Amazon Prime and the Subscription Flywheel

The Amazon Prime subscription flywheel is the hub of Amazon’s whole business, and the membership fee is the least interesting number on the page.

The subscription line that everyone associates with Prime, $49,619M in FY2025 (Amazon Form 10-K, FY2025), is the smallest of the three engines it powers. Its real job is to lift purchase frequency, which feeds third-party seller services ($172,162M) and advertising ($68,635M, up about 22%), while AWS ($128,725M revenue, $45.6B operating income) funds the machine and its AI capex (Amazon 10-K FY2025; Amazon FY2025 results).

That is the whole structure in one sentence: Prime sells frequency, frequency sells seller services and ads, and AWS pays for the infrastructure that makes all of it possible. The membership fee is the entry token, not the prize.

This piece reads that flywheel through Amazon’s own filings, not the press releases about Prime Day. Every number below ties to a specific SEC filing and fiscal period. The framing is analytical, not advisory: this is how to think about the position, not what to do about the stock.

Key takeaways

• The Prime “Subscription services” line was $49,619M in FY2025, the fifth-largest of seven reported revenue lines and the smallest of the three service engines (Amazon Form 10-K, FY2025).
• Third-party seller services ($172,162M) and advertising ($68,635M) both dwarf the subscription line. They monetize the frequency Prime creates (Amazon 10-K, FY2025).
• Advertising grew about 22% year over year, from $46,906M (2023) to $56,214M (2024) to $68,635M (2025), and is now larger than the Prime fee line (Amazon 10-K, FY2025).
• AWS supplied $45.6B of about $80B total operating income in FY2025, more than half the profit on under a fifth of consolidated sales (Amazon FY2025 results).
• Free cash flow fell from about $38B to about $11B on a roughly $50.7B year-over-year increase in property-and-equipment purchases, the AI build-out AWS profit absorbs (Amazon FY2025 results).

The thesis: the fee is the smallest engine

Most coverage of Prime fixates on the wrong line. It asks how many members there are and what the annual fee buys them. That treats Prime as a subscription product, the way Netflix is a subscription product.

It is not. Netflix collects a fee and delivers the thing you paid for. Prime collects a fee and changes your behavior, and the changed behavior is where the money is.

A Prime member shops more often, consolidates more of their spending into one storefront, and does it with less price comparison. That higher frequency is the asset. It is what makes Amazon’s marketplace worth selling into, and what makes Amazon’s ad inventory worth buying.

So the $49.6B subscription line is not the payoff. It is the cover charge that turns on two much larger meters: seller services and advertising. Read the disaggregated revenue table and the ordering tells the story.

What is the Amazon Prime subscription flywheel?

The Amazon Prime subscription flywheel is the loop in which the membership fee lifts how often a member buys, and that frequency is then monetized twice, once through third-party seller fees and once through advertising, while AWS profit pays for the underlying infrastructure (Amazon 10-K, FY2025; FY2025 results). The fee is an input, not the output.

Here is the loop in plain language, with each arrow anchored to a revenue line.

Prime lifts frequency. A member with free, fast shipping and bundled content shops Amazon more often and defaults to it before checking elsewhere. The fee buys behavior, not margin.

Frequency feeds the marketplace. More member purchases mean more demand flowing through third-party sellers, who pay Amazon for fulfillment, referral fees, and logistics. That is the $172,162M third-party seller services line, the single largest service revenue category outside AWS (Amazon 10-K, FY2025).

Frequency also feeds advertising. Every search a member runs is purchase intent at the moment of decision. Sellers pay to be the answer. That is the $68,635M advertising line, growing about 22% year over year (Amazon 10-K, FY2025).

AWS funds the machine. The retail and logistics infrastructure, and increasingly the AI capex, is paid for out of cloud profit. AWS produced $45.6B of operating income in FY2025 against $39.8B in 2024 (Amazon FY2025 results).

The arrows only run one way at the hub. Prime does not need to be profitable on its own fee, because its output is frequency, and frequency is monetized twice over: once by the marketplace and once by the ad system. The subscription is a loss-leader for two higher-margin meters.

