American Express AXP

Thesis

American Express is a closed-loop payment network bolted onto a premium-customer franchise. The same company issues the card, owns the network rails, signs the merchants, and underwrites the credit. That structure produces three streams: discount revenue (merchants), card fees (cardholders), and net interest income (lending). The first two are network/franchise economics; the third is bank economics. The blend is unusual and durable.

The scorecard composite is 76/100 (profitability 15, balance sheet 18, capital allocation 20, valuation 23). Five-year FCF conversion is 1.33x — owner earnings exceed reported net income, which is exactly what you want from a fee-rich, light-physical-capex business. TTM owner earnings of ~$9.18B at a current price of $319.68 imply an EV/FCF of ~20.4x and a P/E of 21.84x versus a 10-year average of 26.31x. The reverse-DCF says the market is pricing in only 4.73% perpetual growth in owner earnings, well below AXP's 10%+ revenue growth in 1Q26 and below its long-run billings CAGR.

The scorer's deterministic IV range is $234.55 (low) / $494.86 (base) / $632.43 (high); current price $319.68 sits at 0.646x base IV. That discount is the margin of safety. The 'low' IV ($234.55) implicitly bakes in a meaningful credit-cycle event; even there, the gap from current price is only ~27%, while the upside to base IV is +55%. Note the financials caveat — reported ROIC of 0 and missing net-debt/EBITDA reflect bank-holding-company accounting, not fundamental capital destruction; AXP's ROE has run in the high-20s% to mid-30s% for years.

Thesis: own the spend network, get the lender for free, and pay a price that already discounts a recession.

Moat

AXP's moat is genuinely multi-source, which is rare. Five lenses:

Network effects (closed loop). Visa and Mastercard run open-loop networks: issuer banks and acquirer banks settle through them and they take a small toll. AXP runs a closed loop: it issues the card, signs the merchant, and clears the transaction itself. That gives AXP first-party data on every transaction across both sides of the market, which feeds underwriting, fraud detection, and merchant marketing. The closed loop is the reason Buffett kept doubling down on the position from the salad-oil scandal in the 1960s through the additions in 1994 and 1998 [1][2][4]. Network effects work because (a) more cardholders make the card more valuable to merchants and (b) more accepting merchants make the card more valuable to cardholders. After ~40 years of MasterCard/Visa parity acceptance in the U.S., AXP still grew billings by 10% in 1Q26 — the network effect is intact.

Switching costs (intangible / behavioral). A premium AXP cardholder is locked in by accumulated Membership Rewards points, airline/hotel status mapped to specific cards, lounge access, statement credits on Uber/Resy/Walmart+, and a credit history they don't want to interrupt. Buffett identified this dynamic as a 'tide-like trend' as early as 1970 with Blue Chip Stamps [6]; AXP has now spent 40+ years compounding that loyalty design. Annual fees of $695 (Platinum) and $325 (Gold) self-select customers who view the card as a lifestyle product rather than a transaction tool. Switching costs are real and growing as the rewards ecosystem deepens.

Intangible asset (brand). The Centurion / Platinum brand is a status good. Premium card fee growth was the largest single driver of net card fee growth in 1Q26. There is no substitute brand at this price point — Chase Sapphire Reserve and Capital One Venture X compete on rewards math but not on the cultural cachet of the metal card. Brand-as-pricing-power is what allows AXP to keep raising fees and still grow proprietary cards in force.

Cost advantage (data + scale). Closed-loop data plus 130M+ cards in force gives AXP underwriting accuracy that pure-play monoline issuers cannot match. AXP's loss rates have been below industry averages for over a decade, even through 2008-09. Lower loss rates per dollar of receivables = structurally lower required reserves = higher ROE. This is a quiet but compounding advantage.

Pricing power (merchant discount). AXP's merchant discount rate is structurally higher than V/MA interchange (typically 230-250 bps vs 150-200 bps). Merchants pay it because AXP cardholders spend more per transaction and skew higher-income. The MD&A explicitly notes 'lower average merchant discount rates primarily due to shifts in geographic and merchant spend mix' — meaning the price is being defended on a like-for-like basis even as mix shifts internationally and into smaller merchants.

Competitor stress test ($10B + 5 years). Could a well-funded entrant replicate this? JPMorgan tried with Sapphire Reserve and pulled rewards back when economics broke. Apple+Goldman tried with Apple Card and Goldman is exiting the partnership. The reason: the closed-loop + premium-brand + 75-year merchant relationship stack cannot be cloned with capital alone. The brand is the bottleneck and brand takes decades.

Erosion risks. (a) Stablecoins / account-to-account rails could compress merchant discount over a 10-year horizon; AXP gets some offset because it owns the issuance side. (b) Open-loop networks plus fintech wallets (Apple Pay, Cash App) compete for transaction layer, but not for the rewards layer. (c) Regulatory: Durbin-style debit caps don't apply to credit; a future credit-interchange cap is a tail risk but politically remote in the U.S.

