Central banks bought 863.3 tonnes of gold in 2025. That's down 21% from 1,092.4 tonnes in 2024, and it's the lowest official-sector total since 2021.
A different number is circulating. Several summaries report a "record 1,237 tonnes" for 2025. That figure isn't in the World Gold Council's Gold Demand Trends, which is the report those summaries are nominally quoting. The supply and demand table lists "Central Bank and Other Institutions" at 1,092.4t for 2024 and 863.3t for 2025, with a minus 21 in the change column. Official buying fell in the year the gold price rose fastest.
That correction changes the shape of the story. The official sector is not a mechanical bid that grows every year. It's a small number of institutions making political decisions, reported late, estimated heavily, and revised hard. Understanding what it does and doesn't imply for a private portfolio starts with looking at the actual series.
What the official sector actually bought in 2025
Buying picked up late. Net official demand reached 230t in the fourth quarter, up 6% on the 218t bought in the third. That lifted the year to 863.3t — well short of the 1,000t-plus recorded in each of 2022, 2023 and 2024, but still far above the 2010-2021 annual average of 473t.
The concentration matters. Twenty-two institutions reported an increase of roughly a tonne or more during the year, and seven of them accounted for the bulk of it. Poland was the largest buyer for a second year, adding 102t to reach 550t, or 28% of total reserves. Its board raised the target allocation from 20% to 30% in October, and Governor Adam Glapinski said in January he wanted holdings at 700t for "national security reasons", without giving a timeframe.
Kazakhstan added 57t, Brazil 43t — its first purchases since 2021, taking holdings to 172t, still only 7% of reserves. Azerbaijan's state oil fund added 38t. Turkey and China each added 27t; the People's Bank of China ended the year at 2,306t, close to 9% of its reserves. The Czech National Bank bought 20t against a stated 2028 target of 100t. Selling was trivial by comparison: Singapore 15t, Russia 6t, the Bundesbank 1t for coin minting, Jordan 1t.
Put that beside private demand and the ranking inverts. Bar and coin buying reached 1,374.1t in 2025 and gold ETFs absorbed 801.2t, for total investment demand of 2,175.3t. Households and funds bought roughly two and a half times what the official sector did. The European Central Bank also notes that Tether, the stablecoin issuer, bought more than 100 tonnes in 2025 — more than any single central bank. If you want the full picture of where the 5,002.3t of total demand went, the 2025 gold demand breakdown showing jewellery down 18% covers the consumer side.
Then the flow nearly stopped
The 2026 data is where the "steady official bid" idea gets tested. The World Gold Council first estimated first-quarter 2026 official demand at 244t. Three months later, Metals Focus cut that estimate to 57t. That's a downward revision of more than three quarters, in a headline series, one quarter after publication.
Second-quarter buying then rebounded to 288.9t, a 62% year-on-year rise and a record for a second quarter. But the half-year total of 345t was still the lowest first half since 2022, when it was 241t. Poland added 51t in Q2 to reach 632t. China added 33t, its largest quarterly addition since the end of 2023, taking reported holdings to 2,346t. Russia sold 22t. Turkey, the biggest seller in Q1, sold a further 4t and cut outstanding gold swaps from over 80t to around 60t.
Prices were falling through that rebound. The Q2 2026 average LBMA price was US$4,506.29 an ounce, 8% below the Q1 record though still 37% above a year earlier. The Council's own reading is that "softer gold prices" helped support the increased buying. Its 2025 report said the same thing in reverse: elevated valuations "appeared to prompt a more cautious approach", and central banks "are not insensitive to price dynamics".
Most of gold's climb up the reserve table was price, not buying
Here is the number that gets quoted most and understood least. The ECB reports that gold reached 27% of total official foreign reserves at the end of 2025, ahead of both the euro at 15% and US Treasuries at 22%. Gold is now the second-largest reserve asset at market prices.