        ┌─────────────┐
        │    PRIME     │  ($49.6B fee, the cover charge)
        └──────┬──────┘
               │ lifts
               ▼
     ┌───────────────────┐
     │ PURCHASE FREQUENCY │
     └────────┬──────────┘
              │ feeds
      ┌───────┴────────┐
      ▼                ▼
┌────────────┐   ┌──────────────┐
│ 3P SELLER  │   │ ADVERTISING  │
│ SERVICES   │   │  $68.6B      │
│ $172.2B    │   │  (+22%)      │
└────────────┘   └──────────────┘

      ┌──────────────────────┐
      │        AWS           │  $128.7B rev / $45.6B op income
      │ funds capex + infra  │
      └──────────────────────┘

Figures: Amazon 10-K, FY2025 (revenue lines); Amazon FY2025 results (AWS operating income).

That diagram is the entire argument. The hub spins; the two large meters run off it; AWS pays the electric bill.

What the filings actually say

Strip away the narrative and look at the structure of the business. Amazon’s most recent full year, fiscal 2025, posted consolidated net sales of $716,924M, up 12% (Amazon Form 10-K, FY2025).

The line that matters is not the total. It is how the total breaks apart. Here is the FY2025 disaggregated revenue, ordered as the filing reports the components (Amazon 10-K, FY2025):

Revenue line (FY2025)	Amount	Role in the flywheel
Online stores	$269,287M	First-party retail; the gravity well
Physical stores	$22,561M	Grocery and in-person footprint
Third-party seller services	$172,162M	Fees on the marketplace Prime feeds
Advertising services	$68,635M	Monetizing the purchase intent
Subscription services	$49,619M	Prime fees plus digital content
AWS	$128,725M	The compute engine that funds capex
Other	$5,935M	Miscellaneous

Source: Amazon.com, Inc., Form 10-K, fiscal year ended December 31, 2025.

Notice the ranking. Subscription services, the line people think of as “Prime,” is smaller than third-party seller services, smaller than AWS, and only modestly larger than advertising. The fee is the fifth-largest of the seven reported lines.

Per the 10-K, “Subscription services” includes Amazon Prime membership fees plus certain digital content subscriptions (Amazon 10-K, FY2025). So even that $49.6B is not pure membership. The headline Prime number is smaller still, and it is the least of Amazon’s worries.

How is the Amazon Prime flywheel monetized twice?

The same purchase intent is sold two ways: once as a transaction fee on the seller side, and once as an ad impression on the search side. A member’s order generates seller fees on the marketplace; the same member’s search generates a paid placement before they even click. Both scale with frequency, not with the fee.

The fastest way to see why the fee is the least interesting number is to score the three service engines on one card. Call it the Flywheel Ledger: a single view that ranks each engine by size, growth, and the job it does in the loop. It is the asset to cite when someone argues Prime is “a subscription business,” because the ledger shows the subscription line is the smallest and slowest of the three (Amazon 10-K, FY2025).

The Flywheel Ledger

Engine (FY2025)	Revenue	2023	2024	Approx. YoY	Job in the loop	Margin character
Third-party seller services	$172,162M	—	—	—	Monetizes member orders	Fee on volume
Advertising services	$68,635M	$46,906M	$56,214M	+22%	Monetizes member searches	Structurally high
Subscription services	$49,619M	$40,209M	$44,374M	+11.8%	Buys the frequency	Thin to negative

Source: Amazon.com, Inc., Form 10-K, fiscal year ended December 31, 2025. Dashes indicate prior-year comparatives not used in this analysis. Margin character is a qualitative reading of the engines, not a disclosed figure.

Read the ledger top to bottom and the structure is obvious. The two engines that monetize the member ($172,162M and $68,635M) sit above the engine that pays to acquire the member ($49,619M). The line everyone calls “Prime” is the one doing the least direct earning. Two readings jump out.

