SoFi Technologies: real growth, but returns on capital still have to be earned

SoFi's growth is real. In Q2 2026, members grew 35% and GAAP revenue 43%, and management kept its FY2026 earnings guidance despite a much higher tax assumption. The harder question is whether that growth turns into durable returns on tangible common equity above the cost of equity, without repeated external capital or understated credit costs.

At $17.16 on September 22, 2026, the shares trade 15.6% above my present Base value of $14.84. The capital-based primary model values SoFi below the market, while the independent earnings model lands almost exactly on the price. SoFi is neither an obviously broken business nor an obvious bargain, and a Confidence C valuation needs a large discount before new money makes sense.

Route: SF-02 Banks, digital bank holding company with lending, financial services and technology platform operations. Economic classification: Quality Growth (7/12), not a demonstrated compounder. Primary method: growth-adjusted justified P/TBV through a staged 10-year residual income model. Secondary method: revenue times normalized net margin times forward P/E. All values are present values per economic share, not future price targets.

Why SoFi is valued as a bank

SoFi operates a regulated bank inside a financial holding structure, funds interest-earning assets with deposits and bears material consumer credit exposure. Its fee and technology businesses can improve capital efficiency, but they do not make consolidated SoFi a subscription software company.

The hard economic measures are common earnings after normal taxes, credit costs and compensation, tangible common equity (TCE), return on tangible common equity (ROTCE), funding quality and regulatory capital. Corporate free cash flow, EV/EBITDA, EV/revenue, SaaS multiples and the company's Rule of 40 presentation are not used. Deposits are operating funding, not debt to deduct, and consolidated cash is not automatically distributable.

Q2 adjusted EBITDA of $357.8 million is not owner-distributable bank income. H1 operating cash flow of minus $6.2 billion reflects held-for-sale loan funding, not a software-style cash burn. Neither number anchors the valuation.

Current financial performance

Current adjusted ROTCE of about 7.1% sits below the 11.5% Base cost of equity. Management's 20% to 30% long-term ROTCE objective is an aspiration, not present earning power, so the valuation must explain a transition from returns below the equity hurdle to returns above it.

Taxes matter. FY2025 pretax income of $525.9 million was taxed at only 8.5%. Re-taxed at the original 26% assumption, net income would have been $389.1 million rather than $481.3 million. For FY2026, keeping $825 million of guidance while raising the tax assumption from about 15% to 22% is an underlying pretax improvement: management put the EPS at about $0.65 on the old tax rate. Reaching the full-year guide requires about $497.9 million of H2 adjusted net income.

A changing business

Fee businesses and the Loan Platform Business can grow revenue without keeping every loan on the balance sheet, which is potentially capital-efficient growth. At the same time, retained lending is expanding fast: personal loans are about 58.9% of unpaid principal in the fair-value loan portfolio. The 2025 capital raises increased capacity to support that growth, which makes proof of returns on the enlarged capital base essential.

Technology Platform revenue fell 23% year over year after the loss of a large client, despite a sequential recovery. Segment contribution margins also declined: Financial Services to 46% from 52%, Technology Platform to 14% from 30%. A high consolidated growth rate does not prove every segment has a durable moat.

On September 22, 2026, SoFi announced it had gone live with stablecoin settlement across Mastercard's global network. It is a genuine product milestone, but payment volume is not SoFi revenue and no separate stablecoin premium is added to the Base value.

Capital, shares and debt

Capital is strong but being consumed by growth. During H1, holding company risk-weighted assets rose from $37.2 billion to $48.7 billion, faster than CET1 capital. Bank CET1 fell from 16.4% to 15.2% and holding company CET1 from 22.8% to 18.7%. At an analyst-selected 12% bank CET1 management floor, not a legal minimum, headroom is about $1,477.8 million, and a further 10% rise in risk-weighted assets would use about $560.1 million of it. That is capital capacity, not cash available for dividends.