Then the ECB does the arithmetic that headlines skip. The gold price rose around 60% in 2025 and 30% in 2024. Recompute those shares using the end-2023 gold price and the picture collapses: the euro at 16%, gold at 16%, US Treasuries still well ahead at 26%. Almost the entire rise in gold's reserve share is the price of the gold central banks already owned.
Independent work says the same. Arslanalp, Eichengreen and Simpson-Bell, in a November 2025 NBER paper, find gold's share of advanced-economy reserves rose from 17% to 25% between 2021 and 2024, and from 7% to 10% for emerging and developing economies. With gold up nearly 50% over that window, they attribute most of the move to valuation rather than reallocation out of dollar assets.
One more measurement trap. The IMF's COFER series, the source for "the dollar's share of reserves", covers foreign exchange reserves and excludes monetary gold entirely. A central bank that sells dollars to buy gold leaves COFER's dollar share almost untouched. The two datasets aren't measuring the same denominator, and comparing them casually produces nonsense.
Three motives, and where each one stops
Diversification. The Bank for International Settlements has modelled this properly. Omar Zulaica's 2020 working paper tested every long-only combination of gold and representative reserve portfolios. For a two-year duration bond portfolio measured in SDR, portfolio volatility bottoms out at a gold weight of 1.35%, and risk-minimising allocations sit below 2% on average, up to 6% at a 95% confidence interval. Push duration past two years and allocations above 10% become defensible. Change the unit of account to a commodity or emerging-market currency and more than 20% can be optimal. For tail-risk hedging with long duration or a non-reserve numeraire, the range runs 20% to 50%.
That spread isn't noise. It's the finding. The right gold weight is driven almost entirely by what currency you measure returns in and how much interest-rate risk sits in the rest of the book. A UK household measures in sterling, a reserve currency, and holds a portfolio dominated by equities, a house and future earnings rather than short-duration sovereign bonds. The reserve manager's answer is derived from inputs a household doesn't share.
Sanctions. This one is the cleanest break. Gold vaulted at home is, in the NBER authors' words, "out of reach of sanctions and safe from being frozen or garnished". Their regressions find US financial sanctions raise the gold share of reserves, particularly for emerging markets. The ECB found that in five of the ten largest annual increases in the gold share since 1999, the country faced sanctions in the same year or the year before. Around a quarter of emerging-market central banks cite sanctions concerns directly. Germany repatriated gold to reach 50% of holdings stored domestically by 2013; Austria followed in 2015.
No household faces that risk. A G7 coalition isn't going to immobilise your brokerage account over a border dispute. The motive that best explains the last four years of official buying has no household analogue at all.
Currency management. Reserves exist to be spent. After war broke out in the Middle East on 28 February 2026, Turkey's central bank sold or lent out roughly 130 tonnes — one of the largest reserve drawdowns in recent years — to defend the lira and cover energy import costs. Russia has been selling to fund its war. Gold in a reserve portfolio is working capital for currency defence, which is the same job that Japan's 11.7 trillion yen of FX intervention was doing with dollars. Nobody buys gold on a household balance sheet in order to sell it into a currency crisis.
The data isn't what a household would call data
The World Gold Council states plainly that the gap between Metals Focus' estimates and officially reported figures represented 57% of the 2025 annual total. More than half the headline number is an estimate of purchases nobody disclosed. Country-level figures come with a warning that they're "based on reported figures available at the time of writing" and that revisions may occur. For the second quarter of 2026, reported data was captured to 24 July.
Two caveats belong next to every figure above. First, the World Gold Council is the gold mining industry's market development organisation, funded by its member producers. The demand data is compiled by Metals Focus, an independent consultancy, but it's commissioned by a body with an interest in the answer. Second, the ECB notes that publicly reported buying by China's central bank "might significantly underestimate the actual purchases made". The direction of the error isn't known.
