Updated November 7, 2026: This article has been substantially rewritten to clarify the interpretation and limitations of Cybiont’s core earnings and equity measures. The numerical results, calculation methodology and position disclosure are unchanged. Technical detail has been moved to an appendix.

Upstart reported a profit in the September quarter. Our core measure showed another loss. The gap matters because the same forecasts that help produce reported earnings also support the equity shareholders would want behind them in a downturn.

Originations reached $4.1 billion, up 44% from a year earlier. GAAP net income was $36.6 million. Cybiont core earnings were negative $58.7 million. Reported tangible common equity was $812.9 million; set aside the projection-dependent balances and the private equity stake described below, removing similarly measured liabilities too, and Cybiont core tangible common equity was negative $19.8 million.²⁸

That is not a claim that Upstart is insolvent, that its retained interests are worthless, or that GAAP profits are fictitious. Core equity and core earnings are our own measures, not GAAP measures, liquidation values or a cash-flow statement. They ask how much equity and earnings remain without recognizing the value of selected future payment streams before those streams pay.

There is a good reason not to dismiss those streams. Upstart puts money into retained interests as its risk-sharing business grows. Charging that investment before it pays makes our earnings measure look worse during expansion. A shareholder who expects the assets to deliver can reasonably prefer the reported answer.

But that preference is a bet on the forecasts. We want to know what supports it.

Selling the loan does not finish the investment

The attraction of the marketplace is easy to understand. Banks, credit unions and institutional buyers supply funding. Upstart earns fees. Contribution profit reached $218.0 million in the September quarter. A fee business with growing volume deserves a different assessment from a lender that has to finance and absorb the risk of every loan it originates.

Upstart is not simply the latter. It is not simply the former either.

Under some arrangements, selling a loan leaves Upstart with a Beneficial Interest Asset (BIA), a claim whose payments depend on how the sold loans perform, sometimes accompanied by a Beneficial Interest Liability (BIL). Securitizations leave notes and residual certificates behind. Servicing loans for others produces servicing assets and liabilities. These balances depend on management’s estimates of defaults, prepayments, duration and discount rates, rather than observable market prices.³⁴⁵

Retaining risk can help keep funding partners involved. It also means that part of the sale remains an investment for Upstart’s shareholders. More originations can produce more fees and more retained claims at the same time. The fees and the claims should not be mistaken for the same thing.

What is behind the equity?

Ordinary tangible common equity (TCE) removes goodwill and intangible assets from common equity. We extend that question to the other balances a shareholder cannot readily draw on without relying on management’s projections. We call the remainder core TCE. The deductions are deliberately symmetrical: if we set aside an estimated asset, we also remove the corresponding estimated liability.

The largest deduction is net beneficial interests: $613.7 million of BIA less $20.6 million of BIL, or $593.1 million. That is about 73% of reported TCE by itself. These are projected payments from loans already sold, not payments already collected.³

Net servicing rights account for another $60.8 million. The $68.0 million asset and $7.2 million liability value future servicing revenue and expenses; borrower defaults and early repayments can shorten the stream.⁴ Notes receivable and residual certificates contribute $144.6 million. They receive what remains after senior securitization investors are paid, so modest changes in loan performance can matter disproportionately to them.⁵

We also deduct $41.0 million of non-marketable equity securities. These are private stakes without readily determinable fair values, about which Upstart discloses little. They are not a source of equity that can quickly be sold to absorb losses.⁶

Finally, we add back $6.8 million of trailing-fee liabilities. These estimate future payments to lending partners from borrower receipts. Removing them is the liability-side counterpart of removing estimated assets, not a claim that the obligation disappears.⁷

Table 1: Reported TCE to Cybiont core TCE at September 30, 2026 ($ thousands)

Source: UPST Form 10-Q for the period ended September 30, 2026, balance sheet and Notes 2, 3 and 5; Cybiont Capital Research (netting and core TCE computation).

Chart 1: Reported TCE to Cybiont core TCE, September 30, 2026 ($ millions)

Source: UPST Form 10-Q for the period ended September 30, 2026, balance sheet and Notes 2, 3 and 5; Cybiont Capital Research.

The resulting core TCE is negative $19.8 million, or −2.4% of reported TCE and −2.2% of common equity. The assets deducted total 103% of reported TCE before the trailing-fee add-back. Chart 2 shows the scale.

