EP18: The Vaults Refill

The Vaults Refill

For twenty years the world's central banks sold gold. Britain sold. Switzerland sold. The metal was called a relic of a dead system. Then the line flipped: central bank gold buying has now run positive for sixteen consecutive years, and since 2022 at roughly double the pace of the decade before. This is the story of the vaults refilling, told from the published ledgers of the institutions doing the buying.

The line that flipped

Take the numbers plainly. Through the 2010s, central banks bought on average 473 tonnes of gold a year, already a reversal of everything the previous generation had done.

Then came 2022. Net purchases reached 1,136 tonnes, the highest level of annual demand on record back to 1950, a figure the World Gold Council amended upward after correcting its own historical data for 1950 to 1969.

Then 2023 above a thousand tonnes, then 2024 above a thousand tonnes again at a restated 1,092.4. Three consecutive years at roughly twice the pace of the 2010s. And 2025 eased back to 863 tonnes, below the thousand tonne level seen in the preceding three years, a fall of 21 per cent, yet still far above the 473 tonne average, with twenty two institutions adding around a tonne or more.

This year is stranger still. The second quarter of 2026 brought 289 tonnes, a record high for a second quarter, roughly five times the revised first quarter figure of 57 tonnes. And here is the honest other half of that sentence, from the same page: central bank demand in the first half of 2026, at 345 tonnes, "was the lowest for a first half since 2022 (241t)". A soft start, then a record quarter. Both are true, and a series that told you only one of them would not be worth reading.

There is a second count, and it has just been read. On 3 September the Council published its monthly figures for July, and central banks "continued their gold accumulation in July with net buying reported at 23t". On that same monthly basis, "central banks reported purchases have totalled around 130t of gold" so far this year, "compared to a reported ~160t which was purchased over the same period last year".

Read those two paragraphs together and you get a number that looks wrong. Three hundred and forty five tonnes in the first half on one count, around a hundred and thirty tonnes to the end of July on the other. They are not the same measurement and they must never be added or set against each other. The monthly series counts only what institutions have publicly disclosed. The quarterly one does not stop there, and the Council says why in its own Q2 note: unreported buying, the activity that has yet to be disclosed, "was again elevated in Q2 and continued to be a notable feature of the market".

So the disclosed line is running below last year, and the estimated line is running at a record for the quarter. We give you both, because the gap between them is itself the finding: a good deal of this buying is not being announced at the time it happens.

Not one buyer, a queue

The interesting thing is the breadth. In the first half of this year the leaders were Poland at 82 tonnes, Uzbekistan at 41, China at 40 and Kazakhstan at 27, with the second quarter alone showing Poland at 51, China at 33, Uzbekistan at 16 and Kazakhstan at 15. Poland's reserves reached 632 tonnes, moving toward a publicly declared target of 700, and it added a further 8 tonnes in July.

July also shows the other side of the ledger, which a queue always has. Russia was the largest seller at 6 tonnes, with Turkey, Jordan and Uzbekistan each selling one. Uzbekistan therefore appears on both sides of the same year, which is what a working reserve looks like rather than a one way bet. Central banks are not a single mind.

The European Central Bank keeps its own league table of purchases since 2022, and it names China at over 350 tonnes, Poland at 320, Turkiye at 220 and India at 130. Overwhelmingly these are emerging market institutions: the countries furthest from the old money centres, buying the oldest money there is.

China deserves its own line. In July 2026 the People's Bank of China added 20 tonnes, its largest monthly addition since October 2023, taking holdings to 2,366 tonnes, or 8 per cent of its total reserves, in a buying streak of 21 consecutive months that is the longest on record. Whether that streak reached twenty two we cannot yet tell you: the August figure had not been published when this piece was updated on 5 September 2026.

February 2022

Ask why, and most people point at one date. Within days of Russia's invasion, the G7 immobilised the reserve assets of the Russian central bank. The Council on Foreign Relations puts the sum at around 260 billion euros, some 301 billion dollars, with over 193 billion euros of it held by Euroclear in Belgium and smaller amounts spread across British, French, Japanese, Luxembourgeois and Swiss custodians, while Brookings notes the estimates range from 280 billion dollars up to 330 billion. What was not frozen, on the CFR's own account, was Russia's gold and its yuan, which "were never at risk of being immobilized" because the metal sat at home.