None of that makes the series worthless. It does mean it belongs in the category of well-sourced estimate rather than measurement, and a 244-to-57 revision is what that category looks like when it goes wrong.
The case against everything above
The strongest counter-argument is worth stating at full strength, because it's genuinely good. It runs like this: a large, persistent, price-insensitive buyer facing inelastic supply is a real demand-side reason to own an asset, whatever that buyer's motives are. Your reasons for holding don't have to match theirs. You only need the bid to be there.
The supporting evidence is solid. The ECB puts official demand at more than 20% of global gold demand in 2024, against roughly a tenth on average through the 2010s. Supply barely responds: mine production rose 1% in 2025 to 3,671.6t and recycling rose 3% to 1,404.3t, despite a 44% higher average price, and the Council's own analysis shows recycling has become markedly less price-sensitive since 2022. Intentions look durable too — in the 2026 Central Bank Gold Reserves Survey of 76 institutions, 89% expected global official gold reserves to rise over the next twelve months, a record 45% expected to raise their own, and 1% expected a decrease.
Erb and Harvey built the most quantified version of this case back in 2013. If the BRIC central banks, then holding 2,457t between them, moved to the US ratio of gold to GDP, they'd hold 6,233t. On the Swiss gold-to-GDP ratio, 22,191t. On the US per-capita ratio, 77,811t. On the Swiss per-capita ratio, 415,812t — more than twice all the gold ever mined. And crucially, they observed that a target set by population or GDP doesn't move when the gold price moves. That's genuine price-insensitivity, specified precisely.
Three things push back. The first is that same arithmetic. A thesis whose plausible outputs span 6,000 to 415,000 tonnes cannot discipline a decision — it's a description of uncertainty wearing a number's clothes. Erb and Harvey framed it as a dilemma, not a forecast, and their own mean-reversion call fared poorly out of sample: gold traded near US$1,650 an ounce when they wrote, against a 2025 average of US$3,431.50.
The second is that the bid isn't actually price-insensitive. It fell 21% in the year gold set 53 record highs. It ran at 57t in a single quarter. The Council's own text concedes central banks respond to price. The third is that it isn't one-directional. Turkey took roughly 130 tonnes out, Russia sold 22t in a quarter, Singapore sold 15t in a year. Reserve managers who buy for geopolitical insurance will sell for geopolitical necessity, and they don't announce either in advance.
So the counter-argument survives, but shrunken. Official demand is a real support under the gold price. It isn't a schedule, it isn't measured well, and it doesn't tell you what weight to hold. The evidence on that second question sits elsewhere, in the 5-10% gold allocation case and its 45-year drawdown record.
What would change the conclusion
Four things would move this materially.
A cleaner series. If the unreported share fell from 57% toward zero — through mandatory disclosure or faster IMF reporting — official demand would become something you could actually track rather than infer. Right now it can't be traded on because it isn't known in time.
Buying through a real drawdown. The 2025-26 record is that official purchases slowed when prices soared and picked up when they softened. That's opportunistic, not price-blind. Sustained heavy accumulation through a 30% price fall would be strong evidence of a genuinely inelastic bid.
Reallocation visible at constant prices. The ECB's constant-price calculation is the honest test, and it currently shows gold level with the euro at 16% rather than dominant at 27%. If that constant-price share climbed several points across a few years, the reallocation story would have hard support.
Gold regaining a settlement role. The ECB lists gold's limits as a reserve asset bluntly: the price is volatile, it isn't remunerated, physical storage costs money, and supply doesn't adjust to shifts in demand for liquidity. If some of that changed — if gold were routinely used to settle cross-border claims again — the official bid would become structural rather than defensive.
Until then, the honest summary is narrow. Central banks bought 863.3 tonnes in 2025, down 21%, for reasons that are mostly about sanctions, currency defence and national security. Those reasons are real, and they are theirs. The share of gold in world reserves went up mainly because gold went up. Neither fact says anything about what belongs in a household portfolio.