Set the shelves against the flow they are chasing. Corporate transactions were R$497 billion of the R$505.5 billion Receita Federal counted last year, 98.3% of the total [2], which leaves individual investors moving roughly R$8.5 billion between them [2], and that R$8.5 billion is the entire pool the retail menus at Itau, Nubank and Banco do Brasil compete for. Banco do Brasil, which opened direct bitcoin and ether buying in January, has moved more than R$11 million through the service [5]: about 0.13% of the retail pool [4], or one forty-six-thousandth of the market as a whole [1]. The reais matter more than the dollar equivalents here, since the conversion in the underlying data implies about 5.1 reais to the dollar [5] and tells you nothing else.
Retail is the thin slice and also the fast one, compounding near 40% a year since 2020 [3], which is a good reason to build distribution and a weak one to buy inventory. Carlos Akira Sato of Syscapital, who told Folha that clearer rules left conservative banks "more secure to launch their products" [9], puts proprietary exposure at own-money purchases that absorb price, liquidity and credit risk, and says no Brazilian bank has crossed that line [10]. The shelves therefore consume compliance capital and no market-risk capital, and neither publisher reports a fee, a spread or a revenue figure for any of them, so the cost of this business can be described and its value cannot.
Banco Safra took the one position with a balance sheet in it, issuing its own dollar-pegged token, Safra Dolar, in September 2025, keeping full custody in-house and marketing it to high net worth clients as dollar exposure without an offshore account [11]. An issuer owes redemption where a broker owes execution, and neither source says what backs the token. Resolution 521, which treats any purchase or exchange of a dollar-pegged token as a foreign exchange operation held to the same reporting bar as sending money abroad [8], drops that product into plumbing the banks already staff, under a framework cryptobriefing dates as effective from February 2026 [13].
The evidence supports at least three endings. The menus stay a retention feature that never earns enough to break out; or the FX-classified stablecoin business becomes the actual product and the retail tokens were a licence application in disguise; or client volume grows until holding no inventory costs the banks the spread, and one of them starts warehousing. My read is the second, with the corporate share of that R$497 billion as the prize and the retail menu as the shop window. It fails if a filing after March 2026 carries a nonzero virtual asset line at any of these banks, or if one of them puts crypto brokerage revenue into group results at a size worth reading, in which case the shelf was the business all along.
