Ten questions from the RWA Summit, answered with ten queries

Earlier this month I spent two days at the RWA Summit in Brooklyn and sat through more than twenty panels. The conversations were good. But a pattern kept repeating: a question would come up, everyone on stage would agree it mattered, and nobody had a number for it. Ten of those questions stuck with me on the train home.

So I did what our customers do. I opened Snowflake, pointed it at the SonarX data shares, and wrote SQL. Nothing clever. Each query below is a dozen lines or fewer, and each one returned an answer in seconds. Most figures come from our Ethereum and Hyperliquid datasets; where a question needed more than one chain, I ran the same query on Solana, Avalanche, Base, BNB Chain, Arbitrum, Optimism and Polygon as well, and the text says so. Everything was queried between 11 and 14 September 2026. Where a number needs a caveat, it gets one.

1. “Nobody trusts published stablecoin numbers”

The complaint, repeated on at least three panels: headline stablecoin volume mixes payments with trading and protocol recycling, some things called stablecoins are closer to funds in a wrapper, and no published figure separates them.

The separation is already in the data. Our stablecoin fact table records, per token per day, supply, holders, mints, burns, transfer value and transfer count. The figures below are Ethereum only. Turnover and average transfer size fall out of a GROUP BY.

SELECT symbol,
       ROUND(AVG(total_supply) / 1e9, 2)         AS supply_bn,
       ROUND(SUM(value_transferred)
             / NULLIF(AVG(total_supply), 0), 1)  AS turnover_30d,
       SUM(transfer_count)                       AS transfers_30d,
       ROUND(SUM(value_transferred)
             / NULLIF(SUM(transfer_count), 0))   AS avg_transfer_usd
FROM   ██████████████████████████████
WHERE  block_date > DATEADD(day, -30, CURRENT_DATE)
GROUP  BY symbol
ORDER  BY supply_bn DESC;

Ethereum, 30 days to 13 September 2026, six of the fifteen tokens with more than $500M of supply:

Three very different economic activities sit under the word “stablecoin”. USDT: 22 million transfers averaging $17k, turning the supply over four times a month. That looks like money moving. DAI: turnover of 299 times supply, average transfer $1.6M. That is protocol plumbing and flash-loan recycling, and it inflates any volume figure it is added to. And USPD: $1.52bn of supply, 114 holders, zero transfers in thirty days. It is in every market-cap total, and it does not move.

None of this needed counterparty labels. Turnover, count and average size do most of the work. Labels sharpen it further.

2. “A thousand different Nvidias”

Someone on stage worried aloud that tokenized equities would produce a thousand tokens all claiming to be Nvidia. It was meant as hyperbole.

SELECT symbol,
       COUNT(*) AS contracts
FROM   ██████████████████████████████
WHERE  is_erc20
  AND  symbol IN ('NVDA', 'TSLA', 'AAPL', 'BUIDL')
GROUP  BY symbol;   -- run once per chain

Run on five EVM chains (Ethereum, Base, BNB Chain, Arbitrum, Polygon) and summed: 1,025 ERC-20 contracts carry the symbol NVDA. 1,357 carry TSLA, 826 carry AAPL. 1,412 carry BUIDL. On Ethereum alone, 271 tokens have “BlackRock” somewhere in their name.

Widen the lens and the scale is clear: of 1.57 million ERC-20 contracts on Ethereum, 83.75% share their symbol with at least one other contract. A large share of that is LP-token boilerplate, which is harmless. The branded symbols above are not.

The symbol is a string anyone can type. What identifies the instrument is the contract address, the deployer, the creation transaction and the proxy pattern, and all four are columns in the same table.

3. “Perps on real-world assets are taking over the venues”

A claim heard more than once: most of what trades on the leading perpetual venues is now stocks, indices and commodities rather than crypto. Nobody offered a measurement.

Hyperliquid publishes a funding event per account per market every hour. Notional open interest is the funding amount divided by the funding rate. Builder-deployed markets, which carry a prefix in their name, are mostly equities, indices, commodities, FX and pre-IPO names.

WITH f AS (
  SELECT coin, user,
         ABS(funding_amount / NULLIF(funding_rate, 0)) AS notional
  FROM   ██████████████████████████████
  WHERE  datetime >= '2026-09-14 11:00'
    AND  datetime <  '2026-09-14 12:00')
SELECT CASE WHEN coin LIKE '%:%' THEN 'RWA / builder perp'
            ELSE 'Crypto perp' END  AS bucket,
       COUNT(DISTINCT coin)         AS markets,
       COUNT(DISTINCT user)         AS accounts,
       ROUND(SUM(notional) / 1e6)   AS oi_usd_m
FROM   f GROUP BY bucket;

At 11:00 UTC on 14 September 2026: 139 builder-deployed markets, 43,354 accounts, $4.14bn of open interest, 28.5% of the venue. Crypto perps: 178 markets, $10.41bn, 71.5%.

