American Express

IThe business

American Express issues cards, lends to the people and companies that hold them, runs the network they spend on and signs up the merchants that accept it. Its 10-K says that direct relationship with both sides of a payment sets it apart from the bankcard networks.³ It is the fourth largest general-purpose card network by purchase volume, behind Visa, China UnionPay and Mastercard, and about 76,800 people work for it.³ The 8-K that carried its third-quarter results was accepted by EDGAR at 7:00 a.m. New York time in Q4 2026, before the market opened. The stock fell 3.7% that day, from $325.17 to a close of $313.22, and closed at $323.71 on 29 October.^1254546

In the twelve months to 30 September 2026, which I call the TTM, revenue net of interest expense was $77.6B, net income was $12.3B and diluted earnings per share were $17.75, or $17.02 without the Q3 gain and charge that I describe under accounting.C (calculated)³⁴⁵⁶ From FY2016 to FY2025 revenue grew 8.2% a year and earnings per share grew 13.1% a year from FY2016 earnings adjusted for one-off gains, or 11.9% as reported, helped by a 24% fall in the share count.C (calculated)³¹⁵¹⁷ FY2016 includes a $1,218M gain on the sale of the Costco and JetBlue portfolios and $410M of restructuring charges, and its revenue includes Costco-related revenue until June 2016; I adjust for the first two under Financials.¹⁵¹⁷

Four segments run the business. U.S. Consumer Services is 48% of segment revenue and earned a 19.6% pretax margin in FY2025. Commercial Services is 23% and earned 21.7%. International Card Services is 18% and earned 12.3%. Global Merchant and Network Services is 11% of revenue and earned 51.1%, which makes it 25% of segment pretax income; that margin depends on how the company prices discount revenue between segments.C (calculated)³

What a revenue dollar is made of

The merchant fee is still the largest line at 51% of revenue, down from 58% in FY2021. Card fees went from 12% to 14% of revenue and net interest income from 18% to 24%.³⁴⁶¹⁰

How it makes money

• The merchant fee. Discount revenue was $39.7B in the TTM.C (calculated) Discount revenue is 2.23% of billed business, down from 2.30% in FY2022, which the company ties to shifts in the geographic and merchant mix of spending. Billed business, the spending on cards Amex issues, was $454.0B in Q3 2026, up 8%.³⁴⁶¹⁰
• Card fees. Net card fees were $11.2B in the TTM, up 16% in Q3 (17% adjusted for currency).C (calculated) The average fee per card was $131 a year in the first nine months of 2026, $117 in FY2025 and $49 in FY2017. The consumer Platinum Card’s fee is $895, and Amex puts its annual value at over $3,500.³⁶¹⁴³³
• Interest. Net interest income was $18.8B, 24% of revenue.C (calculated) Card balances were $221.1B at 30 September 2026, of which $63.1B are pay-in-full balances that earn no interest. The net interest yield was 8.5% in Q3.⁴⁶
• Service fees and network partnerships. $7.8B, 10% of revenue.C (calculated)³⁴⁶

How it is funded

Customer deposits of $155.0B fund most of the card book, up from $53.0B at the end of FY2016, and long-term debt was $58.5B.⁴⁶¹⁷ A parallel rise of 100 basis points in rates would cut annual net interest income by $210M, or 1.1% of the TTM figure, which is small beside the credit cycle.C (calculated)⁴ Interest expense is deducted from revenue throughout this page, as the company does.

What $100 of spending earns

Revenue per $100 rose from $4.10 in FY2016 because fees and interest grew faster than spending; the cost of rewards and services rose from $1.03.C (calculated)³⁴⁶¹⁵

Amex earns $4.35 on every $100 its members spend and pays back $1.55 of it in rewards and benefits.

IIThe moat

I think the moat is the brand, which lets Amex charge $895 for the Platinum Card.³³ The cost of keeping it has risen as fast as the fee.

The moat is a premium brand on a network where the spender is the customer. In the usual terms it is an intangible-asset moat, a brand that supports a price, with a network effect that comes from the spending of high-spending members. Card count matters less. The brand is what lets Amex charge for the card: 76% of new accounts in Q3 2026 were on fee-paying products, the average fee per card rose from $49 in FY2017 to $131 in the first nine months of 2026, and net card fees grew from $2.9B in FY2016 to $11.2B in the TTM, 3.9 times.C (calculated)³⁶⁷¹⁴ The network is what makes merchants accept it: Amex’s payments volume per card is about 4.3 times Visa’s, a figure that mostly reflects the mix of cards.C (calculated)²³

Do members stay?

Amex does not publish card renewal or attrition rates. Management said in the Q3 release that it saw “strength in billings, new acquisitions and Card Member retention”.⁵ The filings give a proxy: cards-in-force at the start of a period plus new cards acquired, less cards-in-force at the end, is the number of cards that left, for any reason. It was 9.5M cards in FY2025, 11.4% of the cards in force at the start of the year, against 10.1% in FY2022 and 12.0% in FY2024. In the first nine months of 2026 it was 8.3M, or 12.8% at an annual rate. Cards sold or acquired with a portfolio, such as the two small business cobrand portfolios sold in 2026, would move the figure.C (calculated)^3491011

Fewer new cards are being acquired. Proprietary new cards acquired were 3.1M in Q3 2026 against 3.2M a year earlier, and 9.2M in the nine months against 9.6M; U.S. Consumer Services new cards fell from 1.5M to 1.3M a quarter while cards-in-force grew 2%.⁴⁷ The next generation is already in the base. Millennials and Gen Z are 38% of U.S. Consumer Services billed business, and they supplied about half of its 9 points of growth in Q3 2026.C (calculated) They were 65% of global consumer new accounts.⁷

