In September 2011, Borders Group closed the last of its 399 bookstores and dismissed the final ten thousand of its employees.1 The proximate cause of death was a company in Seattle that sold the same books without the store. Jeff Bezos’s fortune, roughly $200 billion today, mainly from his shares in Amazon.com, Inc., was built in meaningful part on graves like this one.2

A fortune that large invites a zero-sum intuition: surely someone, somewhere, is $200 billion poorer. Economics calls this mechanism a transfer, where one’s gain is another’s loss. But there is a second mechanism, creation: surplus that did not exist before a transaction occurred and is divided among its parties. Nobel Prize-winning economist William Nordhaus estimated that innovating firms have historically captured only about two percent of the social surplus their innovations generate; the rest spills to consumers.3 The question phrasing, “at the expense of,” quietly assumes the arrow points the other way.

This essay argues two things. First, the bulk of Bezos’s fortune was not taken from anyone; rather, it was created: a surplus that did not exist before Amazon did. Second, for the portion that was transferred, the biggest losers are not the customers or workers the question names, but third-party sellers and the rival firms Amazon destroyed. 

Handled with assumptions tilted against Bezos wherever the evidence is disputed, transfers sum to roughly a sixth of the fortune; to reach even a third requires assumptions the data contradicts. The largest transfer channel is a party the question never names: Amazon’s third-party sellers. The cleanest zero-sum losers, rival firms, are not named either. The question is a magnitude problem.

No Expropriation at the Checkout

The conventional defense is quick. More than two hundred million Prime members repeatedly choose Amazon over alternatives that sit one click away. Every voluntary purchase generates consumer surplus with lower prices, wider selection, and two-day delivery, which brick-and-mortar retailers had treated as physically impossible.4 Getting rich by selling people things they keep choosing, even when the alternative of leaving their home costs nothing, means buyers value the product above its price. That is consumer surplus, not extraction.

But the quick defense leans on revealed preference, and revealed preference is contested. Behavioral economics has documented how defaults, friction, and engineered choice architecture can manufacture “voluntary” behavior. Amazon itself paid $2.5 billion in September 2025 to settle Federal Trade Commission charges (without admitting wrongdoing) that its Prime enrollment and cancellation flows demonstrated this very same behavior.5 A serious answer cannot pretend that customers are frictionless sovereigns.

Thus, I will defend a deliberately weaker claim. Repeated, low-switching-cost transactions cannot prove that customers maximized their welfare. They do rule out expropriation. A fortune assembled from exchanges that either side could costlessly refuse is not money taken from customers, unless the prices themselves were rigged upward by something other than competition. The one economically coherent route to that conclusion runs not through the checkout, but through the sellers whose fees are passed onto buyers. Thisargument, focused on pricing power, is taken up in Section IV.

Employees: A Real Markdown, and its Ceiling

Wage-setting power is real, it is general (not peculiar to Amazon), and it runs in Bezos’s favor, but it is bounded, and the bound is small. The textbook dismissal statesthat wages equal a worker’s outside option. However, Amazon’s starting pay exceeding the local retail median no longer survives the labor-economics literature. David Card’s presidential address to the American Economic Association marked the mainstream’s turn: labor markets are monopsonistic.6 Meta-analyses find low quit elasticities, meaning most workers will not leave over a ten percent pay cut. This gives employers, Amazon included, genuine latitude to mark wages down below marginal product.7 I concede the direction.

Two questions remain: the counterfactual and the magnitude. The counterfactual is genuinely unsettled. Studies of warehouse openings considered flat employment-adjusted effects on local warehousing wages: fulfillment centers add warehouse jobs without raising overall county employment. This suggests the new positions largely displace existing ones rather than expanding the local labor pool.8 Attempts to find a positive wage effect have fared no better. A widely cited finding that Amazon’s 2018 $15 minimum lifted wages at other employers in the same commuting zones was later withdrawn by its own authors, who concluded that the spillover did not survive proper testing.9 When even the direction of the net effect is contested, the claim “he got rich by marking wages down from the competitive level” cannot be carried; there is no agreed competitive level to mark down from.

Magnitude is more decisive because it can be bounded. Table 1 stacks every assumption against my thesis: a fifteen percent markdown, above what Amazon-specific studies support. Of today’sfull 1.16-million-person operations workforce, applied retroactively across thirty years, Amazon’s headcount, for most of this period, has represented a small fraction. It also does not adjust the wage stream to present value. The method is deliberately crude in Bezos’s disfavor: rather than discounting each past year’s dollars to present value, I simply sum the full annual transfer across all thirty years at face value, which overstates the total; I then assign Bezos his roughly nine percent equity share, about $20 billion of his stake, or roughly a tenth of the fortune. A realistic estimate, using markdowns, the Amazon-specific literature actually supports, and historical headcounts, lands nearer a twentieth.

Table 1. Labor-transfer Ceiling

InputValueBasis
Operations workforce1,167,000 employeesEvidence: 75% of 1,556,000 total employees.10
Mean annual wage$45,760Evidence: $22/hr 2,080hrs11
Annual operations wage bill~53BDerived: workforce wage
Assumed markdown15%Assumption: above Amazon-specific findings, high end of cited ranges
Implied annual transfer~$8BDerived: markdown wage bill
Thirty-year undiscounted sum~$240BDerived: ~8B 30 yrs at today’s scale (an over-count)
Bezos’s ~9% stake~$20BEvidence: ~9% beneficial ownership12; equals about one-tenth of  ~$200B fortune.

