How I Used AI To Position Around China's Treasury Exit

Last week, Chinese regulators told banks to limit new purchases of US Treasuries and reduce exposure.

This comes after China’s Treasury holdings fell to $682.6 billion in November 2025. 17-year low. Down 47% from the $1.32 trillion peak. They dropped to 3rd largest foreign holder behind Japan and the UK for the first time in 17+ years.

The narrative everywhere: China is dumping Treasuries. De-dollarization is accelerating. Buy gold, sell bonds.

I'm fascinated by investing x geopolitics, so I started digging into this. If China is genuinely moving away from Treasuries, the implications for the US are massive, both economically and geopolitically. Dollar reserve status, US borrowing costs, the world order. Not small questions.

But I think the market is misreading this story.

If China is pulling hundreds of billions out of Treasuries … where is that money going? And are there assets that could be mispriced?

Tracking sovereign capital flows manually is a multi-day project: US Treasury data, PBOC disclosures, World Gold Council reports, IMF data, sell-side desks.

I built a two-prompt chain that does this in about 15 minutes.

Prompt 1 (Sovereign Reserve Analysis) maps where capital is actually flowing
Prompt 2 (Trade Menu) builds trade ideas with scenarios, invalidation triggers, and risk/reward frameworks

Both prompts are templatized. Swap “China” for Japan or Saudi Arabia and you get the same depth.

Was this email forwarded? Subscribe to get more content on using AI for investing.

Subscribe

Part 1: Where is the money actually going?

The Prompt

Prompt 1: https://www.finprompter.com/share/abd958bf-9df1-42ab-b6d3-e1b0aaef50e8

The Result:

Full output here: https://claude.ai/public/artifacts/c7ffab9c-22bb-41e4-b9a2-e3d3a31f3f0a

The biggest finding: the headline number is misleading.

Everyone quotes the $634B selldown. But China doesn’t hold all its Treasuries under its own name. It routes a large portion through custodian banks in Belgium and Luxembourg. Think of it like buying a house through an LLC. You own it, but your name isn’t on the public record.

The Treasury’s own data: Belgium went from $374.6B to $481.0B in one year (+28%). Over that same window, China’s direct holdings fell from $759.0B to $682.6B. If China is truly “dumping,” why are the custody hubs surging?

AI also flagged China rotating into US agency bonds (Fannie Mae, Freddie Mac). Still dollar assets, just a different instrument.

The real picture: China’s total dollar exposure has barely changed at ~$1.8 trillion. The money moved from Treasuries to agencies, from US custodians to European custodians, and from long-term bonds to short-term bonds. A reshuffling within the dollar system, not an exit from it.

So what is actually changing?

Gold. That’s the real structural shift. The PBOC has been buying for 15 straight months. Official holdings: 2,307 tonnes. But AI flagged multiple independent estimates converging on actual holdings of 4,000-5,400 tonnes … potentially 2-3x what’s reported.

European government bonds are the second destination. The euro’s share of global reserves hit 20.3%, highest since late 2022. And for what it’s worth: the yuan’s share has actually declined 30% since 2021. The world isn’t replacing dollars with yuan. It’s replacing them with gold.

Part 2: How could you position?

The Prompt

Prompt 2: https://www.finprompter.com/share/7d2e282b-d553-4cd3-887b-fea787d7c4f6

The Result:

Full output here: https://claude.ai/public/artifacts/04add324-f7f8-453c-8519-a23e2cc400c3

AI stress-tested the research against three scenarios: steady diversification (55%), geopolitical acceleration (20%), and a US-China deal that stabilizes things (25%). From the full menu of 10 trade ideas, here are three worth exploring.

Idea #1: Long gold and gold miners (GLD, IAU, GDX, GDXJ). Central banks have been buying 1,000+ tonnes per year for four consecutive years. The buyer isn’t price-sensitive. They’re buying for geopolitical insurance, not returns. That creates a demand floor different from previous gold cycles.

Gold is 8-9% of China’s reserves. The global average is 15-17%. The US and Germany hold 78%. That gap suggests a potential multi-year buying runway. Gold miners are particularly interesting here because mining costs are mostly fixed (~$1,400-1,600/oz) while revenue scales with gold price. If gold stays elevated, miners could see outsized margin expansion that hasn’t fully shown up in valuations yet.

Key risk: if the yuan weakens and the PBOC sells reserves to defend the currency (the 2015-2016 playbook), gold could correct meaningfully.

Idea #2: Treasury curve steepener (Long 2Y / Short 30Y). A “curve steepener” profits when the gap between short-term and long-term yields gets wider. Two forces could push in the same direction: Fed cuts pull the short end lower, while China selling long-dated Treasuries + expanding US deficits could push the long end higher.

Key risk: a recession scare driving flight to safety into long-dated bonds.

Idea #3: Dollar weakness (UDN, FXE). The dollar’s share of global reserves is at a 30-year low (57.7%). But DXY has already fallen ~10% from late 2024 peaks. Much of the move may be priced in.

Three consensus traps worth flagging

Trap #1: “Sell all your Treasuries because China is dumping.” As covered above, China’s total dollar exposure is roughly stable at $1.8T. The Belgium TIC number ($481B, up $120B in one year) is the tell. The real question isn’t whether China is leaving the dollar. It’s where within the dollar complex flows are shifting. That’s why the steepener may be more interesting than an outright short.

Trap #2: “Go long yuan assets because China is de-dollarizing.” This sounds logical but the data says the opposite. The yuan’s share of global reserves has fallen from 2.8% to 1.93% since 2021. Even China’s own allies aren’t stockpiling yuan. Capital controls, limited convertibility, and geopolitical risk make the yuan a poor reserve asset. China is diversifying away from dollars, but the rest of the world is not diversifying toward yuan.

Trap #3: “Gold is a bubble at these levels.” The “bubble” framing assumes the buyer is speculative. But the marginal buyer here is central banks with multi-decade time horizons who don’t care about the price. That’s structurally different from 2011 when gold was driven by retail and ETF flows. Could gold correct? Absolutely, especially if we get a US-China deal. But calling it a bubble misidentifies who’s driving the demand.

Bottom line

The market is treating this as a “sell bonds, buy gold” story. It’s more nuanced than that.

Gold and gold miners may have the most direct connection to the sovereign flow thesis. The curve steepener is worth studying. And several popular trades around this theme could be based on incomplete data.

Try Prompt 1 on Japan. Largest foreign holder at $1.2T, and Japanese insurers are already repatriating capital. Same prompt, different sovereign.

Personal

Just got back from a road show across Asia. The AI hype in finance over there is very real. Everyone I met is actively ramping up + PMs, analysts, ops teams. Impressive effort.