HOOD and the agentic stack.
A retail broker, a national bank, and a crypto exchange walked into the same week — and finally agreed on what comes after the app.
For ten years, fintech meant putting a brokerage and a bank inside a phone. The next phase puts a machine inside the brokerage and the bank — and rebuilds the rails underneath in programmable dollars. The week of May 27, 2026, three companies declared which layer of that stack they intend to own.
What "agentic finance" actually means.
Three definitions to clear before the rest of the document can do any work.
Agentic finance is what happens when an LLM stops being a chat window and starts being a counterparty. The agent reads your portfolio, decides what to do, places the order, settles the cash, and reports back. Three pieces have to be true at once for this to work: the agent has to be allowed to act, it has to have money it can move, and the money it moves has to settle somewhere a regulator will recognize. Each of those is a different business.
The rails an AI uses to take an action against a real account. Robinhood's Agentic Trading product, Coinbase's Payments MCP. The "give the agent a steering wheel" layer.
Money the agent can hold, transfer, and reconcile without waiting on bank hours. Stablecoins live here — USDC, SoFiUSD, JPMD. The "give the agent a wallet" layer.
Where digital assets become economically usable in the traditional system — to borrow against, post as margin, or pledge for a down payment. The "make the agent's wallet matter offline" layer.
The plumbing that lets any LLM connect to any service. Robinhood publishes one. Coinbase published Payments MCP in September 2025. The Internet got HTTP; agentic finance gets MCP.
If you only remember one thing: agents do not have credit scores, social security numbers, or bank accounts. They cannot pass KYC. So for an agent to participate in finance at all, a human has to lend it a sandbox — a ring-fenced account, a virtual card, a wallet with rules — and the rails underneath have to settle fast enough that the agent's decision-making doesn't outrun its money. That is the entire problem.
May 27, 2026.
Two product launches on the same Wednesday, plus the September 2025 launch they both inherit.
It is unusual for two listed financial companies to ship the same idea on the same day. On May 27, both Robinhood and SoFi did exactly that, from completely different starting positions. Robinhood opened its brokerage to AI agents; SoFi opened its bank to programmable dollars. Coinbase had already shipped the third piece nine months earlier and quietly become the connective tissue.
What this week clarifies, in retrospect, is that the agentic finance stack already exists across these three names — it just hadn't been packaged in one sentence yet. Robinhood owns the interface: the place the agent takes its action. SoFi owns the settlement: the programmable bank dollar the agent moves. Coinbase owns the custody: the asset the agent ultimately holds and, increasingly, the asset that gets pledged back into the real economy. Each of the three has now also stepped one layer above its starting position — HOOD is buying crypto exchanges, SOFI is operating a public-chain stablecoin, COIN built the agent payment protocol. The lanes are overlapping, but they are not yet the same lane.
Why crypto-backed mortgages matter more than they look.
A small program with an outsized signal.
The Fannie Mae crypto-mortgage program looks tiny on the page. A dual-loan structure: a standard 15- or 30-year conforming mortgage from Better, plus a separate loan secured by your bitcoin or USDC to fund the down payment. You keep your crypto exposure. You avoid the capital gains tax that would have hit if you'd sold to put cash down. If you fall 60 days delinquent, the crypto gets liquidated to cover the deficiency, the same window any Fannie loan applies.
The structure is conservative. The signal is not. The Federal Housing Finance Agency, under Bill Pulte, issued the directive in June 2025 ordering Fannie and Freddie to treat crypto assets as reserves in risk underwriting. That is the regulatory inflection point: bitcoin is now a financial asset the GSEs will look at the same way they look at brokerage statements and 401(k) balances. Roughly 20% of American adults hold digital assets. Many of them are asset-rich and cash-poor. The 2025 Redfin survey said over 10% of millennial and Gen Z homebuyers had sold crypto to fund a down payment. This product cancels that forced sale.
The parable of the second key.
