Upstart Holdings, Inc. UPST

Key data

The business

Upstart is a lending marketplace, not a bank. Its machine-learning models score borrowers on more than a credit report captures, then it originates unsecured personal loans, and increasingly auto and home-equity loans, that other people's money funds: banks, credit unions, and capital-market buyers. Upstart collects fees for finding the borrower, underwriting the loan, and servicing it. The pitch is that its models approve more people at lower loss rates than a FICO cutoff, so a partner books more good loans for the same risk appetite. The catch, and the reason the stock has been a rollercoaster, is that when funding dries up, as it did in 2022 when rates spiked, Upstart cannot originate no matter how sharp the model is, and revenue collapses. It has also started keeping some loans itself, through warehouse lines and risk-retention facilities, so a slice of credit risk now sits on its own balance sheet rather than someone else's.

The numbers

The story is a recovery off a deep hole, and the recovery is real but jagged. Revenue has climbed every quarter for a year, and the last quarter grew about 42% against the year before.

The whole bet is visible in those quarters. Q1 2026 was ugly, a loss and an adjusted-earnings miss so wide that the market cut the stock in half. Then Q2 came back to a record top line and a profit. Sequential revenue accelerated to about 18% quarter over quarter, so the specific worry that growth was slipping toward low single digits did not materialize; if anything the growth leg strengthened. What the pack cannot confirm is the credit leg, whether charge-offs on the loans Upstart now carries are turning up, because origination volume, take rate and loss detail live in the earnings release, not this filing pull.

The five-year picture is why 15 times forward earnings is not obviously cheap. Revenue nearly halved from 2021 to 2023 and the company burned through a quarter-billion of losses before clawing back to profit in 2025. First-half 2026 revenue is up about 43% over the same span last year, matching the two-year pace off the trough. Whether that compounds depends entirely on operating leverage holding, and the third table is where the doubt lives.

An 8.5% operating margin in one quarter, a loss two quarters later on higher revenue: that is the signature of a business whose costs are steadier than its funding-driven revenue. The variant here is simple. The market prices this at 15 times next year because it treats Q3 2025 as the fluke and Q1 2026 as the truth. If the reverse is right, that the Q1 relapse was a funding air-pocket and 8% margins are the trajectory, earnings power sits well above consensus. The print that settles it is two straight GAAP-profitable quarters with double-digit sequential revenue, nothing else.

Management

The insiders voted with real money, and they bought. CEO Dave Girouard put $4.99M of his own into the stock in May 2026, right after the miss knocked it down, and co-founder Paul Gu added $3.14M last November. Against that, 42 insider sales totaling $7.1M, with plan status not disclosed in the filings, look like the usual option-and-grant housekeeping, and Girouard himself sold $2.26M last September. Net, insiders bought $10.3M and sold $7.1M. The company also repurchased roughly $100M of stock in Q1 2026, spending real cash into weakness, while stock-based comp still runs near 11% of revenue, the ongoing tax on shareholders that a young tech-lender pays.

How it fails or surprises you

Funding freezes again. Q1 2026 is the live example: revenue kept rising yet the company swung to a loss and missed adjusted expectations by a mile, because the appetite of loan buyers, not the model, sets the ceiling. If rates back up or capital markets pull in over the next year, originations and take rate compress together and the recovery dies.

The profitability whipsaw repeats. An 8.5% operating margin became a loss two quarters later on more revenue. If the answer is structural rather than a one-off, then the profit is not repeatable and 45 times trailing earnings is the honest multiple, not 15 times forward.

Rate cuts unlock the model (right tail). If cuts land and bank and credit-union funding returns at scale, revenue re-accelerates off proven operating leverage, and consensus already sketches $2.4B of revenue and $2.95 of earnings by 2028, which would put today's price under 10 times a number nobody yet believes. The market will not pay for it until the funding actually shows up, because it was burned so badly in 2022.

Closing thoughts

The next two earnings reports settle most of it. Sequential revenue and charge-offs on retained loans convert the doubt into fact, because the question is whether Q1 2026 was a funding air-pocket or the real trajectory reasserting itself. The tails are both fat and both real: the left tail is a genuine 2022 repeat, when a funding winter took this stock from over $400 to about $12, a permanent-looking loss that the balance-sheet credit risk now makes slightly worse than last time; the right tail is a full rate-cut cycle that re-rates a proven-profitable Upstart hard. On the evidence, the founders buying into the drop and the sequential acceleration tilt the odds rightward, but that is judgment, and the left tail is severe enough that survivability, not upside, is what you check first.

The bet is still that Upstart's computers judge who will repay a loan better than a credit score does, and that banks keep paying Upstart to do the judging, and that as interest rates ease the investors and banks who fund these loans come back in force. What breaks it is funding, and the pair that tells you first is sequential revenue and the retained-loan loss rate. Two more quarters like Q1 2026 and the recovery was a head-fake; two more like Q2 and 15 times earnings will look like a gift.