Among the largest RWA markets by open interest: S&P 500 $423M (5,651 accounts), SK Hynix $382M, gold $339M, Brent $267M, WTI $233M. SpaceX pre-IPO exposure: $162.5M across 4,842 accounts, and in the 13 September position snapshot, 5,027 open positions of which 73.5% are long.

Substantial. Not a majority. And measurable to the hour.

4. “Fixing stablecoins at par liquidated the wrong people”

The October 2025 flash crash came up in the pricing panel. The question underneath it: what did the stablecoins actually trade at that night?

SELECT token_id,
       MIN(low)                 AS min_low,
       MIN_BY(price_date, low)  AS at_time
FROM   ██████████████████████████████
WHERE  token_id IN ('ethena-usde', 'usd-coin', 'tether',
                    'dai', 'binance-peg-busd')
  AND  price_date BETWEEN '2025-10-10' AND '2025-10-12'
GROUP  BY token_id;

On 10 October 2025 at 21:30 UTC, USDe printed a 30-minute low of 0.9572, 428 basis points below par. Binance-peg BUSD reached 0.8969. DAI dipped 42 bps. USDC’s low was 0.9996 and USDT never left 1.0000. Those are 30-minute aggregate bars, so single-venue prints went lower: on Binance, which marked collateral off its own order book, USDe was widely reported near $0.65.

Our price history carries stablecoins at the price they traded, in 30-minute bars, with each bar flagged if the source was stale. A margin engine’s choice to use par is a design decision. What the market was actually paying that night is now a lookup.

5. “Has this counterparty ever been frozen by an issuer?”

The compliance panels wanted screening, provenance and audit trails as a precondition for touching public chains. One concrete piece of that is issuer enforcement: which addresses have Tether, Circle, PayPal or Paxos ever frozen, and when.

SELECT event_type,
       COUNT(*)                        AS events,
       COUNT(DISTINCT account_address) AS addresses,
       MIN(datetime)                   AS first_event,
       MAX(datetime)                   AS last_event
FROM   ██████████████████████████████
GROUP  BY event_type;

Ethereum: 8,899 freeze events on 5,452 addresses across ten stablecoins, from 28 November 2017 to 11 September 2026, plus 582 unfreezes. As of 13 September the frozen stock is $857M of USDT across 1,406 addresses and $120.6M of USDC across 146, with smaller balances frozen on PYUSD, BUSD, XSGD, USDP, TRYB and USDA.

Almost nine years of enforcement history, both directions, as rows, each with a transaction hash.

6. “Tokenized AUM should come with a utilization figure”

Several speakers made the same point from different angles: tokenized funds are being counted, but not examined. Who holds them, how concentrated is it, how much has actually been redeemed?

The cleanest way to answer is to replay the transfers. Every mint is a transfer from the zero address, every redemption a transfer to it, and the difference per address is the holding.

WITH flows AS (
  SELECT to_address AS addr, source_value / 1e6 AS amt
  FROM   ██████████████████████████████
  WHERE  token_address = '████████████████████████████████████████'   -- BUIDL
  UNION ALL
  SELECT from_address, -source_value / 1e6
  FROM   ██████████████████████████████
  WHERE  token_address = '████████████████████████████████████████')
SELECT addr, SUM(amt) AS balance FROM flows
WHERE  addr <> '0x0000000000000000000000000000000000000000'   -- mints and burns
GROUP  BY addr HAVING SUM(amt) >= 1
ORDER  BY balance DESC;

BlackRock’s BUIDL (the USD Institutional Digital Liquidity Fund) is usually quoted as a single AUM figure. On chain it is a set of contracts spread across networks, and the same query runs against each of them. Seven contracts on six networks, as of 11 to 14 September 2026:

Roughly $2.4bn across 85 holding addresses, and in six of the seven contracts a single address holds more than half. The original Ethereum BUIDL contract, the only one with a broad holder base, has minted $2.26bn and redeemed $2.05bn since March 2024; its outstanding balance is the residue of a great deal of movement. Concentration, redemption history and holder count per network all come out of the same query, and the cross-chain total comes from running it seven times.

One more thing the table search turned up along the way: on Avalanche there are 13 contracts with “Institutional Digital Liquidity” in their name and on Polygon 10. One on each is real. Section 2 again.

7. “Where is tokenized collateral actually being used?”

Lending, several panels argued, is where the value in tokenization will accrue. The follow-up question went unanswered: which tokenized assets are pledged where, and what is borrowed against them.