Chart 2: Reported TCE against the assets Cybiont deducts, September 30, 2026 ($ millions)

Source: UPST Form 10-Q for the period ended September 30, 2026, balance sheet and Notes 3 and 5; Cybiont Capital Research (net balances).

Core TCE was $305.0 million at the end of 2024 and $162.9 million at the end of 2025. It turned negative in June 2026. This is not a proposed dollar-for-dollar write-down. It shows how little of the reported cushion remains once these estimates are set aside.

The distinction matters most when the cushion is needed. Upstart’s own sensitivity table says a 10% adverse change in expected credit risk rates reduces BIA fair value by $114.8 million; a 20% change reduces it by $226.9 million. Those are about 14% and 28% of reported TCE. The table changes one input at a time. In a downturn, credit, prepayments and required discount rates can move together.¹

Allen Puwalski’s November 5 note asked how a bank supervisor would treat these marks. For the shareholder, the concern is more immediate: the assumptions supporting the cushion can weaken alongside the business it is meant to protect.

The earnings that arrive before the payments

The balance-sheet treatment has an earnings counterpart. We calculate core earnings from the sequential change in core TCE, excluding signed capital movements and the acquisition and initial-investment effects specified in the appendix. Stock-based compensation remains a cost; we remove the matching credit to equity, not the expense.⁸

This removes the creation and revaluation of the deducted future-payment streams. Receipts from existing beneficial interests, residuals and servicing rights receive credit as they increase core TCE. A new asset does not receive that credit merely because a model assigns it a value. Favorable and unfavorable changes in those marks are both excluded.

This is an equity-based earnings measure, not cash earnings in the literal sense. It does not strip every accrual from the accounts, nor forecast what a stopped-origination business would earn after its costs and collections changed. It isolates a particular dependence on projected future payments.

In the September quarter the gap between GAAP net income and core earnings was $95.3 million. Net beneficial interests increased $57.3 million, residual notes $24.2 million and net servicing $14.4 million. Those net balance-sheet increases account for the reconciliation gap. Table 4 in the appendix supplies the capital-flow bridge.

There is more than one way to describe this recovery. Upstart’s contribution profit deducts borrower acquisition and verification and servicing costs from net fee revenue, but leaves out engineering and general and administrative costs, interest income and fair-value adjustments. Adjusted earnings before interest, taxes, depreciation and amortization (EBITDA) makes further exclusions from GAAP earnings, including stock-based compensation. Neither answers our question about the retained claims.⁹

Chart 3: Management measures, GAAP net income and Cybiont core earnings, 1Q 2021 to 3Q 2026 (quarterly, $ millions)

Source: UPST quarterly earnings press releases, 1Q 2021 to 3Q 2026 (contribution profit, adjusted EBITDA, net income); UPST Forms 10-Q and 10-K for the same periods (balance sheets and statements of stockholders’ equity); Cybiont Capital Research (core earnings).

Table 2: Management measures, GAAP net income and Cybiont core earnings by year, 2021 to 9M 2026 ($ thousands)

Source: UPST quarterly and annual earnings press releases, 2021 to 3Q 2026; UPST Forms 10-Q and 10-K for the same periods; Cybiont Capital Research (core earnings; annual figures are sums of quarters).

Across the 23 quarters from the start of 2021, cumulative contribution profit was $2.66 billion and adjusted EBITDA was $698 million. GAAP earnings were a loss of $242 million. Core earnings were a loss of $1.0 billion.

The chronology matters. Core earnings were positive from 1Q 2021 through 1Q 2022, then negative in all 18 quarters from 2Q 2022 through 3Q 2026. Beneficial interests first appeared in March 2023. The earliest losses cannot be blamed on that book.

When buyers pulled back and rates rose in 2022 and 2023, core earnings reached negative $126.4 million in 1Q 2023 while contribution profit was still positive $67.6 million. GAAP and core earnings moved together more closely in that period, when marks were being cut rather than built.

Since 2024 they have separated. Over the eleven quarters through September 2026, GAAP losses totalled $28.5 million and adjusted EBITDA was positive $447.8 million. Core losses totalled $742.6 million. Stronger originations have not yet produced positive earnings on our basis.

Yes, this measure charges growth before it pays

That objection is not an inconvenience to be tucked into a footnote. It is central to interpreting the result.