Imagine a friend who owns a paid-off motorcycle and wants to buy a house. The motorcycle is worth $40,000. He can sell it for the down payment, which means losing the bike. Or — if his banker will accept it — he can pledge it as collateral on a small loan that funds the down payment instead. He gets the house. He keeps the bike. The bank holds a second key.
That is what the Fannie Mae program does for crypto. The asset stays in the borrower's wallet, still appreciating or correcting, still on the holder's balance sheet. A separate loan against it provides the cash. The home gets a normal Fannie-conforming mortgage. The second key is the only innovation — but it's the one that lets a whole asset class start paying for housing.
The current stress test is the price chart. Bitcoin opened today at $66,667 — down roughly 47% from its October 2025 all-time high near $126,200, the level around which the very first of these mortgages were originated. The mortgage program is now living through exactly the collateral-volatility scenario the four Democratic senators warned the FHFA about in 2025. If the conservative haircuts and 60-day delinquency window hold up through this drawdown without forced-liquidation cascades, the program graduates from pilot to template. If they don't, the experiment ends quietly and gets cited in every future regulatory hearing.
For the equity stack: the program is most directly accretive to Coinbase, which is the custody-and-USDC anchor of the loan structure and gets a fee leg every time a homebuyer pledges. It is indirectly bullish for the entire on-chain dollar ecosystem (and therefore SoFiUSD), because it normalizes the asset class. Robinhood is, for now, a spectator — but the Robinhood Chain currently under development is the project most likely to bring HOOD into this lane directly.
Three tickers, three layers.
Prices and fundamentals at the June 3, 2026 close. Each name handled as its own card.
| Ticker | Layer | Price | TTM EPS | P/E (TTM) | Market cap | Verdict |
|---|---|---|---|---|---|---|
| HOOD | Interface | $87.12 | $1.80 | 46.6× | $79.4B | Pure-play, missing the rail |
| COIN | Custody + payments | $165.19 | $2.63 | 66.1× | $45.8B | Embedded, BTC-correlated |
| SOFI | Settlement (bank) | $16.63 | $0.48* | ~35× | $18.5B | Quietly owning two layers |
Does the agentic launch matter, or is HOOD just selling the same flywheel with a new wrapper?
Q1 2026 revenue grew 15% YoY to $1.07B — down from 27% in Q4 and 100% in Q3 2025 — and crypto transaction revenue fell 47% YoY to $134M. The market saw the decel and cut the stock in half from its $153 October high. Then on May 27 it shipped Agentic Trading and the Agentic Credit Card and rallied 28% in eight sessions on what was, fundamentally, a developer announcement. That is the tell: the multiple was already pricing fatigue, and even a credible new growth narrative bought roughly $20B of market cap back.
The product is real. A ring-fenced sub-account you fund with a fixed dollar amount, an MCP server that lets Claude or ChatGPT take actions against it, real-time trade feed, a kill switch, push notifications on every order. Equities only at launch; options, crypto, futures, event contracts to follow. The Agentic Credit Card extends the same primitive to spending — a virtual card with daily caps, optional manual approval, and the Gold Card's 3% cashback on agent-driven purchases. Combine this with the PDT rule expiring June 1 and HOOD now structurally re-engages two cohorts it had lost: small accounts boxed out of active trading, and the AI-native users who never wanted to click "buy" themselves.
The structural risk is the layer below. HOOD doesn't issue a stablecoin. Robinhood Chain is "under development" but not shipped. Today the agent's cash settles in fiat through standard brokerage rails — which is fine for equities but limits the agent's reach into 24/7, on-chain, or cross-border use cases. The WonderFi close gives them a regulated Canadian crypto base and Bitstamp gives them Europe; the Open Transaction Layer support gives them an onchain coordination posture. But until there is a HOOD-native dollar instrument the agent can actually hold, Robinhood is renting the rail SoFi just minted.