What the fee costs to keep

The fee shows what the brand can charge and what the benefits cost. Net card fees grew 14.8% a year from FY2016 to FY2025. The 10-K describes Card Member services expense as the cost of the benefits and services members use, such as travel benefits and the new Platinum benefits, and it grew faster: from $1.1B in FY2016 to $7.9B in the TTM.³⁴ Card Member services expense equals 70% of net card fees in the TTM, up from 38% in FY2016. Amex does not allocate that expense to fees, because the same benefits support spending and balances, so I read the ratio as a measure of cost and mix. Points are a cash cost as well: about 96% of Membership Rewards points are ultimately redeemed.C (calculated)⁷

could continue to increase as we evolve our value propositions

The ratio peaked right after the Platinum refresh. Card Member services expense was 58% of net card fees in Q3 2025, 74% in Q4 2025 and 68% in Q3 2026, because fees are recognized over the membership year while the cost of the new benefits starts at launch. Rewards, services and business development took 44.7% of revenue in Q4 2025 and 44.3% in Q3 2026, so the step has been taken and the share has not risen since. Year-over-year growth in these costs laps the launch from the fourth quarter of 2026, so that quarter is the first clean read.C (calculated)⁶³³⁴⁴

Per card in force, net card fees rose $81 between FY2016 and the TTM and Card Member services expense rose $72. Per $100 of spending, revenue rose 25 cents and rewards plus Card Member services rose 52 cents, twice as much; salaries and other expense fell 32 cents.C (calculated)^3461517

Pretax profit per $100 of spending was $0.84 in FY2016 on adjusted earnings and $0.83 in the TTM.

On the merchant side the moat holds at a slightly lower price. Discount revenue was 2.30% of billed business in FY2022, 2.24% in FY2025 and 2.23% in the TTM. The 10-K attributes the decline to “lower average merchant discount rates due to shifts in geographic and merchant spend mix”. Seven basis points on $1,783B of billed business is $1.2B a year, 1.6% of revenue and 8% of adjusted pretax income.C (calculated)³⁴¹⁰ Amex card acceptance reached more than 190 million merchant locations in July 2026, including an estimated 80 million in China and Brazil, much of it through digital wallets.⁵

Returns show what the moat is worth. Return on risk-weighted assets, the cleaner measure, was 3.7 cents on the dollar in FY2016 on adjusted earnings, 4.2 cents in FY2025 and 4.4 cents in the TTM.C (calculated)³⁴¹⁷ Return on average equity was 33.5% in FY2018 and 33.9% in FY2025, on equity 50% larger, and 38% in the first nine months of 2026, about 36% without the Q3 gain net of the charge; return on average tangible common equity was 41% in FY2025 and 46% in the nine months.C (calculated)^3571415

IIIThe competitive advantage

American Express’s advantage over Visa and Mastercard is the spender, and over Capital One it is the credit record. Its disadvantage against all four is scale.

Visa processed 7.7 times Amex’s payments volume in calendar 2024 on 33 times as many cards, so each Amex card carried 4.3 times Visa’s payments volume and 4.7 times Mastercard’s; counting cash volume the multiples are 3.6 and 3.9.C (calculated)²³ The comparison mostly measures the mix of cards: Visa and Mastercard count debit, prepaid and emerging-market cards, and Amex’s figure includes cards that other banks issue on its network, which were $227.2B of processed volume in FY2025, 12% of its network volume.C (calculated)³²³ A like-for-like test on credit books is spending per dollar of balances. In Q3 2026 Amex’s billed business was $2.05 per dollar of Card balances; Capital One’s domestic card purchase volume was $0.96 per dollar of loans.C (calculated)⁴⁶²⁶²⁷

Amex’s Q3 net write-off rate of 1.8% is on principal only, the basis Capital One uses for its 3.98%: Capital One deducts charged-off finance charges and fees from revenue, and its supplement says so.²⁶²⁷ Amex’s balances include $63.1B of pay-in-full charge balances, 29% of the total, which lower its rate. Even if those balances wrote off nothing, the rate on the other 71% would be about 2.5%, below JPMorgan’s 2.91% and Capital One’s 3.98%. On loans alone, with pay-in-full receivables apart, Amex’s FY2025 rate was 2.2%, the same as in FY2019.C (calculated)^3462527 Capital One now owns Discover and Diners Club International. Its domestic card purchase volume was $250.0B in Q3 against $454.0B of Amex billed business, and its $260.3B of card loans exceed Amex’s $221.1B of card balances.C (calculated)²⁶²⁷

Where American Express leads

The first lead is the customer: a card that carries four times the volume gives a merchant a reason to accept it. The second is credit, in the table above. The third is the premium product: 76% of new accounts in Q3 2026 were on fee-paying products.⁷

Where it trails

Amex is far smaller than Visa and Mastercard, and it has to pay for the product it sells: variable customer engagement costs, which are rewards, Card Member services and business development, grew 16% in Q3 2026 against revenue growth of 9% and were 44.3% of revenue, a growth rate that still includes the Platinum step.⁵⁷ The 10-K says competitors “may offer richer value propositions”, and it lists agentic commerce, stablecoins, real-time payments and wallets, among others, as paths that could weaken its customer relationships.³ Its reply so far is a Resy integration with OpenAI and a framework called Intent Intelligence for agentic commerce, both announced in Q3 2026.⁵ Delta is another weakness. The Delta cobrand was about 13% of worldwide billed business and 21% of Card Member loans at the end of FY2025, and its agreement runs through 2029.³