Advertising is now the fast engine. It grew from $46,906M (2023) to $56,214M (2024) to $68,635M (2025), about 22% year over year, and it has passed the subscription line in absolute size (Amazon 10-K, FY2025). The ad business monetizes the same purchase intent that Prime frequency creates, and it does so at structurally higher margin than shipping a box.

This is the same move Apple runs through its Services line, where a high-margin layer is bolted onto an installed base built by a lower-margin hardware product. The mechanics differ, hardware versus a membership fee, but the logic is identical: build the base cheaply, monetize it richly, as laid out in Apple Services as the margin engine inside iPhone. Amazon’s version puts the high-margin layer (ads) on top of frequency rather than on top of devices.

Subscription is the slow, steady engine. It grew from $40,209M to $44,374M to $49,619M, about 11.8% year over year (Amazon 10-K, FY2025). Healthy, but it is the smallest and slowest of the three, and it is doing a different job: it is the input that makes the other two grow.

The advertising line growing faster than the subscription line is the flywheel working. Frequency, bought with the fee, is being monetized more efficiently by ads than by the fee itself. This is the same pattern operators chase when they shift from fixed billing toward revenue that scales with use, the dynamic explored in usage-based pricing versus seat-based pricing.

Where the profit and the capex actually sit

Revenue mix tells you what spins the wheel. Operating income tells you what pays for it. They are not the same segments.

Amazon reports its business in three segments: North America, International, and AWS. Here is where FY2025 operating income lands (Amazon FY2025 results):

Segment (FY2025)	Net sales	Operating income
North America	$426.3B	$29.6B
International	$161.9B	$4.7B
AWS	$128.7B	$45.6B

Source: Amazon.com, Inc., FY2025 segment results, fiscal year ended December 31, 2025.

Read the AWS row against its sales. AWS generated $45.6B of operating income on $128.7B of sales, while North America produced $29.6B on $426.3B (Amazon FY2025 results). The smallest-revenue segment is the largest-profit segment by a wide margin.

Total FY2025 operating income was about $80B, an 11.2% operating margin, up from 10.8% in 2024 (Amazon FY2025 results). AWS supplies more than half of that operating income on under a fifth of the sales. That is what “AWS funds the machine” means in numbers: the retail flywheel runs at thin margins, and cloud profit subsidizes the infrastructure underneath it. The same gross-margin gravity decides which businesses can carry heavy spend, the logic in why gross margin is destiny in SaaS.

The capex confirms the dependency. FY2025 free cash flow fell from about $38B to about $11B, driven by a roughly $50.7B year-over-year increase in property-and-equipment purchases, the AI and infrastructure build-out (Amazon FY2025 results). The profit pool that absorbs that spend is AWS. Without the cloud margin, the capex bet would be far harder to carry.

The frequency lever: an illustrative model

The filings do not disclose per-member purchase frequency or the dollar value of a single incremental order. So the next step is a model, and it is labeled as one.

The point is to show the mechanism, not to claim a precise figure. The following uses illustrative, hypothetical numbers, chosen only to demonstrate how frequency, not the fee, drives the economics.

Suppose a member pays an annual fee that nets Amazon a small margin after the cost of shipping and content. On the fee alone, the member is roughly break-even or a modest loss. Now suppose that same member, because they are a member, places more orders per year than they otherwise would, and a slice of those orders flow through third-party sellers who pay fees, and a slice of their searches surface a paid placement.

In that model the fee is the smallest contributor to member lifetime value. The seller fees on incremental marketplace orders and the ad revenue on incremental searches are the larger contributors, and they scale with frequency, not with the fee. Double the frequency and you barely change the subscription line while you meaningfully grow the seller and ad lines.

That is the whole reason Prime can hold its price flat and still get more valuable: the value is downstream of the fee, in lines the member never sees on a receipt.