For the existing position: the thesis got better, not worse. The May rally repriced sentiment; the fundamentals (Gold subs +36% YoY to 4.3M, 27.6M funded accounts, $18B Q1 net deposits at a 22% annualized rate, Trump Accounts going live July 4) still compound. The trade-management question is whether to size up into the next leg or wait for the next deceleration print. The valuation is full at 46× TTM and 42× forward, but the growth bands have widened, not narrowed.
If agentic finance is the future, why is COIN down 61% from its July 2025 high?
Because the stock is two businesses welded together and the market is paying for the wrong one. Roughly 60% of revenue still comes from trading fees, which scale with bitcoin price and retail attention. Bitcoin opened today at $66,665, down 47% from the October 2025 all-time high of $126,200, and COIN's chart has tracked that drawdown almost tick-for-tick. The trading book is in a cyclical trough; Q1 2026 reportedly printed a net loss of $394M on $1.41B in revenue.
The other Coinbase — the infrastructure Coinbase — is exactly where the agentic stack lives. Payments MCP (open-sourced September 2025) is the de facto agent-wallet primitive. The x402 protocol sits underneath it as the HTTP-native payment standard; by end of 2025 it had processed 100M+ payments at ~$600M annualized volume, with Solana taking the majority of that flow. USDC custody and stablecoin reserves continue to earn float income that does not depend on trading volume. And the Better/Fannie Mae mortgage rail launched March 26 puts Coinbase inside the underwriting flow for every conforming GSE-backed crypto mortgage written in the U.S.
The integration with Kalshi for U.S. perpetual futures and the investment in ProShares' GENIUS Money Market ETF (designed for stablecoin reserves) extend the same theme: COIN is becoming the place where every part of the on-chain dollar ecosystem clears. The infrastructure pieces don't show up in the trading-revenue line, which is why the market keeps marking them at zero whenever bitcoin sells off.
The valuation is the hardest to defend mechanically — 66× TTM and 57× forward against $6.56B TTM revenue and a 12% net margin. But this is the same kind of misfit between business and ruler we worked through on INTC. Coinbase trading is cyclical; Coinbase infrastructure is recurring. The right way to value the second is on revenue and float, not the cycle-trough earnings of the first. The market is saying: prove the second business is real by the next cycle peak.
Why is the stock down 35% YTD while the fundamentals just inflected?
SoFi is the name in this trio whose price action is most disconnected from its product trajectory. Q1 2026 revenue hit $1.10B (+5% versus consensus) and EPS of $0.12 met expectations. Net income jumped 134% year over year to $167M. Loan originations set an all-time record at $12.18B. Membership grew 35% to 14.7M, with deposits at $40B. The stock closed today at $16.63, down 35% from its December $26 print and well off the $32 fifty-two-week high.
What the market missed on May 27 was the structural significance of the SoFiUSD launch. This is not a third-party stablecoin partnership. SoFiUSD is issued by SoFi Bank, N.A. — an OCC-regulated national bank. That makes it the first stablecoin in the U.S. that comes with the supervisory regime of a chartered bank attached. Reserves are held 1:1 in cash and cash equivalents inside the bank itself. Ethereum and Solana at launch. Integrated into Galileo (SoFi's BaaS platform, 160M accounts) as a settlement option. Wired into Mastercard's network as a card-settlement currency under the March 2026 agreement.
Translate that into the agentic stack: when an agent on Robinhood needs to move dollars, the dollars it moves can be SoFiUSD. When a Mastercard transaction settles cross-border, it can settle in SoFiUSD. When a Galileo-powered neobank wants to offer a stablecoin-native checking account, the rails are already there. SoFi is selling picks-and-shovels into a market its competitors are only just identifying. The GENIUS Act limits SoFi from paying interest on the stablecoin itself, but tokenized deposits (which can pay interest and stay FDIC-eligible) are already on the roadmap. That is the chess move: a stablecoin for movement and a tokenized deposit for storage, both inside the same charter, with conversion between the two.