IVManagement

Stephen Squeri has been chairman and chief executive since February 2018 and has worked at American Express for over 40 years. He is 67. Christophe Le Caillec has been chief financial officer since August 2023. Douglas Buckminster, the vice chairman, retired on 6 March 2026, and Anre Williams, who led enterprise services, left in February 2025.³⁸³⁵³⁸

Since 31 January 2018, the last month-end before Squeri became chief executive, AXP has returned 16.0% a year with dividends, 3.66 times, against 13.9% for the S&P 500, measured by SPY, and 9.7% for the Financial Select Sector SPDR.C (calculated)³¹⁴⁷ Diluted earnings per share rose from $7.91 in FY2018 to $15.38 in FY2025, 10.0% a year. Return on equity was 33.5% in FY2018 and 33.9% in FY2025 on an equity base 50% larger, and the pretax margin was 20.1% in FY2018 and 19.1% in the TTM.C (calculated)³¹⁴¹⁵

The board has 13 nominees, all independent except Squeri. John J. Brennan, a former chairman and chief executive of Vanguard, is the lead independent director. Three directors joined in 2025. Every nominee received a majority of the votes at the May 2026 meeting, and say-on-pay received 93.32%.^8343637

Ownership

Berkshire Hathaway held 151,610,700 shares at 30 June 2026, 22.1% of the class in the proxy and 22.6% of the 670.6M shares outstanding at 15 October, worth about $49.1B at $323.71. Its 13F for 30 June 2026 shows the same count as the 13F for 31 December 2025.C (calculated)⁴⁸²⁸⁴⁵

A 1995 agreement keeps Berkshire passive. Berkshire follows the board’s recommendations on voting up to 17% of the shares while Squeri is chief executive and it owns at least 5%, and it may not dispose of its shares to anyone who holds over 5% or seeks control.⁸

All 27 directors, nominees and executive officers together own 501,133 shares and hold rights to 448,594 more, under 1%.⁸ Squeri owns 223,990 shares, including 163,292 in a grantor retained annuity trust of which he is the trustee, about $72.5M at $323.71, or 1.6 years of his reported pay. The executive officers may not hedge, pledge or hold company stock on margin.C (calculated)⁸⁴⁵

Pay

Squeri’s 2025 pay was $46.2M on the summary compensation table, up from $37.2M in 2024 and $35.7M in 2023. The committee puts it at $48.1M on its own basis, 37% above target and 96% performance-based. His pay is 814 times the median employee’s. Compensation actually paid, which marks stock to market, was $95.4M for 2025 and $122.5M for 2024, and $100 invested at the end of 2020 was worth $326 at the end of 2025 against $203 for the S&P Financials Index.⁸

The plan pays on revenue growth, earnings per share and return on equity, which are sensible measures for this business. It names no measure of credit quality, of what the proposition costs, or of how much of each fee dollar reaches profit. Effective risk management sits inside a 20% strategic category, and a risk review can cut the award.⁸ Earnings rise when reserves are released and when loans grow, and both lift return on equity. In Q3 2026 provisions fell 23% on a reserve release, and adjusted pretax income rose 2% while adjusted earnings per share rose 13%.C (calculated)⁵⁶⁷⁸ Under Squeri, return on equity is unchanged on a larger equity base and earnings per share grew with the share count; the pretax margin slipped from 20.1% in FY2018 to 19.1%. Management has grown fee income and returned capital, and its plan names no credit or cost measure.C (calculated)³⁸¹⁵

Of Squeri’s $46.2M in 2025, $28.9M was stock and options, and the plan pays for revenue, earnings per share and return on equity, so a reserve release pays him too.C (calculated)⁸ He owns 223,990 shares, about $72.5M, and cannot hedge or pledge them.⁸⁴⁵ Over the cycle the company bought back $41.2B of stock at $132.88 a share, against $323.71 now. Since FY2025 it has paid $313 to $319, near my base value, where a buyback does little for a holder who stays.³⁴¹⁵¹⁷

Capital allocation

American Express buys its own stock far more than it buys companies, so the price paid is the main capital decision.

From FY2016 to FY2025 the program bought about 310M shares for $41.2B at an average of $132.88, against $323.71 today, 144% higher. Shares outstanding fell from 904M to 686M, 24%, and to 671M at 30 September 2026.C (calculated)^34151720 The fall of 218M shares is smaller than the 310M bought, because about 92M shares, 30% of the purchases, replaced stock issued to employees; net of issuance, the cost of shrinking the share count was $189 a share.C (calculated)^3151720 The recent record is different. The program paid $242.65 in FY2024, $312.87 in FY2025 and about $318.67 for 17.2M shares in the first nine months of 2026, 2% above today’s price and 2% under my base value of $324.C (calculated)⁴¹⁸¹⁹ Buybacks and dividends took 75% of net income to common over the three years to Q3 2026.⁷ The dividend declared per share rose from $1.22 in FY2016 to $3.28 in FY2025, 11.6% a year, and to $0.95 a quarter from March 2026, up from $0.82.C (calculated)³²¹

The policy is to hold CET1 between 10% and 11% and return the rest. CET1 was 10.9% at 30 September 2026, about $2.3B above the 10% floor and $0.4B short of the 11% top.C (calculated)⁴⁶ Deals are small and several have no disclosed price:

Capital spending rose from $2.4B in FY2025 to $3.7B in the TTM, and the new headquarters is a commitment of up to $2.8B.³⁴

VFinancials

Revenue doubled from FY2016 to FY2025, 2.04 times, and net income grew faster, 2.2 times, from an FY2016 base of $4.9B that I adjust for one-offs. Earnings per share grew 3.0 times from an adjusted $5.10 because the share count fell 24%.C (calculated)³¹⁵¹⁷

Net income grew faster than revenue, 2.2 times against 2.0 times, and buybacks lifted earnings per share to 3.0 times.

