2026-09-01 · Gold & commodities

Gold: four wrappers, one metal — GLD, PAXG, XAUT and the perp

Data as of 01.09.2026 unless another date is given next to a figure.

Verdict: gold wrappers aren't junk, but they aren't metal either. Three of the four are claims on real metal, and the metal does exist; the fourth, the perpetual, has no metal at all. The question that actually decides it: how many links of trust sit between the holder and the bar, and which of them can cut the chain. In a crisis each wrapper protects against something different: GLD against a crypto crisis, PAXG in self-custody against a brokerage crisis, the perp isn't gold at all. In November 2022 Paxos froze 11,184 tokens worth $19 million on law-enforcement instructions.

What it is — four different things, not four ways to buy gold

Prices and size: PAXG $4,431.92 per token, market cap $1.91bn (31.08.2026) · XAUT about $2.79bn, 16.2 tonnes as of December 2025, about 590k tokens · the whole tokenized gold category about $5.62bn. GLD's size wasn't checked — it's orders of magnitude larger, but there's no sourced figure here.

The key question: chain length

A useful frame is to count links of trust: issuer, custodian, transfer agent, chain, bridge, oracle, redemption rail, auditor, base-chain consensus. It doesn't hand you a verdict — it tells you what to count.

Chains are drawn from descriptions of the structures, not from each issuer's legal documents.

Findings

1. Freezes aren't theoretical — one has already happened

November 2022: Paxos froze 11,184.38 PAXG, about $19 million, on instructions from US federal law enforcement, across four addresses linked to FTX. The terms of use state plainly that Paxos may freeze any tokens, wherever they're held.

Both tokens — PAXG and XAUT — have admin keys that can pause transfers and blacklist addresses. Self-custody doesn't help: the wallet belongs to the holder, the off switch to the issuer.

That doesn't make them bad — any regulated issuer can do the same. But it answers "is this real gold?": a bar in a basement has no admin key.

2. The perpetual has already shown what it is

On 4 July 2026 gold on the Hyperliquid perp dropped about $100, below 4,090, and recovered within a minute. The cause: thin liquidity plus an oracle price plus leverage. A self-correcting price is no help to a trader liquidated inside those 60 seconds. In May 2026 a similar episode on another contract liquidated $1.51 million across 1,393 positions.

The gold market there is deployed via HIP-3: a third-party team lists it, picks the oracle and is responsible for liquidity, with 500k HYPE staked as collateral. Validators can slash it if it harms the network, but that's compensation after the fact, not protection during.

Lighter: no confirmation that it lists gold at all. That's "unconfirmed", not "no".

3. Redemption — where the differences get stark

In practice: a retail GLD holder will never get metal, only dollars at market. A PAXG holder will, if they accumulate a full bar and go collect it. An XAUT holder will, if they aren't a US person and are ready for Switzerland.

4. Supervision and audit — PAXG is ahead here

5. Liquidity — and why it lies precisely in a crisis

In calm markets PAXG trades tight: spreads often under 50 cents on a ~3,300 price (below 0.015%), with monthly turnover around 65% of supply. But that liquidity sits on centralized exchanges, which is a structural flaw: when things get scary, exchanges protect themselves, and depth vanishes exactly when it's needed.

GLD trades on an exchange with market hours — a minus for an overnight panic, a plus for an orderly exit: there are trading halts, market makers with obligations and a creation-redemption mechanism that keeps price close to NAV.

The core point: different insurance for different crises

"Which one is real gold?" has no answer, because "real" depends on what breaks.

Neither form is strictly better — ETF and tokens carry different sets of risks. Tokens suit someone whose main worry is a broken brokerage rail. For someone who simply wants gold exposure, an ETF offers mature infrastructure, while tokens add young infrastructure plus an issuer admin key.

Yield on tokenized gold in DeFi

A different frame. Here yield isn't swamped by price moves the way it is on ETH: gold moves 15–20% a year, so 5–8 points of yield equal 27–56% of its annual range (versus 2.5 points on ETH being just 3–5%). Against holding an ETF the gap is even starker: the fund charges about 0.40% a year and pays nothing, so the spread is 5.4–8.4 points.

The real numbers: there's essentially no yield market for tokenized gold

A spread from 0.00% to 12% for the same quantity is a finding, not noise. Three primary sources give roughly zero; only review articles give double digits, and none of them match what the protocols themselves show.

So where do 5–8% come from?

Three possibilities, none of them passive:

• A narrow range. On Uniswap v3 capital only works inside a set band. Turning the pool-wide 0.77% into 5–8% requires liquidity 6.5–10.4x denser than average. A ±1% range gives ~100x concentration versus full range, ±1.5% roughly 67x — so those figures are arithmetically explained by a tight band. The price is active management: once price leaves the band the position flips 100% into one of the two tokens and stops earning; every re-centre costs gas and a tax event.
• Token incentives. They end when the programme ends.
• Leverage. Collateral plus a stablecoin loan — a different risk, not yield on gold.

The APR shown in a pool's UI is usually extrapolated from the last 24 hours; with a couple of hundred trades a day and lumpy volume it swings by multiples. An annual figure computed from one day isn't an annual figure.