Moat verdict: WIDE.

Management & Capital Allocation

Stephen Squeri (CEO since Feb 2018) inherited a company that had just lost the Costco co-brand and was widely seen as ex-growth. Eight years later, billings, cards in force, and net card fees have all reaccelerated. The 1Q26 print — revenues +11%, EPS +18%, billings +10%, premium card fees driving growth — validates that the premium-refresh strategy worked. That is the macro management track record.

The five capital-allocation choices:

1. Reinvest in the business. AXP spends heavily on rewards (the 'card member rewards' line is the largest expense after interest), marketing, and technology. The right way to read these is as customer-acquisition and customer-retention spend; they are operating expenses that protect the moat. Rewards expense scales with billings, which is fine. Marketing has been disciplined — AXP doesn't chase every co-brand at any price (notably let Costco walk in 2016 rather than overpay).

2. Acquisitions. Squeri's M&A has been bolt-on and digestible: Kabbage (small-business lending platform, 2020), Resy (restaurant booking, integrated into card benefits), Tock (acquired 2024). Accertify was divested in 2Q24, which the 10-K filing references. No transformative deals, no goodwill blowups. This matches Buffett's preferred profile: management that doesn't 'do something' for the sake of it.

3. Debt. AXP is a bank holding company so leverage is regulator-managed. Net debt / EBITDA is not a meaningful metric here (the scorecard correctly shows it as null). The relevant number is CET1 capital ratio, which AXP runs in the 10-11% range — comfortably above regulatory minimums. The bank-holding structure is a constraint but also a discipline: the Fed effectively backstops capital allocation.

4. Buybacks. Share count is down only 2.97% over 10 years — modest by S&P 500 standards. This understates the underlying repurchase activity because AXP also issues equity for compensation; gross buybacks are larger. The relevant question is whether buybacks were done above or below IV. With the stock having traded between roughly $50 (2016) and $320 (today), and the scorecard's base IV at $494.86, most repurchase years were below IV. Average P/IV on buybacks is plausibly in the 0.5-0.8x range — value-accretive on average. Not best-in-class (a Henry Singleton would have repurchased aggressively in 2016 and 2020), but not destructive either.

5. Dividends. Modest payout ratio (~20%); the dividend is a reliable check on cash discipline without consuming the reinvestment runway. Twelve consecutive annual dividend increases.

Communication quality. Squeri's investor day cadence, segment disclosure, and the recent reclassification of card balances (combining loans and receivables to match industry convention, 1Q26) all point to clean, evolving disclosure. The 10-K and 10-Q language is straightforward — no euphemisms when credit costs rise.

Incentive alignment. Long-term comp tied to ROE and EPS growth. Insider ownership is modest (this is a large bank, not a founder-led company), and Berkshire Hathaway remains the largest holder at ~21%. Buffett's continued ownership is, in itself, a governance signal — Berkshire has held the position essentially untouched since 1998 [2][4].

Knocks: (a) buybacks could have been more aggressive at the 2016 and 2020 lows; (b) the Kabbage acquisition has been a slow burn; (c) executive comp is high in absolute dollars even by big-bank standards.

Capital allocator: B+.

Industry Structure

Porter's Five Forces applied to the global premium-payments industry:

1. Rivalry among existing competitors — MODERATE. The card industry looks crowded (Visa, Mastercard, Discover, JCB, UnionPay, plus issuer banks like JPM, C, COF, BAC) but the premium-charge-card-with-closed-loop subsegment is essentially AXP alone. Visa Infinite and Mastercard World Elite are issuer-bank programs running on open-loop rails; they compete on rewards but not on the integrated experience. Discover is closed-loop but mass-market, not premium. Rivalry exists at the rewards-arms-race level (Chase Sapphire Reserve raised its annual fee to $795 in 2025), but that competition is rational — everyone is chasing fee income, not undercutting on price.

2. Threat of new entrants — LOW for premium closed-loop, HIGH for transaction layer. Building a global premium card network from scratch is essentially impossible because you need (a) two-sided scale, (b) decades of brand, (c) bank-holding-company licensing. The transaction layer is a different story: Apple Pay, Cash App, Venmo, PayPal, stablecoin wallets all compete for the act of payment. The crucial point is that they compete with the form factor, not with the underwriting and rewards layer that produces AXP's profits. AXP cards run inside Apple Pay just fine.

3. Bargaining power of buyers — MIXED. Cardholders have low individual power but high collective optionality — they can switch to Sapphire Reserve. Merchants have higher power than they used to: the EU has capped interchange, Australia has limited surcharging restrictions, and large U.S. retailers periodically threaten AXP non-acceptance (Costco 2016 was the canonical case). The defense is cardholder spend per transaction: merchants accept the higher discount because the AXP customer's basket is larger and incremental.