Reported FY2016 net income of $5.4B carries $756M of after-tax gains on the Costco and JetBlue portfolio sales and $266M of after-tax restructuring costs; adjusted for both it is $4.9B, and reported earnings per share of $5.61 become about $5.10.C (calculated)¹⁵¹⁷

Where the margin went

Pretax margin was 20.4% in FY2016 on my adjustment, 19.1% in FY2025 and 19.1% in the TTM without the Q3 gain and charge. As reported, FY2016 was 22.7%, because it includes a $1,218M gain on the portfolio sales, which sits in Other expenses and so inside salaries and other expense, less $410M of restructuring charges. Adjusted, the margin was 20.1% in FY2017 and FY2018, 19.4% in FY2019, between 17.4% and 18.8% in FY2022 to FY2024, and swung with credit reserves in FY2020 and FY2021.^3451517

Amex spent 10.5 points of revenue more on rewards and Card Member services and recovered 9.3 of them by cutting salaries and other expense. Headcount rose from 56,400 to 76,800, so the saving came from revenue growing faster than people.C (calculated)³¹⁷ The saving is the part of my base case I trust least, because it has to continue: my base needs salaries and other expense to fall another 2 points of revenue by year 5, and the $336M restructuring charge in Q3 for efficiency initiatives is the company’s own step in that direction.E (estimate)⁴⁵

My base slows both lines. The table sets the pace they ran at against the pace in my cases; over the last four quarters rewards, services and business development did not rise at all.

I set the cost creep at 0.3 point a year because rewards, services and business development took 44.3% to 44.7% of revenue in each of the last four quarters, after the Platinum step. I set the saving at 0.4 point a year because it has slowed: salaries and other expense fell 0.7 point in FY2025 and 0.3 point in the three quarters since. If both ran at the historical pace my base would be $315; if the cost creep ran at the historical pace and the saving stopped, it would be $227.E (estimate)⁴⁶⁴⁴

Credit

Net write-offs on loans, principal only, were 2.2% in FY2019 and 2.2% in FY2025, with 0.9% in FY2021 and FY2022 and 1.8% in FY2023 between them; in Q3 2026 the principal-only rate on consumer and small business balances was 1.8%, and 2.1% with interest and fees. Credit is back at its pre-pandemic level and is not deteriorating.³⁴¹⁰¹⁴ The reserve is 2.5% of card balances, down from 2.9% a year earlier. Provisions as a share of revenue were 6.1% in the TTM against 7.3% to 8.1% in FY2023 to FY2025 and 8.2% in FY2019, so my base provisions of 7.0% are a return to normal levels.C (calculated)³⁶¹⁴ Credit is the cost that moves most between good years and bad ones: provisions were 13.1% of revenue in FY2020, when revenue fell 17% and pretax income halved, and negative 3.4% in FY2021, when reserves were released.C (calculated)¹⁰¹²

Capital and returns

Return on equity is high because Amex earns about 4 cents a year on each dollar of risk-weighted assets and holds about 13 cents of equity against each dollar.

The rise in return on equity from 25.8% in FY2016 came from earning more per dollar of risk-weighted assets and holding less capital against it. Equity per dollar of risk-weighted assets fell from 15.7 cents in FY2016 to 12.7 cents in FY2022 and has been steady since; the TTM figure of 4.4 cents includes a lower provision.C (calculated)³⁶ The room to lift the return further is small: CET1 is 10.9% against a 10% to 11% target, and American Express became a Category II firm in Q2 2026, which phases in heavier capital, liquidity and reporting requirements.⁴

Cash

I do not use operating cash flow or free cash flow for a lender, because card balances grow by billions a year and swing it. The measure I use is distributable earnings: adjusted net income to common less the increase in CET1 capital. In the TTM that was $8.5B, which is $11.6B of adjusted earnings to common less the $3.1B by which CET1 capital rose, from $26.2B to $29.3B, a rise that includes retained gains and other equity effects as well as growth. Cash paid out was $9.2B, 75% of net income and 4.4% of the market value, which is $0.7B more than distributable earnings. The CET1 ratio still rose from 10.5% to 10.9%, helped by the Q3 gain, which is in capital but not in my adjusted earnings.C (calculated)⁴⁵³⁴⁶

Accounting to watch

1. Reserves. The reserve fell from 2.9% to 2.5% of card balances in a year. Q3 provisions of $985M were 23% lower than a year earlier because of a reserve release against a build in the prior year; the $235M release on card balances is mostly the $215M benefit from a model update.C (calculated)⁴⁵⁷ My base puts provisions at $5.5B in year 1, up from $4.7B in the TTM.E (estimate)
2. One-offs in Q3 2026. Five items sit inside the quarter’s earnings, and the table shows which of them adjusted EPS removes.⁴⁵⁷
3. Other, net. The 10-Q names gains in Other, net that it does not size: a remeasurement gain on buying the rest of Swisscard, a release of a reserve for international non-income tax, a gain on selling a small business cobrand portfolio and Amex Ventures gains; unrealized gains on investments without readily determinable value were $306M in nine months against $91M.⁴⁶ Without the GBTG gain and the legal reserve, Other, net was $1.2B in the TTM against $1.6B in FY2025. The $0.4B gap is about the size of the $400M legal reserve, so the two roughly offset in my starting margin, and the one-input table in the valuation section shows each alone.C (calculated)⁴⁶⁴⁴
4. Yardsticks that changed. Loans and receivables became one line, Card balances, in Q1 2026. The net interest yield moved from a non-GAAP 11.9% for FY2024 to a GAAP 8.1% for FY2025, and a 2022 recast moved processed revenue out of discount revenue, which is why I show discount revenue per $100 only from FY2021. I do not splice these series.³⁴⁹¹¹