What an LP in a gold/gold pair is actually selling

A pair of two tokenized golds looks like a pair with no divergence risk: both legs track the same metal, so impermanent loss is near zero. But it shows up exactly when one of the two issuers breaks. An LP by construction sells what's rising and buys what's falling: if Tether runs into trouble, the LP automatically ends up mostly in XAUT.

There's a recorded precedent: in March 2023, when USDC fell to 0.87, the Curve pool went lopsided and LPs were left holding the asset that dropped. A gold/gold pool is a short position on Paxos–Tether divergence that pays a few percent a year.

Where the yield comes from: swap demand between one gold and another is naturally small — few people need to swap PAXG for XAUT. Fees are thin, so most of the yield is likely token incentives. If more than half the yield comes from emissions, it's a subsidy, not income; the industry benchmark is grim — real fees average about 14.5% of emissions. When the subsidy ends, what's left in the pool is the short divergence position, now unpaid.

On small pools separately: the gold/gold pool on Curve holds about $831k. If token rewards are fixed, every new deposit into a pool that size dilutes everyone's yield pro rata.

A simpler alternative

Lending PAXG on a major money market: 2–6% APY. Less than pool headlines, but three things disappear at once: issuer divergence risk, reliance on token incentives and the tiny-pool problem. PAXG stays PAXG, exit is usually instant, and liquidity is orders of magnitude deeper — the token is integrated into 100+ protocols. The yield gap is about 2–4 points; that's the price of removing three risks.

Other routes, in order of rising risk

1. Gold basis trade

Buy the metal as a token and short the gold perp. Funding on the Hyperliquid gold contract: 0.0006% per hour, 7-day average 0.0008%, i.e. 5.3–7.0% annualized. Open interest $166.5m, daily volume $26.5m — roughly eighteen times deeper than all tokenized gold pools combined.

Pros versus a pool: similar yield on a deeper market, no issuer divergence risk, no range rebalancing. Cons: separate margin needed, funding can flip sign, and there's the risk of a profitable position being force-closed in a panic — on an oracle price that already spiked $100 on 4 July. Crucially, the trade removes gold exposure entirely: it's yield, not gold.

2. Tighter range plus automated rebalancing

Going from ±1% to ±0.5% doubles concentration — from 100x to 200x. Yield roughly doubles, and so does the frequency of exiting the band. That can't be managed by hand; automated range managers exist. No vaults dedicated to a gold/gold pair were found — "unconfirmed", not "none". It adds an intermediary's smart contract and someone else's strategy.

3. Levered gold isn't a yield strategy

The loop from review articles: post gold tokens as collateral, borrow stablecoins, buy more gold, repeat — ending at 1.5–2x the metal. The arithmetic says no: supply yield on gold in lending is zero (see the Aave reserve above), stablecoin borrowing costs about 4–6% a year. 2x leverage earns zero minus ~5% in borrow cost, plus a 6% liquidation penalty and a 75% liquidation threshold. It's a leveraged bet on the gold price with negative carry, and calling it yield is a category error.

Comparison

What would break it, and by how much

The 2% and 1% thresholds are working assumptions, not an industry standard.

Forecasts (first check 31.12.2026):

• By 31.12.2026 there will be at least one episode where gold on the Hyperliquid perp diverges from spot by more than 1% for longer than 30 seconds. The previous such episode: 4 July 2026, about $100.
• By 31.12.2026 PAXG will not lose its spot peg by more than 2% on a daily close. A forecast in the instrument's favour: if it holds, the wrapper has gone a year without divergence.

What we don't know

No instrument for it. Whether Lighter lists gold — unconfirmed either way. Who deployed the gold market on Hyperliquid and which oracle they chose — the most important unknown for that instrument, since the oracle produced the 4 July spike. Funding on the gold perp as a holding cost wasn't measured — over a long horizon it could eat the whole idea. GLD's dollar size wasn't checked. XAUT tonnage (16.2 t) is as of December 2025, nothing fresher available.

Nobody knows. How PAXG redemption behaves if thousands of holders show up at once — the mechanism is documented, untested at scale. Whether tokenized gold survives a crisis in which crypto exchanges themselves break: the category is young — about $5.6bn — and hasn't been through a full cycle.

Summary: who each wrapper suits

Ranked by "realness" in a crisis, strongest first: physical metal in hand → GLD (for a crypto crisis) → PAXG (for a brokerage crisis) → XAUT (same, weaker links) → the perpetual (not gold at all, a bet on price).

PAXG beats XAUT on three measurable counts: monthly audit vs quarterly, OCC supervision vs issuer transparency, redemption for any holder vs a 50-oz minimum with Swiss-only delivery and no US persons.

If the goal is yield rather than storing value: the idea makes sense, but not through a gold/gold pool that pays the LP to be short divergence between two issuers. Lending PAXG at 2–6% does the job more simply. The perpetual is for a short-term bet on price, not for storing value: a third-party oracle, a $100 divergence episode already on record, and in a real panic a profitable position can be force-closed — exactly when the hedge is needed.

Research for information only. Not investment advice.