How AI Sovereign States Will Weaken the US Dollar

October 21st, 2024

I was talking to a VC friend today who is bullish on AIs using crypto and sees the future economy running on stablecoins. “Denominated in dollars?” I asked. “Of course” she said, “the most interesting AI stuff is happening in America.”

She’s not wrong. Of the frontier research labs: OpenAI, Anthropic, Meta, XAI, Mistal, only 1 is outside the US. The best AI agent companies are predominantly being built in the US in the Bay Area. The same is true for Nvidia, new AI hardware startups (Cerebras, Groq), etc.

Foundational models are typically based in English, as are the programming languages that make the internet run. Why shouldn’t we expect even more dollar dominance in this next era?

Europe has lagged in the last decades. The EU is starting to unravel, and other countries are already adopting a dollar standard (e.g. El Salvador & Ecuador). The joint powers of China, Russia, etc could perhaps bootstrap a new currency to displace the dollar, but the US’ dominance in AI combined with the dollar’s existing standing probably wins that fight.

So why not the dollar? Two words: sovereign states.

Everyone agrees the foundational model companies are going to be powerful. Whether they’re independent, or controlled by a larger bigtech funder (e.g. Meta), these models will be the basis for the new economy. They will control intelligence and with it, net new GDP creation.

In a world where the foundational model companies create most of the world’s GDP, the meaning of the public sector erodes. These few private companies will be responsible for wealth redistribution. Private companies will likely control transit & infrastructure (e.g. Tesla). Affordable housing will be built by robots developed by private companies. Sam Altman is planning for this future — that’s why he’s already been experimenting with UBI using his personal capital and via his project Worldcoin.

AI will create abundance, the AI companies will work to distribute that abundance to the masses, and the government will be left with little to do or extract. How does the dollar have more power in that world than say, OpenAI credits which govern model usage, UBI, etc.?

Resource deployment & AI’s takeoff moment

Historically, big companies retain their monopoly through regulatory capture -- paying off lobbyists and donating to PACs to get politicians on their side. But what does it mean to have the White House on your side when 99% of GDP creation is created online by AI? Bits are much harder to regulate than atoms, and the biggest AI companies will have the leverage to enact their own rules, currency, monetary policy, etc.

The level of resource deployment of AI companies will soon start to look like the resource deployment of nations. 3 OOMs from GPT-4 (~2028) will require the equivalent of an entire state's worth of energy and exceed the cost of the International Space Station. The US government is likely too bureaucratic to keep up with this pace of resource deployment, whereas focused private companies can likely find a path.

Meanwhile, every AI company will be fighting for a takeoff moment: the company most capable of automating AI research & deployment could turn a slight technological into a compounding advantage that quickly leads to a single dominant foundational model provider. An easy way to understand this is the first AI company that can automate a senior AI researcher, can then perhaps do the next 2 years of AI research in a few weeks. It’s not hard to see how this could lead to a monopoly. An AI company that has this sort of monopoly could be more responsible for the future of the land mass we call “America” than the US government itself.

The US government will try to slow AI, which further incentivizes getting off the dollar-standard

The government, if it can’t yet, will discover the threat of AI sovereign states before long. These arguments are clearly articulated in Leopold’s Situational Awareness. The US will do what it can to regulate AI companies -- to use monetary policy and taxation to limit their power accumulation.

Politicians will also be fighting for votes which may include appealing to populist policies around wealth redistribution, and rules about AIs replacing humans in jobs -- policies that limit both economic growth and the AI companies’ ability to scale.

One of the US government’s strongest assets is the control of the banking system. They can control inflation & real rates which impacts investment (including CapEx for model companies). They can also cut off companies & individuals from the banking system entirely if they misbehave.

Foundational model companies at the scale of threatening the government’s power would be much better off if they had control of the financial backbone for AI. If AI companies couldn’t be unbanked, could control monetary policy & inflation, and could earn directly from introducing the currency itself, they would accumulate much more power.

With enough time, I’d expect all the largest AI companies to at least introduce a robust credits system with a secondary market that mimics a currency. The biggest AI companies should go beyond a centralized credits system, using blockchain rails to offer a proper digital currency that can be used as a medium of exchange across the new AI economy.

The future of currencies -- what’s after the dollar?

1) Energy/Compute Denominated

One possibility for the future of currencies is what I’ve just mentioned — an effective unit of currency denominated in usage of these models. That is analogous to energy usage at the limit. So another way to frame this new reserve currency is an energy-denominated token. I would posit this token is either directly developed by a foundational model company (or a consortium of them) or by an energy provider (e.g. one of Sam Altman’s fusion bets, if they pay off). Bitcoin is highly correlated to this but represents spent compute rather than spendable compute.

2) Denomination-Agnostic Exchange

The second possible outcome is an unstable medium of exchange currency. AI agents and humans hold value in scarce resources (e.g. Bitcoin and gold). They then convert these scarce resources into digital money for transactions. The medium of exchange currency they use may sometimes be dollar-denominated (e.g. USDC) perhaps sometimes it’s Bitcoin denominated, ETH, or SOL, and other times it’s a native token for a specific chain or protocol (e.g. a compute token to exchange for a training run). So long as swapping costs trend to 0 and instant transaction times, nobody should care about the medium of exchange for transactions.

3) Subsidized Dominant Currency

The third outcome is we have a dominant, consistent currency that isn’t the dollar, and isn’t energy denominated. How could this happen? Agents will use whatever currency is fastest, cheapest, and most convenient. In order for a single token to win out, it would have to somehow have a durable technological advantage, with no better chain existing for any use cases. Even with a clear technological leader, it’s likely that certain use cases would have better tradeoffs between speed/cost/security with another chain & token.

The other possibility is that some token has a significant subsidy — this would incentivize everyone to transact in this token even if there’s other tokens that offer other slight advantages in other ways. Who could subsidize such a token? Well only a company interacting with nearly every onchain transaction and earning a transaction fee from it (e.g. a foundational model company, compute provider, or a flow of funds / payment infra provider. Think about Stripe offering 0% transaction fees if you use the fictional $STRIPE token — then every superintelligent AI would transact in $STRIPE.

Conclusions and open questions

It is unclear to me exactly what the future currency of the world is. I think that #3 — advantage of a single currency by subsidy seems the most likely, but it’s possible an energy denominated token wins out instead if the infra stack for crypto payments and AI ends up being distributed between many smaller players instead of concentrated in a few monopolies. Regardless though, it’s clear if you’re bullish AI long term, you must be bearish government-run financial institutions.