Adjusted EPS removes only the gain net of the charge, so it still includes the model benefit, the tax benefit and the legal reserve, a net of about $0.02. Adjusted pretax income rose 2% and adjusted net income 11%; the 31% rise in earnings per share to $5.42 includes the gain.C (calculated)⁵⁷

VIValuation

How this goes badly wrong

Before any numbers I ask what would most likely hurt an owner, and work back. I see three ways.

1. The product costs more each year. Rewards and services have risen about 1.0 point of revenue a year since FY2016. If that continues and the salary saving stops, my base is $227. This is the likeliest.E (estimate)³⁴¹⁵
2. A credit cycle on top. In FY2020 provisions took 13.1% of revenue, against 6.1% now. My bear case is a milder version, worth $189.E (estimate)⁴¹⁰¹²
3. A cap on card interest. Net interest income is 24% of revenue. Caps are proposed, not enacted. If that income fell 25% with costs unchanged, my base would be $207.E (estimate)³⁴

What an owner can take out

Adjusted earnings were $17.02 a share. An owner can spend only what is left after stock pay and the capital needed to grow. Stock pay, $0.55B in FY2025, is already inside expenses; but about 30% of the shares bought from FY2016 to FY2025 only replaced shares issued to staff.^3151720 At 9% growth in year 1, with CET1 held at 10.5% of risk-weighted assets, about $2.5B of the $11.6B of net income to common stays in the business. That leaves about $9.0B, or $13.40 a share, 78% of earnings, a 4.1% yield at $323.71 against 4.98% on the ten-year Treasury.C (calculated)⁴⁶⁴⁹

I cannot know the cost path or the next credit cycle. On cost alone my base runs from $227 to $324, and across every input I change from $189 to $455.E (estimate)

What the price implies

At $323.71 the market value is $217.1B. The stock fell 3.7% to $313.22 on the day of the Q3 results and has since regained most of it. At $323.71 that is 18.2 times the midpoint of FY2026 adjusted guidance of $17.75 and 19.0 times TTM adjusted earnings of $17.02. It is also 16.0 times the 2027 consensus of $20.21, which predates the results: the feed still lists the September quarter as a forecast of $4.47, against the $4.68 adjusted result.C (calculated)⁴⁵⁵³² The earnings yield on TTM adjusted earnings is 5.26% against a 4.98% ten-year Treasury yield, and the dividend yield is 1.17%. The stock is 6.3 times book value per share of $51.02.C (calculated)^454672949

Against the other card companies

The market pays about 34 times earnings for the two networks, which earn operating margins near 60% with no credit risk and grew revenue 14% in the latest quarter, and 12 to 14 times for the two lenders. It pays 19.0 times for Amex. Against the lenders, Amex earns a higher return on equity (34.5% against 9.7% and 18.0%), reports lower card write-off rates (1.8% against 3.98% and 2.91%) and grew revenue faster (9% against 6% and 8%), but its 19.1% pretax margin is below Capital One’s 20.9% and JPMorgan’s 42.0%. Against the networks it has about a third of the margin and slower growth (9% against 14%), and it carries credit risk. I think the market prices Amex between a network and a lender.C (calculated)^{45456232425262739404142}

What the price needs

Run my model backwards, holding its margins and capital rules. The price then needs revenue to grow 6.6% a year for ten years, to about $147B in the twelve months to September 2036, against $149B in my base and $179B in my bull.C (calculated) Revenue grew 8.2% a year from FY2016 to FY2025 and 11.0% a year from FY2022 to FY2025, and the company guides about 10% for FY2026.C (calculated)³⁵¹¹¹⁵ Billed business grew 7.6% a year over the nine years and revenue per $100 of billed business rose from $4.10 to $4.33, which is the other 0.6 point; at a flat $4.35 per $100, the price needs billed business to grow 6.6% a year.C (calculated)³⁴⁶¹⁵

The price needs revenue to grow about 6.6% a year for ten years if salaries and other expense fall 2 points of revenue by year 5, and about 8.2% a year if they do not.

