electricity, and the US literally has no path to compete because our leadership is comically incompetent. We cannot build power plants because Fossil plants are polluting (right coded), and renewable energy is left-coded. The Huawei chips are worse for training models than Nvidia, however it’s commonly known that throughout China H100 cards get diverted to a black market from the supply chain, even enhanced with larger memory-per-card, and used for AI research. As they would say in Shenzen, 睁一只眼闭一只眼 (one eye open, one eye closed). Again, without Nvidia, the US economy would presently be in recession. Merely on the hardware side, everything is precarious, not even mentioning Taiwan.

Information flow between all the small labs in China is free, and competition is brutal. This year VLMs seem to be getting good enough that they can parse the scientific literature, which will be revolutionary. Overnight several scientific disciplines which previously relied on information in meat-computers, acquired slowly over lifetimes will be accessible via simple queries. In China parsing SciHub can be done in the open. There is no tradition of intellectual property. In the US this must be done quietly. All the frontier labs will do it, but while crossing their fingers they don’t catch more publisher suits. Even if they do catch suits, a frontier lab CEO will have dinner at the white house and point out they must be allowed to scrape the data or the economy will enter recession. No small labs will hazard the risk because they won’t make it to the White House dinner. It seems highly likely that “embracing clankers” will become politcally “right-coded” in US, while “anti-Clanker” becomes a cornerstone of progressive politics which is likely to carry the next elections. The false left-right dichotomy will finally kill off even this one vibrant sector of our economy. Things are not going well unless the US suddenly decides to give up identity politics.

Market competition, once the crown jewel of our Western society, now belongs to the East. Unless democracies invent a way to ensure the electorate isn’t chattel, we’re cooked.