Methodology: how to read the frequency model

• Inputs: the disaggregated revenue lines (subscription $49.6B, third-party seller services $172.2B, advertising $68.6B) and their growth rates, all from Amazon’s 10-K, FY2025.
• Assumptions: that membership materially raises purchase frequency, that a portion of incremental orders are third-party (generating seller fees), and that incremental searches generate ad impressions. Each is consistent with the filing structure but not separately quantified in it.
• Sensitivity: if membership lifts frequency only marginally, the flywheel weakens and the fee carries more of the burden. If frequency response is strong, the seller and ad lines dominate member economics and the fee becomes almost irrelevant to the math.
• What this misses: Amazon does not disclose member count, per-member order frequency, or the marketplace-versus-first-party split of member orders in the cited filings. The model is directional, not a measured result, and the illustrative figures are not Amazon disclosures.

Where this is genuinely vulnerable

A credible analysis names the holes. There are three.

The frequency assumption is inferred, not disclosed. The flywheel rests on the claim that membership lifts purchase frequency enough to drive seller fees and ads. That is highly plausible and consistent with the revenue structure, but the cited filings do not isolate member frequency. The argument is a reading of the mix, not a measured causal chain.

Advertising growth could be borrowing from retail margin. Ads grew about 22% to $68,635M (Amazon 10-K, FY2025), and ad load on a storefront is not free. Past some point, more sponsored placements degrade the search experience that creates the purchase intent in the first place. The fast engine and the frequency engine can come into tension, and the filings do not show where that line is. It is the same ceiling that ad-supported video runs into, the tension mapped in why the Netflix ad tier changes everything.

The capex bet is unhedged and AWS carries it. Free cash flow fell from about $38B to about $11B on a roughly $50.7B increase in property-and-equipment purchases (Amazon FY2025 results). That spend is a wager on AI and infrastructure demand, and the profit pool absorbing it is AWS. If AWS operating income growth stalls, the capex load becomes a multi-year drag, the same structural risk that runs through every hyperscaler’s balance sheet right now. The cloud-margin side of that bet is the subject of AWS margin pressure and the cloud reset.

None of these is fatal on today’s evidence. All three are why this is a flywheel under load, not a perpetual-motion machine.

The bear case: what the skeptics get right

The strongest objection to the whole framing is simple: the flywheel is a story imposed on a revenue table, and a skeptic can read the same lines a different way.

The subscription line is not actually small. $49,619M is the fifth of seven reported lines, but in absolute terms it is larger than most public companies’ entire revenue (Amazon 10-K, FY2025). Calling it “the least interesting number” is a rhetorical device. A line growing about 11.8% a year, durable and recurring, is exactly the kind of revenue investors prize. The bear says: do not talk yourself out of valuing the most predictable engine just because two others are bigger.

There is weight here. Recurring revenue carries a higher quality of earnings than transactional fees, which is precisely why the market rewards it. The flywheel framing should not be read as “the fee does not matter.” It is “the fee’s job is upstream of where it gets counted.” Both can be true.

The two large engines may not depend on Prime as much as the loop implies. Third-party seller services and advertising would exist even with a weaker membership. Plenty of non-members shop Amazon; plenty of ad impressions serve them. The filings do not isolate how much of the $172,162M seller line or the $68,635M ad line is member-driven (Amazon 10-K, FY2025). If the member contribution is smaller than assumed, the “fee buys the frequency that powers everything” claim is overstated, and Prime is closer to one product among several than the hub of the wheel.

This is the most honest hole in the thesis, and it is the same one named in the vulnerability section above: the frequency link is inferred, not disclosed. The flywheel is a strong reading of the structure, not a measured causal chain.

AWS dependency cuts both ways. The bull says AWS profit funds the machine. The bear says that makes Amazon a cloud company wearing a retail costume, and the retail flywheel is a low-margin appendage kept alive by one segment. On that reading, the interesting analysis is AWS, and Prime is a distraction. The counter is that the retail surface is what creates the ad inventory and the marketplace in the first place, so the segments are not separable. But the bear is right that if AWS stumbles, the whole structure gets re-priced fast.