The first row is the price against my base case as it stands. The second removes the expense saving and leaves everything else: the price then needs more growth than my base assumes, 8.2% against 6.7%, which is about the same as the 8.2% of the last nine years.C (calculated)³⁴

The discount rate

I discount at 9.5%. The capital asset pricing model gives 9.52%, which I round to 9.5%: the 4.98% ten-year Treasury yield in Q4 2026 plus Aswath Damodaran’s 4.20% implied equity risk premium for 1 October, times a beta of 1.081, which is AXP’s against the S&P 500 over 60 monthly returns to September 2026.C (calculated) Damodaran struck his 4.20% against a 5.29% Treasury yield; on that pair the rate is 9.83% and my base is $307, 5% below the price. The 116-month beta of 1.164 gives 9.9% and a base of $304.C (calculated)^29493031

A half-point on the discount rate moves my base by 8% to 9%: $353 at 9.0% and $299 at 10.0%.C (calculated)^45293031

Bear, base and bull

The model runs fifteen years from the TTM, with the Q3 gain and charge removed: revenue of $77.6B, pretax income of $14.8B, a 19.1% margin, and $11.6B of net income to common, which is $17.02 a share on 682M diluted shares.C (calculated)⁴⁵⁶ Revenue grows at a rate I set for each case and then fades in a straight line to the rate after year 15. The pretax margin is what remains after four costs, each a share of revenue: provisions; rewards, Card Member services and business development; marketing; and salaries and other expense. Tax is 21.5%, the rate for FY2024 and FY2025, which the 10-K says includes discrete benefits; at 22.5%, close to the 22.4% I get for the nine months of 2026 without the Q3 discrete benefit and the GBTG items, my base is $319.³⁴⁵

Distributable earnings are net income to common, after preferred dividends and the earnings allocated to restricted stock awards, less the CET1 capital needed as risk-weighted assets grow with revenue. Risk-weighted assets were $270.1B, 3.48 times TTM revenue, and I keep that ratio flat although it has fallen from 3.68 in FY2022. Capital is held at its high-water mark, so a recession releases none. Stock pay is inside expenses, so it is a cost in every case. I divide by 674M shares, the Q3 diluted average; 670.6M were outstanding at 15 October.⁴⁶

My values are about 5% conservative on timing. Cash arrives at each year end and values are dated 30 September, the date of the last balance sheet. Mid-year timing would add 4.6% to every case, and dating the values from the price date instead of 30 September would add 0.7%: with both, my base is about $341, my bear about $199 and my bull about $480, and the price would imply about 6% growth.C (calculated)⁴⁶

My base is below consensus, which predates the results. It grows net income 7% to 8% a year for five years.C (calculated) Consensus earnings per share rises 14.5% in 2027 and 12.0% in 2028, which is 11% and 9% growth in net income at the 2.7% yearly fall in the diluted share count, from 693M in Q3 2025 to 674M in Q3 2026.C (calculated)⁶³² My bull case grows net income 11% to 13% a year, below consensus in year 1 and above it from year 2, so consensus falls between my base and my bull.

The bear case is a recession on top of a costlier product. Revenue falls 8% in year 2 and provisions reach 12% of revenue, a milder version of FY2020, when revenue fell 17% and provisions were 13.1% of revenue. In each of the first five years rewards and benefits take 0.5 point more of revenue than in the year before, provisions settle at 8.0% and the margin stays near 15.5%, 3.6 points under the TTM. No capital is released in the recession.C (calculated)¹⁰¹² That is worth $189, or 11.1 times TTM adjusted earnings.

In the base case revenue growth fades from 9% to 4.5% in year 10, which is 6.7% a year. Provisions return to 7.0% of revenue. Rewards, services and business development take 0.3 point more of revenue each year for five years, against about 1.0 point a year since FY2016, and salaries and other expense fall 0.4 point a year for five, against about 0.9, so the margin holds near 19%. That is worth $324, which is 19.0 times TTM adjusted earnings and 16.0 times the 2027 consensus.E (estimate)⁴⁵³²

The bull case is that Millennial and Gen Z customers keep raising their spending. Growth stays at or above 8% for seven years, provisions stay near 6.5%, and salaries and other expense fall 0.7 point a year for five, which takes the margin to 21.4%, above the 20.4% of FY2016 on my adjustment and the 19.1% of the TTM. That is worth $455, or 26.7 times TTM adjusted earnings.C (calculated)³¹⁵¹⁷

The two groups of inputs that matter most are revenue growth and the costs together; the capital rule matters little. Putting the bull’s revenue growth into my base adds $52 and the bull’s costs and credit add $46. A one-point-higher CET1 target takes only $8 off, because the capital stays in the business.C (calculated)⁴⁶

The cost rows are the ones to read first. The saving is a $35 lever; cost creep at 0.9 point a year is a $53 lever; creep at the historical pace of about 1.0 point with no saving takes $97 off. At the historical pace on both lines my base barely moves, because the two offset. A net interest income cut is the largest single revenue risk: 10% takes $47 off and 25% takes $117, with costs unchanged.E (estimate)⁴⁶

One item sits outside the cases. The Kabbage bankruptcy estate alleges that Amex’s purchase of Kabbage’s lending platform and other assets was a fraudulent transfer and seeks up to about $746M, or $1.11 a share before tax. A second complaint against Kabbage’s former directors and officers seeks some or all of the same amount, and they say Amex must indemnify them, which Amex disputes. Amex’s range of reasonably possible loss above accruals for all disclosed matters is zero to $150M.C (calculated)⁴ The $400M added to legal reserves in Q3 is inside my cases: the cost base my model starts from keeps it, 0.5 point of revenue, and every case carries it forward as a share of revenue. The 2026 gains inside Other, net are inside it too. If the reserve does not recur my base is $334; if the gains do not recur it is $315.E (estimate)⁴⁵⁶

The margin of safety

• Price$323.71Close, Q4 2026 [45]
• Base value$324My estimate E (estimate)
• Margin of safety below$22730% under my base value C (calculated)