The model front-loads the existing award pool and includes the option cash. Economic shares are 1,354.9 million in Bear, 1,375 million in Base and 1,405 million in Bull, with the extra Base and Bull shares reserving for net share settlement of convertibles after capped calls. No cash proceeds are assumed for that reserve. Future compensation is charged as a recurring cost, not also counted as annual share issuance.

$428.0 million of 2026 convertible notes mature on October 15, 2026, with principal settled in cash and any conversion spread in shares (conversion price about $22.41). $862.5 million of 1.25% notes mature on March 15, 2029. Repaying principal reduces cash and liabilities together, so it is not an extra deduction from TCE. June consolidated cash of $3,126.2 million does not prove how much the parent company can access, because bank distribution restrictions apply. The 2025 equity raises are already inside June equity and are not added again.

Credit and the fair-value loan book

The personal loan portfolio is carried at fair value, so expected credit economics sit in valuation marks and related income. The Financial Services provision is not the total loss cost of the consolidated personal loan business. Fair-value marks, delinquent loan sales and realized charge-offs have to be read together, not added mechanically.

Three different loss measures exist: management's all-in personal loan charge-off estimate including delinquent loan sales of about 3.7%, the reported portfolio charge-off rate of 2.62%, and the 4.77% annual default assumption in the fair-value disclosure. None should be mislabeled as another.

The all-loan fair-value premium of $2,027.1 million equals about 21.4% of reported TCE. That is a valuation-sensitive exposure, not proof the asset is worthless. An extra 100 basis points of annual loss on $26.1 billion of personal loan principal costs about $195.8 million after tax, or $0.142 per Base share per year.

Primary valuation: residual income on tangible equity

A bank is worth its tangible common capital plus the present value of profits earned above the return required on that capital. SoFi is not valued as an immediately mature 20% to 30% ROTCE franchise. Each year, net income equals beginning economic TCE times normalized ROTCE, and residual income equals net income less the cost of equity on beginning capital. The horizon covers ten rolling years from September 2026 to September 2036.

$ million except EPS. ROTCE is on beginning economic capital, while the company's reported ROTCE uses average reported capital.

Base cost of equity is 11.5%, terminal growth 3.5% and terminal ROTCE 17%. The Year 1 profit of about $861.9 million is only modestly above the FY2026 guide, and the ramp is deliberately slower than a straight-line EPS extrapolation. Future returns assume roughly 25% normal taxation. All earnings are retained for five years, and later distribution capacity is a valuation mechanism, not a forecast of an announced dividend.

The terminal justified multiple is 1.69x TCE, below today's economic P/TBV of 2.46x, so the Base value does not rely on a higher terminal multiple than the market pays now. The primary Base value is $14.07 per share. About 75.7% of it comes from the discounted terminal value, a real weakness that lowers Confidence.

Secondary valuation: forward earnings

The cross-check starts from the $4,800 million midpoint of FY2026 adjusted revenue guidance as a run-rate proxy, grows it for three years, applies a normalized net margin after taxes, credit costs and compensation, divides by scenario shares and applies a forward P/E, then discounts three years at the scenario cost of equity.

The 18x Base multiple is an estimate for a profitable financial platform still capable of material per-share growth, not a peer median. This method can justify roughly today's price, which is important counterevidence to a simple "expensive bank" conclusion. It can also hide the capital needed to support its EPS, so it receives 25% weight against 75% for the capital-explicit model.

No synchronized peer panel was built, so peers carry no quantitative weight. Nu's Q2 2026 ROE of 33% shows that scaled digital banking can earn high returns, but different geography, products and accounting make it unsuitable as a direct multiple.

Historical cross-check

Today's price to tangible book is lower than at the last three year-ends. That is a legitimate favorable cross-check, but not proof of undervaluation: the 2025 equity raises changed the denominator, and three observations from a young bank configuration are too few to force a reversion to 3x to 4x book.

Valuation triangulation

The primary model receives 75% and the earnings model 25% in every scenario. A capital-regulated lender must first justify returns on the capital it needs.

Base divergence of 19.6% passes the framework's 20% test, just barely. Bear divergence of 27.3% sits in the warning band, because the Bear P/E capitalizes continuing profits more generously than the capital model does when returns stay below the equity hurdle.