Weighing it: the bear case does not break the flywheel, it bounds it. The fee matters more than the framing’s sharpest phrasing admits, the member contribution to the big engines is unquantified, and the AWS dependency is real. The thesis survives as “frequency is the asset and it is monetized downstream,” not as “the fee is irrelevant.” Held to that, it holds.

What operators should take from this

If you build software, the transferable lesson is not “launch a membership program.” It is the ordering underneath it.

Amazon is demonstrating, at the largest possible scale, that the headline subscription line can be the least important revenue you have, if it buys behavior that monetizes elsewhere. The fee is an input. The output is frequency, and frequency is what you actually sell.

Build your own Flywheel Ledger. Take your revenue lines and tag each one with its job (acquires behavior, or monetizes it) and its margin character, exactly as the table above does for Amazon. The moves below fall out of that exercise.

• Separate the acquisition line from the monetization line, on paper. Before optimizing pricing, label every revenue line as either “buys behavior” or “monetizes behavior.” If your subscription is doing acquisition work, stop measuring it on its own margin. Amazon runs Prime at thin or negative fee economics on purpose; the return shows up two lines over.
• Ask what behavior the fee buys, not just what it collects. A founder reflexively prices a subscription to be profitable. The Prime logic says price it to maximize the downstream behavior, usage, frequency, attach rate, that a higher-margin line can monetize. The fee is a lever on the meters, not the meter.
• Monetize the same intent twice where you honestly can. Amazon sells a member’s intent once as a seller fee and once as an ad placement (Amazon 10-K, FY2025). If your product generates purchase or usage intent, look for a second, structurally higher-margin way to monetize the same event rather than raising the price of the first.
• Know which line is the engine and which is the funder, and never confuse them. AWS pays the bills while retail spins the wheel (Amazon FY2025 results). Map your own version: the fastest-growing line is rarely the one that should be funding the capital-heavy bets. Confusing the two leads founders to starve the engine to protect the funder.
• Watch the ad-load tradeoff before you chase it. Amazon’s ad line is the fast engine, but the bear case flags that more sponsored placements can degrade the experience that creates the intent. If you add a high-margin monetization layer on top of behavior, instrument the behavior for decay, not just the new revenue. A monetization layer that quietly erodes its own input is a slow leak.
• Treat recurring revenue as quality, not just size. The bear case is right that a durable, predictable line has value beyond its rank in the table. Use the ledger to see the structure, but do not talk yourself out of valuing the steadiest engine you have.

This is the same prioritization that runs through Google’s playbook, where the surface, not the product, is the moat, the argument in Google’s AI strategy as a distribution war. Amazon’s version: the fee, not the membership, is the cover charge, and the meters that run off it are where the business is.

How the pieces of the flywheel fit together

Amazon’s flywheel is not one bet. It is a stack of reinforcing ones:

1. Sell Prime to lift purchase frequency, accepting thin or negative margin on the fee itself.
2. Monetize that frequency through third-party seller services, the $172.2B line that dwarfs the subscription fee (Amazon 10-K, FY2025).
3. Monetize the same purchase intent again through advertising, the $68.6B line growing about 22% (Amazon 10-K, FY2025).
4. Run the retail flywheel at thin segment margins, knowing AWS carries the profit.
5. Use AWS’s $45.6B operating income to fund the AI and infrastructure capex that the retail margins could not (Amazon FY2025 results).

The companies treating Prime as a subscription business are reading the smallest number on the page. The number that decides the business is frequency, monetized twice, and paid for by cloud.

That is the whole structure. The fee is just the turnstile, and the meters that run off it, not the membership, are where the business lives.

---

Analysis, not investment advice. Figures are drawn from Amazon.com, Inc.’s public SEC filings (Form 10-K and FY2025 results, fiscal year ended December 31, 2025) and cited inline by fiscal period. Frameworks here are for understanding business strategy and tradeoffs, not for making buy or sell decisions.

Want the full toolkit for reading filings like this, the segment-margin worksheet, the flywheel-mapping framework, and the Flywheel Ledger template used above? It’s in the Tech Business Analysis Playbook.