I set the margin of safety at 30% below my base value, which gives $227. My rule for a discount to base value is 25% for a business with net cash and no customer above a tenth of revenue, 30% when one cycle or two customers drive results, and 35% when fixed charges such as leases sit ahead of the shares. Results here turn with the credit and travel cycle: provisions swung from 13.1% of revenue in FY2020 to negative 3.4% in FY2021, travel and entertainment is 29% of billed business, and Delta is 13% of billed business and 21% of loans.³⁶¹⁰¹² I use 30% because the 35% tier is meant for leases and fixed rents, and American Express’s deposits and debt are matched by $221.1B of card balances and $43.2B of cash at 30 September 2026, with the interest on them already deducted from revenue. Assets were 9.0 times equity at the end of FY2025. At 35% the margin-of-safety price would be $211.C (calculated)³⁴⁶

The price is 43% above that line, so the stock would have to fall about 30% to reach it.C (calculated) The lowest close of the past year was $292.27 on 27 March 2026, still 29% above it. The stock is down 12.5% in 2026 and 9.8% over the year to the Q4 2026 close.C (calculated)⁴⁵⁴⁶ My bear case of $189 is 42% below the price and 17% below the margin-of-safety price, so the discount covers a downturn milder than my bear case.C (calculated)

The same dollar could go to Mastercard, whose base return is 7.6% a year on its page, or Visa at 7.1%. Mine is 9.5%, against 4.98% on a ten-year Treasury.E (estimate)⁴⁹ Both sit above their base values, by 13% and 18%, and American Express is level with its own.

The business and the price are two questions. The business earns 33.9% on equity. The price gives no margin of safety, which starts at $227. This looks like a very good business at a fair price, not a cheap one.³⁴

Where this could be in 5 and 10 years

These are my model's values at later dates. Each is the present value, at 9.5%, of the owner cash that comes after that date, per today's 674M shares and in that year's dollars. Owner cash is net income to common less the capital needed to grow. It is what the business could pay out. I assume no buybacks. Dates are 30 September.

For the base to stand at 5 years, revenue has to reach about $114B, which is 8% a year, with the pretax margin near 19%. That is $16.6B of earnings. To stand at 10 years, growth has to fade to 4.5% and revenue reach $149B, with $21.7B of earnings. Value then grows about 4% a year and the owner cash adds about 5 points, so the return is the 9.5% discount rate: today's price sits on my base value.E (estimate)

The bear has a recession in year 2, with revenue down 8%, a margin near 15.5% and growth near 3% a year, to $104B. The bull has growth at or above 8% for seven years and a margin of 21.4%, to $179B. The bear's return is low mostly because the price starts $136 above its value.C (calculated) Growth and costs drive the gap most. The discount rate comes next: a half-point moves the base 8% to 9%.

The most likely way my base is wrong is cost. It has rewards and benefits taking 0.3 point more of revenue each year for five years. They have taken about 1.0 point a year since FY2016. It also has salaries and other expense falling 0.4 point a year. If benefits keep rising at the old pace and that saving does not come, my base value today is $227, and the later values fall with it.E (estimate)

These are estimates of value, not forecasts of the share price. The market price can sit above or below value for years.

VIIThe case for the price

What most people miss

The fee and the cost run on different clocks. The richer Platinum benefits started on 18 September 2025.³³ Existing U.S. consumer members were due to pay the $895 fee only at their first renewal from 2 January 2026, per the travel press,⁵³ and Amex books each fee evenly over twelve months.³ So the cost landed at once and the fee is still arriving. The last reset by January 2027, so a shrinking lift to year-over-year growth runs through 2027, after the cost has lapped. One year on, the CFO, Christophe Le Caillec, put spending retention on the portfolio at 99%, in a machine transcript I could not match to Amex’s documents.⁵⁰

The filings put the $336M charge down to “efficiency initiatives”.⁴ On the call Stephen Squeri said it removes about 5% of positions.⁵⁰⁵⁵ If pay falls with positions, that is about $0.5B a year on about $9.6B of salaries without the charge, or 0.6 point of revenue. My base asks for 0.4 point a year.E (estimate)³⁶ Squeri gave no savings figure and said it frees capacity to invest, so it may not all reach the margin.⁵⁰

The bull case

The bull case is a membership business that compounds at 19.0 times earnings, against about 34 for the networks. The multiples are my comparison. The compounding is Le Caillec’s, from 23 October. Return on equity is around 35%, against 25% to 28% a decade ago.⁷ Revenue is on track to more than double and earnings per share to triple over that time, with shares down over a quarter. Squeri said 2026 is on track for 10% revenue growth and mid-teens growth in earnings per share.⁵⁰

Greg Abel’s first letter, dated 28 February 2026, names Amex among businesses Berkshire expects “will compound over decades”. The $1.3B stake was worth $56.1B at year-end.⁵¹ Berkshire’s 13F lists 151,610,700 shares at 30 June 2026.²⁸ Warren Buffett wrote in February 2023 that the dividend checks “seem highly likely to increase”.⁵² MarketBeat reports that JPMorgan raised its price target to $365 on 26 October, level with MarketBeat’s average. I have not read the note.⁵⁴

Against reality

The price sits at my base, so the test is simple. If revenue grows 6.6% a year for ten years, rewards and services rise no more than 0.3 point of revenue a year for five years, and salaries and other expense fall 2 points, then $324 is possible.C (calculated) Today’s numbers meet two of those three, for now. Revenue grew 9% in Q3. Rewards and services took 44.3% of revenue, under the 46.1% my base allows by year 5. Salaries and other expense are the gap: 21.3% without the Q3 legal reserve, 21.8% with it, against the 19.8% my base needs. That is 1.5 to 2 points to go.C (calculated)⁴⁶ Without that saving the price needs 8.2% growth. History is the catch: rewards and services rose about 1.0 point a year from FY2016, and at that pace my base is $262.E (estimate) Three things have to go right at once. Say 65% for growth, 40% for the cost share and 50% for the saving: 13% multiplied.E (estimate) They lean on each other, so this is not the odds of $324, only how much must line up.