Part of the negative core earnings since 2024 represents cash committed to new beneficial interests and residuals. A profitable investment can depress this measure when made and improve it when it pays. The cumulative core loss is therefore not proof that an equal amount of economic value has been destroyed. Accept the assets’ booked values and those investments remain assets, which is the GAAP treatment.

Warren Buffett put the other side of the growth argument plainly: “In the case of a low-return business requiring incremental funds, growth hurts the investor.”¹⁰ He was distinguishing growth from value creation, not saying that investment is itself a loss. Our deduction does not establish the return on each retained interest. It makes the question of that return harder to avoid.

The fair question for us, then, is whether the assets are performing as their valuations imply. Recognition under GAAP is not an assurance that every estimate will be realized. Nor does the use of Level 3 estimates establish that they are wrong. We need evidence beyond the label.

What the marks have delivered against their own schedule

A discounted asset should generally accrete as its expected payments draw nearer, if the underlying expectations hold. We estimate that accretion from opening BIA and the average beginning and ending weighted-average discount rates, then compare it with the disclosed BIA fair-value earnings line. We count any BIL fair-value increase as adverse, following the November 5 analysis. The full calculation is in Table 6 and the appendix.

On that basis, nine of the twelve complete quarters from 4Q 2023 through 3Q 2026 show an adverse residual. The cumulative residual is $110.7 million. In the September quarter, expected BIA accretion of $16.9 million compared with actual fair-value earnings of $6.1 million; adding the $13.2 million BIL increase produces a $24.0 million adverse residual. June’s was $23.1 million.

There are favorable quarters too: 1Q 2025, 4Q 2025 and 1Q 2026. They reduce the cumulative adverse result. Leaving them out would turn a diagnostic into an argument assembled to win.

The test has a substantial limitation. Upstart does not disclose a bridge separating accretion from credit, discount-rate and prepayment effects. Our calculation assumes accretion passes through the fair-value earnings line. It also lacks precise intra-quarter timing of additions and is affected by settlements and transfers. An adverse residual is an implied miss under those assumptions, not a separately disclosed credit loss or estimate revision, and not evidence of intentional misstatement.

It is not a second haircut to equity either. Prior fair-value changes already passed through earnings. The $110.7 million is a flow diagnostic, not another deduction from today’s book value.

Even with those limits, the pattern gives us reason not to accept the growth explanation as the whole answer. BIA grew from $176.8 million at the end of 2024 to $613.7 million in September 2026, about 3.5 times. Weighted-average discount rates fell from the mid-13% range in late 2023 to about 12.4%. A lower discount rate supports a higher present value for otherwise unchanged payments, while lowering the percentage rate of expected accretion. Growth in the book nonetheless increased the dollar accretion implied each quarter.

Credit risk rate spreads were positive in most quarters since early 2024, indicating expected credit losses above initial underwriting assumptions; the weighted average was 1.05% in both June and September 2026. That is consistent with pressure on the marks. We do not add it to the residual calculation.

The record does not settle the ultimate value of the retained book. It does weaken the case for treating its growth as sufficient evidence of recovery.

The loans we cannot test

The same calculation cannot be applied reliably to loans carried at fair value: the public rollforwards do not separate fair-value marks from charge-offs and borrower interest.

Those loans originate from the same engine, but they are not the same portfolio as the sold loans. Upstart says 51% of the loans on its balance sheet in September were held for research and development, testing newer products and borrower segments and helping price discovery. They have less performance history than the loans investors buy. The shortfalls on retained interests in sold loans do not measure losses on this different book. They do make us less comfortable assuming that its less-tested marks are more dependable.

We make no further deduction for this in core TCE. The disclosure does not support one.

A recovery with a dependent cushion

Upstart’s origination recovery is real. So are the claims left behind when some of those loans are sold. Their eventual receipts matter to shareholders, and dismissing them as worthless would be as unhelpful as treating the forecasts as cash in hand.

Our conclusion is narrower, and sufficient. Reported tangible equity depends on estimated future payments; the equity remaining after our adjustments is negative. Core earnings have stayed negative through both the buyers’ retreat and the recent expansion. The accretion comparison, while imperfect, offers little comfort that the retained book is consistently delivering what its opening marks imply.

That combination deserves a higher assessment of risk than a clean fee marketplace. When funding and activity pull back, fee income is exposed just as the assumptions supporting retained assets may weaken. The reported equity cushion and earnings recovery are not independent protections against that event.