What is the market already pricing in?

No reverse valuation identifies a unique market belief. Revenue is not the decisive disagreement, because the earnings model already accommodates strong growth. The real questions are how long it takes to earn attractive returns on the expanded capital base, how much capital must be retained, and what premium durable fee earnings deserve. A one point lower cost of equity alone lifts the blended value to about $17.15, almost the price. That limits conviction in an outright overvaluation claim.

Analyst price targets average $20.34 (median $19, range $12 to $30, 26 analysts). A 12-month target discounted one year at 11.5% is about $18.24. It is more favorable than the bank model, but without a transparent credit and capital bridge.

At $17.16, SoFi trades at 28.6x FY2026 guided adjusted EPS, a forward guidance multiple rather than a trailing GAAP P/E.

Bear, base and bull scenarios

Bear: lending and fee growth slow, credit normalization absorbs more revenue, and returns stay below the cost of equity despite continued profits. No rescue equity is assumed, and more severe regulatory or liquidity distress could produce values below the Bear case. Base: fee monetization and operating leverage improve, credit losses stay manageable and the enlarged capital base is deployed gradually, with returns rising above the cost of equity but short of management's full objective. Bull: fee mix and operating leverage support a high-return platform, with returns approaching the lower or middle part of management's ambition. There is no separate crypto or stablecoin value in any case.

The probability-weighted present value is $16.11, 6.1% below the price. The probabilities are subjective, not market-implied, and a weighted present value is not a future price.

Sensitivity

Only the secondary model gives explicit three-year prices: $6.96 in Bear, $23.75 in Base and $46.84 in Bull, or annualized price returns from $17.16 of -26.0%, 11.4% and 39.8%, before dividends, costs, taxes and currency effects. These are not the triangulated present values. Calibrated 12-month price touch probabilities are not available.

Where the stock becomes attractive

No industrial leverage surcharge is applied to bank deposits, and no extra divergence or accounting surcharge duplicates what Confidence already reflects. Applied to the $14.84 Base value, the maximum buy price is $8.16. At a 50% discount, the stronger threshold is $7.42, giving a purchase review band of $7.42 to $8.16. That means re-underwrite if the same fundamentals still apply, not that the company becomes safer automatically at that price.

At $17.16, the shares trade 110.3% above the maximum buy price and would need to fall about 52.4% to reach it, absent an upward revision. That is not a prediction that the stock will get there.

SoFi paid no common dividends in H1 2026, and no dividend is assumed. Income is not the basis of this thesis.

Reasons to own or follow SOFI

• Integrated distribution. The reported existing-member cross-buy share was 51%, supporting a credible distribution advantage. The test is higher profit per customer, not simply more counted products.
• Deposit funding. A predominantly deposit-funded platform can earn better economics than one reliant on warehouse facilities, if retention does not need uneconomic pricing.
• Capital-light fees. Loan platform, interchange, brokerage and service fees can raise revenue per unit of tangible capital.
• Execution. The 2025 revenue and EBITDA outperformance and the tax-adjusted 2026 guidance upgrade support management's delivery record.
• An identifiable Bull mechanism. If margins and capital efficiency approach management's long-term objectives, value can materially exceed today's price.

What could go wrong?

• Credit and true earnings. Rapid growth and delinquent loan sales can make a single loss ratio misleading, and fair-value assumptions must reconcile with realized credit. Three years of 100 basis points of extra personal loan losses cut the Base value to about $13.84.
• Capital and liquidity. Capital is healthy but being consumed, and parent-level cash access and the October 2026 note maturity matter separately from consolidated liquidity.
• Dilution and capital allocation. Further dilution, technology spending or acquisitions may add revenue without per-share returns. 5% more economic shares lower the Base value to $14.13.
• Regulation and execution. Consumer lending, deposit competition, third-party loan demand and newer payment activities face changing economic and compliance demands.

