"The AI Overwatch Act Would Help China." Neil Chilson's Op-Ed And What It Tells Us About The Regulatory Setup.
Neil Chilson, head of AI policy at the Abundance Institute, argues in today's WSJ Opinion section that the proposed AI Overwatch Act — the latest legislative attempt to formalize U.S. controls on AI exports — would, in his framing, "help China."
The core claim: where the U.S. has restricted exports, Chinese companies have made the biggest advances. Chilson cites Huawei's Ascend 950 chips, which he says are nearly three times as powerful as the H20 Nvidia ships into China and comparable to the H200 in aggregate system performance. The implication: U.S. controls aren't slowing China; they're creating the demand for domestic substitutes that wouldn't otherwise exist.
The Specific Argument
Chilson's framing: "China has its own systems and hardware that can run AI and wants to compete against ours for global market share. Proponents of export controls argue China is indigenizing tech as fast as it can regardless, but that isn't true. The massive share of China's CPU market that Intel (INTC) and AMD (AMD) have enjoyed for decades inhibits the development of Chinese alternatives."
The Huawei LogicFolding announcement from Monday (covered separately in this edition) actually supports Chilson's case. If Huawei can match 1.4nm chip density by 2031 without ASML machines, the U.S. export-control regime is, at best, slowing China rather than stopping it — and at worst, accelerating Chinese self-sufficiency by giving Beijing a clear strategic priority.
The Counterargument
The pro-controls camp would argue: export controls bought the U.S. five-to-seven years of leadership-stack dominance. During that window, Nvidia (NVDA), Microsoft (MSFT), and OpenAI / Anthropic built the model-and-tooling ecosystem that is now the global standard. The controls have not been free for China; the WSJ has reported repeatedly that Chinese AI labs are bandwidth- and compute-constrained relative to U.S. peers.
What the op-ed correctly notices: that five-to-seven-year window is closing. Whatever the U.S. wants to do next on AI policy — tighten, loosen, or radically restructure — needs to be decided in the next 18 months, not the next 5 years.
Why This Is Investable
The op-ed itself is interesting context, not actionable. What is actionable: the regulatory setup is now bipartisan and live. Both parties are now writing AI policy bills. The Pope's encyclical (this week), the chatbot-psychosis stories (this week), the data-center protests (ongoing), and the Huawei announcement (this week) all create pressure for Washington to act on AI policy in the next 12-18 months.
That doesn't tell you the direction of the policy. It does tell you that policy uncertainty is going to be elevated. Elevated policy uncertainty on a single sector (hyperscalers) means lower multiples until the uncertainty resolves. We saw the analogous setup in pharma in 2017-2019 around drug-pricing debates; valuations compressed even though earnings kept growing.
What We're Doing
Not selling the hyperscaler exposure (MSFT, GOOGL, NVDA, AMD, AMZN, META). The underlying demand — per Tuesday's "84% of small businesses using AI" piece — is too strong.
Pairing the exposure with deliberate hedges:
- Gold (IAU) at 12% across all tiers.
- Defense (LMT, RTX, GD, NOC, LHX) at 6-8% in Aggressive tiers.
- Energy (XOM, CVX, LNG, WMB) at 8-10% in Aggressive tiers.
- Dividend ballast (JPM, BAC, MO, JNJ, PG, KO) at 15-20% in Income tiers.
If Washington holds 30 days of high-profile AI hearings, expect a 5-10% drawdown in the hyperscaler basket. That would be a reinforcement opportunity, not a sell signal.