The next filing will put those expectations to work again. We will compare the beneficial-interest rollforward with the roughly $17–19 million of quarterly accretion implied by the current book and a discount rate near 12.4%, alongside credit risk rate spreads, core TCE and the GAAP-to-core earnings gap. Subscribers receive the quarterly reconciliation in the Upstart Fair-Value Rollforward Monitor.

Disclosure: Cybiont Capital holds no position in UPST. Robert Lacoursière has no personal position in UPST. This report is for informational purposes only and is not investment advice or a recommendation to buy or sell any security.

Technical appendix

Equity and capital reconciliation

Core TCE = reported TCE − net loan servicing rights − notes receivable and residual certificates − net BIA/BIL − non-marketable equity securities + fair-value trailing-fee liabilities. Reported TCE itself is common equity less goodwill and intangible assets. No additional deductions for the loans discussed above enter this calculation.

Table 3: Reported tangible common equity at September 30, 2026 ($ thousands)

Source: UPST Form 10-Q for the period ended September 30, 2026, condensed consolidated balance sheet; Cybiont Capital Research.

Core earnings equal the change in core TCE less signed equity effects from capital movements. Equity inflows are subtracted; equity outflows are reversed. The capital reconciliation includes share issuance and option exercises, stock-based compensation credited to equity, repurchases, capped-call purchases and dividends (Upstart pays none). The check using unadjusted common equity is change in common equity less signed capital movements = GAAP net income.

Acquisitions and the initial purchase of the non-marketable stake in 3Q 2021 are excluded from core earnings because they exchanged cash for assets this walk deducts. Later small changes in the stake stay in core earnings. These are the treatments in the published series, not new adjustments.⁸

Table 4: Cybiont core earnings for 3Q 2026 ($ thousands)

Source: UPST Forms 10-Q for the periods ended June 30 and September 30, 2026, balance sheets and statement of stockholders’ equity; Cybiont Capital Research (core TCE and core earnings).

Table 5: Management measures, GAAP net income and Cybiont core earnings, 4Q 2024 to 3Q 2026 ($ thousands)

Source: UPST quarterly earnings press releases, 4Q 2024 to 3Q 2026 (contribution profit, adjusted EBITDA, net income); UPST Forms 10-Q and 10-K for the same periods; Cybiont Capital Research (core earnings and core TCE).

Accretion calculation and coverage

Expected BIA accretion ≈ opening BIA × average of beginning and ending weighted-average discount rates ÷ 4. Asset residual = expected accretion − disclosed changes in fair value recorded in earnings. Total adverse residual = asset residual + BIL fair-value change, with a liability increase adverse and a decline favorable. Positive totals are adverse under this test.

Coverage is 4Q 2023 through 3Q 2026. The second and third quarters of 2023 have asset marks but incomplete liability rollforwards and are excluded from the total. The calculation does not isolate credit, rates or prepayments and cannot resolve the timing of additions, acquisitions, settlements and transfers. Favorable quarters offset adverse ones. Nothing in the cumulative total is an automatic incremental equity loss.

Table 6: Twelve-quarter beneficial-interest accretion scoreboard, 4Q 2023 through 3Q 2026 ($ thousands, except discount rates)

Source: UPST Forms 10-Q and 10-K for periods ended December 31, 2023 through September 30, 2026, beneficial-interest Level 3 rollforwards and weighted-average discount rates (fair value measurement note); Cybiont Capital Research (expected accretion and residual calculations).

Notes

Sources

1. Allen Puwalski, Nov 5, 2026 — https://cybiontcapital.com/2026/11/05/upstart-bank-capital-ssfa/

2. Allen Puwalski & Robert Lacoursière, Nov 2, 2026 — https://cybiontcapital.com/2026/11/02/upstarts-long-road-to-becoming-a-bank/

3. UPST Form 10-Q for the period ended September 30, 2026 (accession 0001647639-26-000084) — balance sheet; statement of stockholders’ equity; Notes 2, 3 and 5 (fair value hierarchy, Level 3 inputs, sensitivities and rollforwards); MD&A (balance sheet loan composition); Item 3

4. UPST Forms 10-Q and 10-K, 1Q 2021 through 2Q 2026 — balance sheets, statements of stockholders’ equity, historical beneficial-interest rollforwards and discount rates

5. UPST quarterly earnings press releases, 1Q 2021 through 3Q 2026 — contribution profit, adjusted EBITDA and net income

6. Cybiont Capital Research — core TCE walk, core earnings, expected accretion and residual calculations