Management execution

Initial FY2025 guidance called for $3,200 to $3,275 million of adjusted revenue and $845 to $865 million of adjusted EBITDA. Actual results were $3,591.4 million and $1,053.9 million. GAAP net income of $481.3 million beat the $285 to $305 million guide, partly because of the low tax rate, but the tax-normalized comparison remains favorable.

The July and December 2025 equity offerings raised about $1.7 billion and $1.5 billion net at prices above book value. The added tangible capital is real, but the return on it still has to be earned. An earlier technology-related goodwill impairment and the later large client loss prevent an exceptional acquisition record.

Stock Analyza scorecard

Under the framework's bands, a Base value 13.5% below the price classifies the stock as fairly valued: above central fair value and above the purchase ceiling, with limited margin of safety.

Confidence score

Confidence measures the reliability of the valuation, not the chance that management hits its targets or that a trade is profitable.

What I would watch from here

The next review points are the October 15, 2026 note maturity and the Q3 2026 results.

The three most important thesis breakers

1. Full-loss credit economics that stay below the investment case.
2. Regulatory or holding company liquidity stress that forces damaging financing.
3. Repeated capital or share issuance without per-share returns above the cost of equity.

Adversarial review

The arithmetic review recomputed the capital and share bridges, every scenario, the terminal multiples, weights, reverse solutions, sensitivities, margin of safety and scores. A separate high-precision implementation revalued the distributions and terminal cash flows without reusing the residual income function. The clean re-audit passed 165 consistency checks and 23 separate precision checks.

The economic review prevented SoFi from being routed as a SaaS company or valued on industrial free cash flow, reconciled issued, diluted and adjusted share counts, added option exercise cash alongside option shares, avoided counting future compensation twice, replaced an unverifiable book value input with the reported tangible book series, recognized the tax-adjusted guidance upgrade, discounted each future horizon to today, separated capital ratios from parent liquidity and gave the stablecoin launch no value without incremental economics.

The strongest counter-thesis is that SoFi can earn 20% to 30% ROTCE, grow fee revenue with less capital and put the raised capital to productive use. The secondary Base already supports about $17.13 and the full Bull case $34.87. The no-new-money conclusion does not survive every plausible Bull outcome. It stands because the evidence does not yet make those outcomes the central case.

At the guidance midpoint with 25x earnings, today's price can be justified. At 18x, it cannot. The disagreement is about the multiple and the durability of returns, not about disbelieving management's earnings guide.

Holding other assumptions fixed, terminal growth of 2.5%, 3.5% and 4.5% gives primary values of $13.59, $14.07 and $14.69. The publication gate status is PASS_WITH_WARNING because future returns, credit seasoning, convertible settlement and terminal assumptions remain uncertain.

Final verdict: hold at $17.16, watchlist for new money

SoFi's growth and improving profitability deserve serious attention, but today's price does not provide the required buffer against uncertain long-run returns on capital, unsecured credit seasoning and dilution. The earnings model is close to the market price, so the evidence does not support treating SOFI as an obvious short or a mechanical sell.

Verdict: WATCHLIST for new money at $17.16, conditional HOLD for existing holders. Triangulated Base value $14.84, downside 13.5%, required margin of safety 45%, maximum buy price $8.16.

For new money, that means waiting for either a much larger discount or better demonstrated returns per share. For existing holders, continuing depends on the original thesis, position risk and the monitoring triggers above, not on the purchase price.

The main thesis breaker is growth that keeps requiring more equity while normalized, loss-inclusive returns stay below the cost of equity. A lower share price alone would not fix that.

Sources

Primary sources include SoFi's Q2 2026 earnings release, June 2026 Form 10-Q, Q2 2026 investor presentation, Q2 2026 earnings call transcript, FY2025 earnings release, FY2024 earnings release and the Mastercard stablecoin settlement announcement. The digital bank return comparison uses Nu Holdings' Q2 2026 results. Analyst targets, consensus estimates and historical year-end market capitalization are from a third-party market data aggregator and serve only as secondary cross-checks.

The reference market price is $17.16 at 3:26 pm EDT on September 22, 2026, regular session, not a closing price. Balance sheet and capital figures are as of June 30, 2026, and the share count as of July 31, 2026.