Aave Horizon, Aave’s institutional RWA market, is one of the lending protocols we index event by event.

SELECT reserve_symbol,
       COUNT(*)                        AS borrows,
       ROUND(SUM(usd_value) / 1e6, 1)  AS borrowed_usd_m,
       COUNT(DISTINCT user_address)    AS borrowers
FROM   ██████████████████████████████
WHERE  protocol = 'aave_horizon'
GROUP  BY reserve_symbol;

Since 1 January 2026: $428M borrowed, split RLUSD $246M (31 borrowers), GHO $128M (20), USDC $53M (25). On the collateral side, the deposit table shows $131.5M of USTB (the Invesco Short Duration US Government Securities Fund, tokenized by Superstate) from 13 depositors, $61.8M of JAAA (the Janus Henderson Anemoy AAA CLO fund, issued via Centrifuge) from 11, and $10.6M of VBILL (VanEck’s tokenized Treasury fund) from 3, alongside $563M of RLUSD from 199 depositors.

Which asset, which address, what amount, what time. Utilization is a division.

8. “Has the curator changed the risk parameters?”

The vault panel proposed a simple test for on-chain credit vaults: has the manager initiated a risk change, and is there time to leave before it takes effect. With thousands of markets on one protocol alone, nobody is checking by hand.

SELECT datetime,
       protocol,
       config_type,
       reserve_address,
       old_value,
       new_value,
       transaction_from
FROM   ██████████████████████████████
ORDER  BY datetime DESC
LIMIT  5;

The two most recent rows, both 7 September 2026 at 17:53:23 UTC: an Aave USDC supply cap raised to 3.0bn (was 2.5bn) and the borrow cap to 2.7bn (was 2.25bn), in one transaction, by one address. Year to date on Ethereum: 273 parameter changes across supply caps, borrow caps, collateral factors and reserve settings, made by 13 distinct addresses.

For the comparability problem, our Morpho registry currently lists 6,225 markets across 3,922 distinct collateral tokens. Before, after, who, when. An alert is a WHERE clause on this table.

9. “Who held this instrument at the record date?”

A dividend that takes twenty days to reach tokenholders. A proxy campaign that could not find them. Both stories reduce to the same requirement: the exact holder list at a given date, reproducibly.

The replay query from section 6 does this with one extra condition, WHERE datetime < :record_date. For the original Ethereum BUIDL contract, the one with the broadest holder base:

Each row is a full address-by-address register, accurate to the block, that anyone can recompute and get the same answer. A distribution waterfall is arithmetic on that table.

10. “When does the day end on a market that never closes?”

The operations panel asked how to mark a quarter-end, and how to reconcile positions, when one side of the book settles at 4pm in New York and the other never stops.

SELECT COUNT(*)                            AS transfers,
       ROUND(SUM(source_value) / 1e15, 1)  AS usdc_bn
FROM   ██████████████████████████████
WHERE  token_address = '████████████████████████████████████████'   -- USDC
  AND  datetime >= '2026-06-30 20:00'
  AND  datetime <  '2026-07-01 00:00';

Between the New York close and midnight UTC on 30 June 2026, $9.0bn of USDC moved on Ethereum in 98,296 transfers from 26,830 sending addresses. On Saturday 4 July, over the Independence Day weekend, $27.0bn of USDC and $9.0bn of USDT changed hands.

Our balance history is timestamped to the block, so “position at 20:00:00 UTC on 30 June” is a legitimate question with one answer, and the on-chain side of a hybrid book can be cut at whatever moment the off-chain side uses.

What the ten have in common

Every query above is short. The questions are hard; the tables make them look easy, because the people who designed them had already been asked these questions by banks, funds and compliance teams, and shaped the data so the answer would be a GROUP BY rather than a project.

That is the part of SonarX that does not show up in a schema. Customers get the broadest indexed history in the industry, 130+ chains from genesis to tip, delivered into their own warehouse. They also get the engineers who built it, sitting with their analysts, turning a question heard at a conference into a result set the same afternoon.

If one of these ten is your question, or the eleventh one nobody put on a panel, send it to us. We will answer it with a query, and we will show you the query.

All figures from SonarX Snowflake data shares, queried 11 to 14 September 2026. Sections 1, 5, 7, 8, 9 and 10 use Ethereum data; section 4 uses our stablecoin price history; section 3 uses Hyperliquid; section 2 sums Ethereum, Base, BNB Chain, Arbitrum and Polygon; section 6 covers Ethereum, Solana, Avalanche, Optimism, Arbitrum and Polygon, and its total covers those six networks. Stablecoin price history is 30-minute bar data. Open-interest figures are derived from hourly funding events and reflect a single tick. Holder counts include addresses holding at least one unit.