Why a Sovereign Wealth Fund AI Indexing Strategy Can’t Stop at the S&P 500
Trammell and Imas argue developing countries can't just buy SPY and call it AI exposure. Anthropic is private, SK Hynix is Korean, and the index misses the supply-side rents that matter most.
In a June 2026 Dwarkesh Podcast conversation with economist Alex Imas, Phil Trammell asked a question every emerging-market finance ministry should be wrestling with: does Nigeria own a lot of SK Hynix and Anthropic? His answer, captured in the Dwarkesh transcript, was probably not.
The line “It’s not enough for them to just own the S&P 500” is the cleanest one-sentence summary of why a sovereign wealth fund AI indexing strategy needs more than SPY.
The naive version of the strategy is straightforward. A developing country with savings to deploy buys a basket of US large-caps, gets paid for the AI boom through the earnings of the firms doing the booming, and avoids the political risk of trying to stand up a domestic frontier lab from scratch.
The argument
That logic has two cracks the index does not fix.
First, a non-trivial share of the frontier sits outside the public market. In his essay “Capital in the 22nd Century”, Trammell estimates the fraction of US corporate capital held by private firms has roughly doubled since 2000 and now sits under 20%, with the trajectory still climbing.
The long-run drift toward private ownership is corroborated by Morgan Stanley Investment Management’s public-to-private paper and by quant work from PitchBook and HarbourVest. The AI buildout has accelerated it. Anthropic, OpenAI, and xAI are not in SPY. Neither is the bulk of the private compute layer that funds them.
Second, the names that own the chip supply chain are not American. Nvidia (NVDA) is, but Nvidia’s designs are fabricated by Taiwan Semiconductor (TSM), the high-bandwidth memory comes from South Korea’s SK Hynix, the lithography from ASML (ASML) in the Netherlands, and the tool stack from a long tail of Japanese and European specialists.
A sovereign fund that caps its public-equity sleeve at the US index buys downstream demand for AI without owning the supply-side rents. Siblis Research’s market-cap dataset makes the home-bias problem visible: SPY is a US-market product, not a world-AI product.
| Exposure layer | Where the wealth accrues | In SPY? |
|---|---|---|
| Frontier model labs | Anthropic, OpenAI, xAI | No |
| AI accelerators | Nvidia (NVDA), AMD (AMD) | Yes |
| HBM memory | SK Hynix (Korea), Samsung (Korea) | No |
| Foundry | TSMC (TSM, Taiwan) | No |
| Lithography | ASML (ASML, Netherlands) | No |
| Hyperscaler capex demand | Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), Meta (META) | Yes |
The table is the argument. Roughly half of the AI value chain is missing from the default index trade.
The strongest counter, and the rebuttal
The most honest counter comes from Trammell himself, a few minutes later in the same conversation. “Every company in the S&P of the future, if it’s going to make it to the S&P 500, it is because it has leveraged AI”, he tells Imas. If AI behaves like electricity, the index becomes the AI trade by default. You hold SPY and you are, mechanically, holding the AI economy. Booboone’s writeup of the same exchange flags this as the more interesting concession.
It is the strongest case for sitting still. It is also incomplete.
Electricity got commoditized because nobody owned the rents of being-an-electricity-company. AI looks less symmetric. The producer surplus appears to concentrate in the labs, the accelerator designers, and the memory and foundry oligopoly.
Even if every S&P 500 firm uses AI, the abnormal returns accrue to a much smaller set of names, and that set is precisely the one the US index misses. Indexing the consumer side of a transformation is a fine way to make ordinary returns. It is a poor way to make sovereign-scale ones.
Conclusion: the stance, qualified
Indexing is a floor, not a ceiling. A developing-country sovereign wealth fund that owns SPY and stops there has not bought AI exposure. It has bought US large-cap exposure with whatever AI tilt the index happens to carry.
A more honest mandate would pair the US index sleeve with a global-equity sleeve covering Korean, Taiwanese, Dutch, and Japanese compute names, a private-markets sleeve sized for frontier-lab vintages, and a separate budget for direct compute infrastructure stakes where capital controls allow them.
That is an opinion, not advice. Nothing here is telling any specific fund what to allocate, only what an index alone fails to capture.
What this does not tell you
This piece does not model returns. It does not tell you which private vehicles a sovereign wealth fund can actually access, since most frontier-lab capital tables are gated by relationship and US national-security review. It does not address CFIUS-style restrictions that already block some sovereign capital from US AI infrastructure.
It treats the private share of US corporate capital as still under 20% per Trammell’s own estimate, with real uncertainty around that figure. None of the public sources used here disclose any individual fund’s positioning.
FAQ
If AI lifts every company, doesn’t owning the index eventually catch the AI trade?
Partially. Indexing captures the consumer-side gains, which are real. It misses the producer-side rents, which historically concentrate in a small number of compute, memory, and model names that are either foreign-listed or still private.
Can a sovereign wealth fund actually buy stakes in Anthropic or OpenAI?
Some can, through late-stage rounds, secondaries, and structured vehicles. Many cannot, because of foreign-investment screening or because allocations are gated to existing relationships. The constraint is a reason to layer foreign listed exposure and direct compute infrastructure on top of US indexing.
Disclaimer. This article is opinion and editorial commentary, not investment advice. It is not a recommendation to buy, sell, or hold any security, index, fund, or asset class. Elite CurrenSea is not advising any sovereign wealth fund, government, or institution on portfolio construction.
Estimates of private-market share, supply-chain concentration, and AI value capture are inherently uncertain and were stated by the cited authors at a point in time. Past positioning and forward-looking views can change without notice. Readers should consult their own licensed advisors and not act on the views above.