Every reserve manager on earth watched that happen. The ECB's researchers later wrote it down: imposing financial sanctions is associated with increases in the share of central bank reserves held in gold, and in five of the ten largest annual increases in that share since 1999 the country involved faced sanctions in the same year or the year before. Roughly one emerging market central bank in four now cites sanctions concerns as a factor.

Now the other side, in the same breath. The buying began in 2010, twelve years before the freeze. And the CFR makes a point that cuts directly against the simple version: those reserves "only have value if a country is willing to barter goods for gold or if the gold is swapped for a currency like the Chinese yuan". Metal at home is metal nobody can switch off, and it is still only worth what somebody will hand over for it. An accelerant, then. Not an origin.

In their own words

You do not have to infer motive. Every year the World Gold Council asks the reserve managers, and in 2026 a record 76 answered, with fieldwork running from 5 February to 19 May. A record 45 per cent expect their own institution's gold to rise over the coming year, 83 per cent expect gold to be a higher share of total reserves in five years, 74 per cent expect the dollar's share to be lower, and a record 90 per cent cite gold's performance during times of crisis, ahead of long term store of value at 84 per cent and diversification at 82. Eighty nine per cent expect global official gold to keep rising, which is a large number and, in fairness, not a record: last year's equivalent was higher. The survey's own chapter puts that five year share expectation slightly differently, at 84 per cent, up from 76 the year before. We give you both rather than pick.

The scoreboard, and the caveat

By the ECB's June 2026 reckoning, gold had reached 27 per cent of total official foreign reserves at the end of 2025, ahead of United States Treasuries at 22 per cent and the euro at 15. Then the ECB states the caveat itself, and so must we: the gold price rose by around 60 per cent in 2025 and 30 per cent in 2024, which mechanically lifts gold's share. Adjust for that and gold and the euro sit level at 16 per cent each, with Treasuries at 26. Much of gold's climb up the table is the price of gold, which is not exactly a point against it.

Two rulers, both official

Is this de-dollarisation? It depends which ruler you use. The IMF's ruler counts currencies and excludes gold by construction. On that ruler the dollar's share of allocated reserves rose to 57.13 per cent in the first quarter of 2026 from 56.42 per cent, with about half the increase down to exchange rate movements, the euro slipping to 20.03 per cent and the renminbi sitting at 1.99. That first quarter reading was still the most recent when this piece was updated. The dollar's share was around 71 per cent when the euro launched in 1999. That is a quarter century of erosion, not a collapse, and nothing is replacing it.

The ECB's ruler counts the metal too, and tells the other story. Both are official. Hold both. Together they describe a slow rebalancing at the margin, away from promises and toward weight.

What the ledgers show

Central banks now hold approximately 38,600 tonnes of gold, worth approximately 5 trillion dollars and amounting to 17 per cent of the more than 220,000 tonnes mined throughout history, on the Council's end of 2025 valuation: close to one ounce in six of all the gold there has ever been, sitting in official vaults.

Priced today the pile is larger, and this next sum is our own arithmetic rather than anybody else's. At 4,431.10 dollars an ounce, 105.3956 pounds a gram, fetched at 08:36 on 5 September 2026, those 38,600 tonnes come to roughly 1.24 billion troy ounces and about 5.5 trillion dollars. The tonnage did the work; the price did the rest.

The institutions that issue the world's promises, the people who know better than anyone alive exactly what a promise is worth, have spent four years accumulating the one reserve asset that is nobody's promise at all. They publish their reasons, answer the surveys and state their targets, and a good part of the buying they do not announce until later. That is not a prediction and it is certainly not advice. It is simply what their own ledgers show.

Fifty five years ago they closed the window and called gold a relic on the way out. Today the vaults are refilling. If you want to understand why the difference between holding metal and holding a claim on metal matters so much to them, read Paper Gold and the Unsecured Creditor. And if you would rather hold the thing itself, our Vault collection is where we keep it, or you can sell your own gold to Chapman Gold.