Management’s framework asks for more. Mid-teens growth in earnings per share, less the 2.7% yearly fall in shares, is about 12% growth in net income. Five years of that takes today’s $11.6B to $20.4B, close to the $20.5B in year 5 of my bull case.E (estimate) Five years is my horizon, not theirs. On 10% revenue growth it takes a 21.1% margin, 2 points above today’s 19.1%. So if they hit those targets for five years and growth stays at or above 8% for seven, $455 is possible. The margin is the hard part. Since FY2016 it has passed 21.1% only in FY2021, when reserve releases made provisions negative.¹² Ten percent growth for five years at my base margins gives about $360, near that $365.E (estimate)

What they have to hit

Where I come down

The fee timing moved me. It makes a flat cost share look reachable, and if rewards and services stay at 44.6% my base is $350.E (estimate) Ten years of those costs rising a point a year outweigh one good quarter, so my 0.3 point stays for now. Two years of rewards and services at or below 45.6%, salaries and other expense below 20.5%, and 2027 revenue guidance above 9% would raise my values.

VIIIRisks, and what would change my view

The company’s own risk factor on the cost of attracting members is the one I weight most:

costs and Card Member attrition could increase

1. The cost of the proposition. Variable customer engagement costs grew 16% in Q3 2026 against revenue growth of 9%, and Card Member services expense grew 36%, but that growth still includes the Platinum step that began in Q4 2025 and laps from the fourth quarter of 2026.³⁵⁷ If rewards, services and business development rise 0.9 point of revenue a year for five years against my 0.3, the base is $271; at the FY2016 to TTM pace of about 1.0 point, with the saving unchanged, it is $262; with no saving as well, $227. If only the saving stops, it is $289.E (estimate)
2. Credit. Provisions are low partly because the reserve is being released. If they settle at 8.0% of revenue the base falls $19, and a year-4 recession takes $39 off it.E (estimate)⁵⁶ Net write-offs on loans were 2.2% in FY2019, 2.2% in FY2025 and 1.8% for card balances in Q3 2026, so credit is at its pre-pandemic level; my base provisions of 7.0% of revenue assume a return to normal.³⁴⁶¹⁴
3. Regulators. On 8 October 2026 American Express Company and its main operating subsidiary, TRS, consented to a Federal Reserve order, and American Express National Bank to an OCC order, over parts of the financial crimes compliance program. The bank agreed to a $350M penalty, partly reserved before. The 8-K says the orders impose no asset cap and that the cost of meeting them is not expected to affect 2027 guidance; the 10-Q says a failure to meet them could bring further penalties or material restrictions on the businesses, and neither document describes a restriction on new products, markets or acquisitions.⁴⁴³ Growth 2 points lower in years 1 and 2 would take $10 off my base, and a compliance build of 0.3 point of revenue in every year would take $6 off.E (estimate) Category II status adds requirements; Amex’s preliminary reading of the March 2026 capital proposals is a modest reduction in risk-weighted assets to about neutral.⁴
4. Rules on pricing. The 10-K lists “recent proposals to cap credit card interest rates” among its consumer-protection risks, and the 10-Q lists governmental actions to cap them among the factors that could change net interest income; neither says that a cap has been enacted.³⁴ Net interest income was 24% of the TTM revenue.C (calculated)⁴⁶ If it were 10% lower from year 1 with costs unchanged the base would be $277, and 25% lower, $207.E (estimate) Two merchant class actions, Pizza Hazel and 5-Star General Store, challenge the anti-steering rules; Amex warns that significantly increased steering after an adverse result could materially affect the business. Australia stopped requiring surcharging to be permitted for Visa and Mastercard on 1 October 2026, and Amex no longer permits it there.³⁴ On the other side of the ledger, discount revenue per dollar of billed business fell 7 basis points from FY2022 to the TTM; another 2 basis points a year for five years would take the base to $284.E (estimate)
5. Partner concentration. Delta is about 13% of billed business and 21% of loans. All cobrand cards are about 26% and 36%. The Delta agreement runs through 2029, and the 10-K says billed business could fall and attrition rise significantly if a partnership ended.³ Amex extended its Marriott partnership in Q3 2026 and exited two small business cobrand portfolios, which cut Q3 billed business growth by about 0.6 point.⁴⁵
6. Technology. The 10-K says agentic commerce, stablecoins and fintech competitors could disintermediate card relationships. It would show first in the share of new accounts on fee-paying products and in the discount revenue rate, and the yield row above is its price in my model.³
7. People and ownership. Squeri is 67. Berkshire’s commitment to follow the board on 17% of the vote lasts only while he is chief executive and Berkshire owns at least 5%. The vice chairman retired in March.³⁸

What would move me

The first test is the fourth quarter. FY2026 adjusted guidance of $17.60 to $17.90 and nine-month adjusted earnings of $13.49 imply fourth-quarter adjusted EPS of $4.11 to $4.41, against $3.53 a year earlier.C (calculated)⁵⁶ It is also the first quarter in which year-over-year growth in Card Member services expense no longer includes the Platinum launch. American Express has not announced a date for its fourth-quarter and full-year results; it released the FY2025 results on 30 January 2026.³ Berkshire’s 13F for 30 September, due by mid-November, will show whether its 151.6M shares changed.²⁸