The World's Top-Performing Sovereign Fund Just Warned of a US Pullback — and It's Sitting on US$31.7B of NVIDIA

The Guardians disclosed portfolio holdings as of December 31, 2025. Its top five positions by value were a NZ$3B stake in NVIDIA, followed by Apple, Microsoft, Alphabet and Amazon — the five largest constituents of the Magnificent Seven by market cap and the spine of the S&P 500's 71.5% five-year rally. The full US book of NZ$31.7B represented roughly a third of the entire NZ Super portfolio, with about half the fund in equities overall. That concentration is the operational reason a top fund's warning on US valuations is more than a quote.

The disclosure matters because it shows that the fund is not a passive index hugger fighting the benchmark. Co-CIO Brad Dunstan told i3 Insights in May 2026 that NZ Super believes global equities are overvalued and are 'likely to revert to historical averages,' but is staying long because the alternative-asset toolkit can still extract alpha. The September message is the public version of that internal debate — and the first time the CEO has named the trigger as the US book specifically.

Strategists have been calling US equities expensive since CAPE crossed 30. What changes when the warning comes from a sovereign wealth fund — even a relatively small one at US$54B — is the signal it sends to the official-sector ecosystem that has been the marginal buyer of US mega-cap risk for most of the cycle. Norway's Government Pension Fund Global, the world's largest SWF at ~US$2.3T, said on Aug 12, 2026 that it has 'no plans' to reduce US assets and posted a record NOK 1,753B (≈US$184.9B) profit in H1 with equities up 13.0%. NBIM's NVIDIA stake alone was reported at 1.3%, worth US$61.8B. When the smaller, more cautious Kiwi fund speaks in public, it is in part speaking for a wider constituency that includes pension funds, endowments and family offices benchmarking against it.

Townsend's reversion argument is anchored in two pieces of math the market can verify. First, the S&P 500 has returned 71.5% over the past five years, or 11.4% annualized, versus a long-run average close to 10% — so the cushion above trend is roughly 140 basis points per year. Second, the Shiller CAPE ratio at 40.16 sits in the top 1% of its 145-year history; only the dot-com peak was higher. NZ Super's cut to its long-term return assumption (7.8% → 7.2%) maps directly onto that valuation gap, and the active-risk reduction is the operational way of saying it intends to underweight the most crowded trades inside its US book.

The 60-basis-point cut in the long-term assumption is the cleanest single number to anchor expectations. If a fund that has beaten its passive reference by NZ$22B over 20 years now expects to add only ~7.2% a year going forward, it is implicitly pricing in a multi-year period of US equity returns closer to 6–8% rather than the 11% trailing clip — a re-rating, not a crash, but a meaningful one for the AI mega-caps whose valuations bake in continued double-digit growth.

A reversion event doesn't have to be a crash. Even a return to the long-run average across US equities compresses the AI mega-cap trade disproportionately because that's where the index premium lives. The listed names fall into three transmission tiers: the mega-caps themselves (the direct mark-to-market hit), upstream semiconductor enablers (revenue tied to mega-cap capex), and downstream AI infrastructure beneficiaries (whose multiple expansion relies on the AI capex narrative continuing).

Over the short window (days to quarters), the trade is reactive: NZ Super's headline has already circulated in FT, CNBC, Reuters and Bloomberg, and the next move belongs to other large SWFs and the November NVIDIA earnings print. If hyperscaler capex guidance softens, the upstream complex (Taiwan Semiconductor, ASML) is the first place the multiple compression shows up because order-book multiples are the most sensitive to forward capex. Over a 1–3 year window, the structural picture diverges by tier: power and grid-constrained plays (Constellation Energy, NextEra Energy, Vertiv) can rerate higher as the binding constraint shifts from capital to electrons, while the mega-caps themselves need to convert their AI capex into earnings growth to defend their current multiples.

Three takeaways. First, the warning is from a real official-sector actor with a track record, not a sell-side strategist — NZ Super's 9.68% 20-year annualized return is the kind of number that funds benchmark against. Second, the warning is not that US equities will crash; it is that they will mean-revert, which is a slower and more grinding kind of headwind that compresses multiples without necessarily triggering a recession. Third, the transmission is asymmetric: the mega-caps that drove the last five years carry the most downside to a reversion, the upstream semiconductor enablers carry the next most, and the AI infrastructure beneficiaries downstream are partly insulated because their thesis is increasingly about physical scarcity rather than equity-market beta.