4. Bargaining power of suppliers — LOW. AXP's main 'suppliers' are co-brand partners (Delta, Hilton, Marriott) and processing infrastructure. Co-brand partners have meaningful power (Delta is ~9% of billings) but are also locked in by joint customer bases. Tech suppliers are commoditized.

5. Threat of substitutes — MEDIUM. Substitutes are debit, ACH/instant payments (FedNow, Pix in Brazil, UPI in India), BNPL (Affirm, Klarna), and stablecoins. The realistic 10-year scenario is that some share of low-margin retail transactions migrates to instant-payment rails, while premium credit holds because of the rewards/insurance/dispute-resolution bundle that those rails don't offer. The premium customer is harder to disintermediate than the average customer.

Value pool location and trajectory. Within global payments, the value pool sits in (a) interchange/discount fees on credit (AXP's home turf), (b) net interest income on revolving balances (AXP and bank issuers), and (c) rewards monetization (AXP, Chase). The pool is growing nominally with global consumption and faster than GDP because of the cash-to-card secular shift in international markets where AXP under-indexes. AXP's international billings are growing faster than U.S., consistent with this thesis.

Industry Verdict: Good. The specific niche AXP occupies (premium closed-loop) is structurally Excellent; the broader payments industry it sits inside is more contested and faces real disintermediation risk over a decade-plus horizon.

Inversion (Bear Case)

I am now the short. I will not hedge.

1. The single event that kills this. A consumer credit cycle worse than 2008, layered onto AXP's structurally higher credit-card lending exposure post the loans/receivables reclassification (1Q26 10-Q makes the combined Card balances explicit). AXP's provision for credit losses was $1.25B in 1Q26, up 9% YoY. In a 2008-style scenario, provisions could 3-4x to $4-5B per quarter. AXP's TTM net income is roughly $11.5B; a $12-15B annualized provision build wipes out most of it and forces capital preservation. The stock would drop 50%+ as it did in 2008-09. Unlike Visa or Mastercard, AXP has the lender's balance sheet and shares the lender's downside.

2. Why the moat is narrower than bulls think. The closed-loop advantage is real for premium cards but the transaction layer is being commoditized. Apple Pay, Google Pay, Cash App, Venmo, FedNow, Pix, UPI — every one of these reduces the consumer's awareness of the underlying card and weakens the brand's emotional pull. Younger consumers think 'Apple Pay' before they think 'Amex.' The Membership Rewards lock-in works only if customers care about the points; a generation raised on instant cashback and price-comparison apps is structurally less loyal. The scorecard's reverse-DCF implied growth of 4.73% may look low to bulls, but if billings growth decelerates to 3-4% nominal as cash-to-card saturates and the youngest demographic skews to debit/wallet, even that growth becomes hard. Furthermore, a wide moat that stops widening is not the same as one that compounds — AXP's pricing power on merchant discount has been declining (the 1Q26 MD&A admits as much), masked only by mix shift.

3. Why management is worse than it appears. Squeri has executed well in a benign credit and rate environment. The test of bank management is the next downturn, not the last upcycle. Buyback discipline has been mediocre — only 2.97% net share count reduction over a decade despite multiple obvious low-price windows in 2016 ($50s) and 2020 ($85). A Singleton would have shrunk the share count 30-40% in that period; AXP shrank it 3%. Management has prioritized growth over per-share compounding, which is a subtle but real form of empire preservation. The Kabbage small-business-lending acquisition is showing slow returns. Executive compensation is high in absolute dollars. Most importantly, when the next credit cycle arrives, the 'we don't take outsized credit risk' narrative will be tested — and AXP's lending mix has been growing, not shrinking.

4. What bulls are extrapolating that won't hold. Bulls extrapolate (a) 10% billings growth indefinitely, (b) premium-card fee growth of 12-15% indefinitely, (c) credit losses staying near current levels, (d) multiple expansion back to the 26x 10-year average P/E. All four can break simultaneously. Premium-card fee growth is being pulled forward by a one-time refresh of Platinum/Gold; once everyone is repriced, the growth rate normalizes to billings growth. The premium-card category is also crowded — Sapphire Reserve at $795, Venture X at $395, Apple Card etc. — which limits AXP's share gains. Credit losses normalizing to 5% of loans (vs ~2% today) on a $130B+ Card balance book is $5-7B of incremental annual provisions. Multiple expansion back to 26x assumes a stable rate regime; if 10-year yields stay above 4%, financial multiples stay compressed.