Note: Table 1 labels each row as evidence (a sourced figure), assumption (a value that I chose that deliberately is tilted to overstate the labor transfer), or derived (a calculation from rows above). The operations share (75%, ~1.167M of ~1.556M) is inferred from Amazon’s reported concentration of headcount in fulfillment and logistics, since the 10-K does not break out an exact operations figure. The result is a ceiling, not an estimate.

One objection deserves pre-emption: money is fungible, so attributing the fortune to “cloud-computing profits rather than wage savings” is accounting fiction. Agreed, which is why the bound concerns magnitude, not attribution. I do not claim the suppressed wages ended up in some other line of the business, like AWS profits; since money is fungible, the tracing runs nowhere. Rather, I claim the pool of plausibly suppressed wages, summed as generously as the arithmetic allows, is an order of magnitude too small to be the engine.

Third-Party Sellers: The Part the Question Forgot

The third-party marketplace is where the extraction story has real teeth, which makes it remarkable that the question does not ask about third-party sellers. More than sixty percent of units sold on Amazon now come from independent third-party sellers, and Amazon’s effective take of a typical seller’s revenue rose from roughly forty percent five years earlier to more than half by 2023, through referral commissions, fulfillment and storage fees, and advertising that has drifted from optional to existential; seller services bought Amazon $156 billion in 2024, a quarter of its revenue.13

The mechanism that converts fees into harm is price coherence. Amazon retired its formal price-parity clause in 2019; however,  the FTC alleges that its Fair Pricing Policy and Buy Box suppression produce the same constraint: a seller who prices lower on Amazon loses visibility on Amazon, so the fee is not absorbed but capitalized into the seller’s prices everywhere.14 This, not below-cost pricing at the checkout, is the strongest form of Lina Khan’s celebrated 2017 argument that Amazon’s harm hides where consumer-price antitrust cannot see it. It is the heart of FTC v. Amazon, which survived the company’s motion to dismiss, meaning the mechanism has cleared a court’s threshold of legal plausibility, though not yet a verdict.15

But a take rate is not a transfer rate. Fulfillment, storage, shipping, and ad placement are real services with real costs, and payment for services at competitive prices is exchange, not extraction. The transfer is only the rent: the excess above competitive service pricing that platform lock-in makes collectible. The decomposition matters because the accounts suggest most of the fee is cost. Amazon’s North America segment, which contains the entire marketplace and its high-margin advertising business, earned only a roughly six percent operating margin, while the company’s profit concentrates overwhelmingly in Amazon Web Services: fifty-eight percent of 2024 operating income on seventeen percent of revenue.16 Were Amazon’s logistics priced far above cost, that margin would be higher. It is not.

Table 2 bounds the channel from both ends. At an implausible ceiling, with half of all seller fees treated as pure rent, the capitalized transfer would rival the fortune itself, which is precisely why the assumption fails: it requires Amazon’s logistics arm to be a hidden profit machine that its own segment accounts contradict. At a realistic rent share of roughly fifteen percent, the seller transfer capitalizes to about eleven percent of Bezos’s fortune: the largest transfer channel, comfortably ahead of labor. The sensitivity deserves honest disclosure, because this is the essay’s most assumption-dependent number, and if the FTC proves its case, the rent share, and with it the transfer, rises.

Table 2. Seller-Rent Bound (2024 basis)

StepImplausible ceilingRealistic estimateBasis
Seller-services revenue$156B$156BEvidence: 2024 seller-services revenue17
Rent share of fees50%~15%Assumption
Seller-paid advertising~$17B~$7BDerived
Annual seller rent~$95B~$30BDerived
Capitalized at Bezos’s ~9% stakeWould rival the fortune~11% of the fortuneDerived
Verdict on assumptionContradicted by segment marginsCentral estimate; rises if FTC prevails

Note: The segment margin evidence is the basis for rejecting the 50% ceiling.

The Cleanest Losses of All

Yet the parties who unambiguously lost, zero-sum, dollar-for-dollar, appear nowhere in the question: Borders, Circuit City, Toys “R” Us, and the rents and shareholders of an incumbent retail sector that Amazon’s gale leveled. Schumpeter named the process creative destruction and insisted that the creation and destruction are inseparable: the same gale that buried those incumbents delivered the lower prices and wider selection the surplus is made of.18 The consumer gain and the incumbent’s grave are one event, not two. Their absence from the question is proof of the misframing: it hunts for victims among the two groups that mostly gained and never glances at those who actually paid.

Verdict

Tally the channels. Customers: no expropriation at the point of sale; their genuine exposure is indirect and through seller-fee pass-through. Employees: a markdown conceded in direction, near a twentieth of the fortune on realistic assumptions and a tenth at a deliberately hostile ceiling. Sellers: the dominant transfer, about an eleventh on realistic assumptions, rising if the FTC prevails. Competitors: clean, large, and unasked-about. Summed against $200 billion, the transfers plausibly account for one-sixth. The remaining five-sixths is surplus that did not exist before Amazon.

So: both, though the components differ by nearly an order of magnitude, and the larger transfers fall outside the question’s answer choices. One final discipline keeps the conclusion honest: everything above is a claim about the fortune’s source, not a verdict on Amazon’s justice. Warehouse injury rates, productivity surveillance, fee escalation, and parity pressure may merit condemnation, regulation, or a lost lawsuit; condemning them does not relocate the origin of the money. Bezos got rich mostly by creating value for the very parties named as his victims, and at a real expense to two parties, rivals and sellers; the question was never thought to be asked. The flaw was never in the answers: it was in the options.

Leave a Reply

Trending