SoFi also shipped SoFi Coach, an AI financial-planning tool, on May 30. The agentic-finance posture is now bilateral: SoFi is both the place an agent does the work, and the place the dollars settle when it's done. The stock didn't get rewarded for any of this in the past two weeks, which is half the appeal. Anthony Noto's framework of 30% revenue CAGR and 38-42% EPS CAGR through 2028 is intact. The disconnect between price and trajectory is the entire opportunity. (Note on the EPS metric: TTM EPS is approximate; SoFi's GAAP earnings are still scaling rapidly off a low 2024 base.)
Three companies, three lanes, one stack. HOOD lets the agent act. SOFI gives it dollars to move. COIN holds the asset on the other side. Buy the lane that's mispriced, not the slogan that's loudest.
The six points that survive.
If you forget the document, keep this list.
- Agentic finance is a stack, not a feature. Interface, settlement, custody. Any product analysis that doesn't name a layer is hand-waving.
- MCP is the standard. Robinhood publishes one. Coinbase shipped Payments MCP in September 2025. The fight is over which app the agent connects to first, not whether the agent exists.
- The May 27 trifecta wasn't coincidence. HOOD shipping Agentic Trading the same day SOFI made SoFiUSD generally available is the market signaling that interface and settlement now scale together. The next product war is over the bundle.
- Crypto-backed mortgages are the bridge from the agentic stack to the real economy. Once Fannie Mae accepts an asset as reserves, that asset has joined the system. The current bitcoin drawdown is the program's first real stress test — and if it holds, the rail extends.
- HOOD is the agentic interface, but not yet the rail. The Agentic Credit Card and ring-fenced sub-account are real product. The missing piece is a HOOD-native settlement asset — currently delegated to whatever fiat or stablecoin the agent already holds.
- SOFI is the quietest beneficiary. A 35-percent YTD drawdown alongside a 134-percent jump in net income and the first stablecoin under a U.S. national-bank charter is the rare moment when the price and the product line genuinely disagree.
Remaining Thoughts.
The gaps I want to close before sizing or rotating.
- Q1 HOOD already runs at 46× TTM and 42× forward, and just rallied 28% on a product launch with no measurable revenue yet. How do I separate the "agentic finance is a real second growth wave" thesis from the "AI-themed re-rating that fades when the next deceleration prints"? Specifically — what would I need to see in Q2 (reported late July) to keep adding versus trim?
- Q2 The structural gap in HOOD's stack is the settlement layer. Robinhood Chain is "under development" and the company is supporting the Open Transaction Layer protocol. Is the right way to play this gap to assume HOOD eventually ships its own dollar instrument — or to assume they partner with SoFiUSD or USDC and stay an interface-only business? The choice has very different valuation consequences.
- Q3 COIN at 66× TTM is the hardest valuation to defend on the surface, but the infrastructure book (Payments MCP, x402, USDC float, GSE mortgage rail) is exactly the part that doesn't depend on bitcoin price. Is there a clean way to value Coinbase as two businesses — trading on P/S at a cycle-aware multiple, infrastructure on float-times-take-rate — and underwrite the sum of the parts? This feels like the INTC ruler-selection problem with a different ticker.
- Q4 SoFi looks like the most asymmetric setup in the trio — 134% net income growth, a structural stablecoin moat, and a 35% YTD drawdown all in the same six months. What is the bear case I'm missing? Specifically: how much of the stablecoin opportunity is competitive (JPMorgan's JPMD, Tether at 58% share, USDC) versus how much is genuinely defensible under the GENIUS Act's bank-charter requirement?
- Q5 The Fannie Mae crypto-mortgage program is being live-tested by a 47% BTC drawdown right now. What's the right way to monitor whether this works? Watching delinquency rates inside the program is too slow; watching BTC price is too noisy. Is there a single proxy — Better's volumes, Coinbase's institutional custody balances, FHFA testimony — that would tell me ahead of time whether the rail extends to a broader rollout or quietly dies?