5. Valuation trap. The scorecard's base IV of $494.86 assumes a 10-year DCF with growth roughly in line with historical, normal credit losses, and stable margins. Substitute (a) 4% billings growth, (b) credit losses normalizing to 2008 + 30%, (c) merchant discount rate compression of 15 bps over a decade, (d) P/E re-rating to 16x in a higher-rate world. The DCF collapses to roughly $200 — below the scorer's low IV of $234.55. The current price of $319.68 looks cheap against $494.86 base IV but expensive against a credibly bearish $200 IV. The 'margin of safety' is partly an artifact of optimistic base assumptions about a financial company at the late stage of the credit cycle. Bank stocks repeatedly trade through 'cheap' multiples on the way to a real downturn.

If I am right, the stock could be worth $180-200 within 2-3 years.

Lollapalooza Bias Check

Biases active in me as the analyst right now:

Authority bias (strong). Buffett owns 21% of AXP, has held it since the 1960s (with re-purchases in 1991-94 and 1998), explicitly named it alongside Coke as a top conviction position [1][2][4]. It is genuinely difficult to write a bear thesis on a stock that one of the greatest investors of all time has held for 60 years and continues to hold. I am consciously discounting the Buffett halo, but it is shaping my prior toward 'durable franchise.'

Confirmation bias (medium). The scorecard composite of 76 and the P/IV of 0.646 are headline-attractive numbers. Once I see those, I am likely to over-weight evidence consistent with 'cheap, high-quality' and under-weight evidence of credit-cycle risk. I tried to counteract this in the inversion section by stress-testing the IV assumptions explicitly.

Recency bias (medium). 1Q26 results were strong (revenue +11%, EPS +18%). I am extrapolating off a recent high-water print. Long-run billings CAGR is closer to 7-8%, not 10%. The reverse-DCF implied 4.73% is a useful anchor against this bias.

Anchoring (medium). The 10-year average P/E of 26.31x makes the current 21.84x look 'cheap,' but the 10-year average includes the post-COVID multiple expansion era. A more honest anchor might be the 18-22x range AXP traded at through most of 2010-2019. Anchored to that band, the current multiple is mid-range, not bargain-bin.

Social proof (low-medium). Active value-investor consensus on AXP has been positive for years; popular financial press treats it as a high-quality compounder. I am aware that this consensus could itself be priced in.

Deprival super-reaction (low). Not buying a Buffett stock at a 35% discount to base IV feels like leaving money on the table — an instinctive aversion to missing the trade. I'm noting this rather than acting on it.

Incentive bias (low here). I have no economic stake in the recommendation, but the framework rewards crisp Buy/Sell calls over 'Hold' — which subtly biases toward action.

Net read: the dominant active biases are authority and confirmation. Both push me toward Buy. The countervailing inversion case is real and was written deliberately to resist them. The recommendation below tries to honor both: a Buy with measured conviction, not a thump-the-table conviction-high Strong Buy.

10-Year Outlook

Will AXP look fundamentally the same in 2036?

Same business model? Highly likely. AXP has been a charge-and-credit-card-issuing closed-loop network for 75 years and has survived multiple technology waves (mag stripe → chip → contactless → mobile wallet). The core economic engine — premium customer pays fee + spends a lot, merchant pays discount, AXP underwrites the credit — has been intact since the 1960s. The form factor will keep changing (more transactions inside Apple Pay, more inside in-app checkout) but the issuer-merchant-network triangle persists.

Customer base larger? Likely yes. Cards in force have grown for 15 consecutive years. International expansion is the multi-decade tailwind: AXP is structurally underpenetrated in Europe and Asia versus its U.S. position. The 10-year base case is +30-50% cards in force, weighted toward international.

Profit per customer higher? Likely yes, with caveats. The premium card refresh playbook (raise fee, add benefits, retain) has worked for two cycles. Card fee revenue per card is up materially over 10 years. The next decade will test whether AXP can repeat this with a Gen-Z customer who wasn't raised on points culture. Probable, but the slope is uncertain.

Moat wider? Probably wider in premium-customer brand and data, possibly narrower in transaction-layer differentiation as wallets and instant-payment rails commoditize the act of payment. Net: roughly stable to modestly wider in the segments that matter for profit.

Single biggest threat. Two related ones: (1) a severe consumer credit cycle that forces AXP to reserve $4-5B/quarter and reveals that the 'we underwrite better' story is only true on average, not at the tail; (2) generational shift away from points-culture loyalty toward debit/wallet/cashback simplicity. Both are real but neither is a death threat — they would compress margins and growth, not break the franchise.

Confidence assessment. The business will exist, will be larger, and will be profitable in 2036. The uncertainty is around per-share earnings growth rate (could be 5%, could be 10%) and credit-cycle timing (will hit at some point in the window). That uncertainty is meaningful but bounded. The financials caveat (ROIC = 0 in the scorecard is an artifact of bank-holding accounting, not a fundamental issue) does not change the qualitative judgment.

CONFIDENCE: medium