Amazon doesn't sell you stuff. Amazon is a landlord.
Galloway picked Amazon for 2026. The chart-watchers laughed - worst Mag 7 performer, capex bleeding, free cash flow underwater.
Amazon Isn’t a Retailer Anymore.
AWS did $128.7 billion in 2025 at 35% operating margins. Advertising cleared $21 billion, growing 22%, with incremental margins north of 50%. Then there’s the rent Amazon charges the millions of third-party sellers stuck on its Marketplace - listing fees, fulfillment fees, ad placements. Plus Prime, the stickiest subscription in consumer tech.
North America retail flipped from a loss in 2022 to nearly $30 billion in operating income. Most of that isn’t from selling toasters. It’s from charging other people to sell toasters.
The Marketplace is the mall. AWS is the data center version of the same trade. Ads are the billboards. Sellers pay rent whether or not shoppers show up.
Comparing this to Walmart is a category error.
The capex “panic”
$200 billion in 2026 capex. ROIC sliding from 14.8% to 12.4%. The bears have a spreadsheet and they’re sticking to it.
Here’s what the spreadsheet misses. Amazon is the only hyperscaler whose AI infrastructure has itself as the anchor customer. Microsoft and Google rent GPUs to outsiders. Amazon uses the same silicon to run AWS, train Anthropic’s models, power its own ad-targeting, and operate its robotics fleet. One dollar of capex, four revenue streams. Depreciation hits one line. The benefits scatter.
Then there’s Trainium, Amazon’s in-house AI chip. Every Trainium chip deployed internally is a check NVIDIA doesn’t get. The business is running around $20 billion and accelerating. The market values it at roughly zero.
The Anthropic stake nobody prices
Microsoft gets enormous credit for owning a slice of OpenAI. Fine. Amazon has put over $8 billion into Anthropic, which has committed to spending $100+ billion on AWS over a decade. Anthropic is probably the second-most-valuable AI lab on Earth. Amazon’s stake is in the model at roughly zero.
That gets repriced the next time Anthropic raises. Or files. Or sells.
Kuiper isn’t a Starlink fight
1,616 satellites. The press calls it a consumer broadband play against Musk. Wrong frame.
The Pentagon has openly said it wants alternatives to Starlink. The intelligence community doesn’t love single-vendor dependence on a network controlled by one volatile billionaire. AWS GovCloud already handles classified workloads. Bolt a non-Musk satellite layer onto that stack and you have a vertically integrated sovereign comms system. Exactly the thing DoD is shopping for.
Defense-tech revenue trades at very different multiples than consumer broadband. Same satellites.
The regulatory thaw
Eighteen months ago, Lina Khan’s FTC was trying to break Amazon up. Today the case sits in purgatory. Bezos shifted the Washington Post’s editorial line, showed up at the Trump inauguration, Amazon wrote a $1 million check to the inaugural fund. The old regulatory risk premium is still in the stock. The thaw isn’t.
The robotics number
Amazon has signaled it could automate 75% of fulfillment work and eliminate roughly 600,000 jobs. Politically toxic. Morally grim. Also the largest labor-cost reduction in retail history.
Strip labor out of logistics-as-a-service and the rent Amazon charges third-party sellers gets dramatically more profitable. UPS and FedEx are the ones who should be losing sleep.
A landlord that automates its workforce, extracts rent from its tenants, and sells the leftover infrastructure to the Pentagon. Recession-proof in a way Wall Street isn’t modeling.
The setup
If you own QQQ or VGT, you already own 6-7% Amazon. The question is whether you want to be overweight the most-discounted name in an index you already hold.
Amazon lagged the Mag 7 in 2026 for mechanical reasons - capex shock, FCF compression, rotation. None of those are the business deteriorating. The business is quietly turning into something more durable than the model in everyone’s spreadsheet.
A tech ETF is the boring, correct default. Overweighting Amazon is the asymmetric one. Galloway landed on the right answer for the wrong reasons. That’s still the right answer.
Companies of interest
Amazon (AMZN) - The thesis. Utility-infrastructure-distribution platform priced like a retailer. Catalysts: Anthropic repricing, ads disclosure, regulatory thaw, robotics margins.
Microsoft (MSFT) - Cleaner version of the same hyperscaler trade. OpenAI exposure already partly in the price. Lower torque, lower risk.
Alphabet (GOOGL) - Owns its own silicon (TPUs), its own model (Gemini), and the biggest ad business on earth. The substitute if you don’t trust Amazon to execute.
Anduril (private, watch for IPO) - Defense-tech name to watch if Kuiper becomes a Pentagon play. Same patron network.
Constellation Energy (CEG) / Vistra (VST) - Picks and shovels. $200B in capex is worthless without electrons. These guys sell the electrons.
Invesco QQQ / Vanguard VGT - The boring default. Own one and go for a walk.