I was on Eric Seufert's Mobile Dev Memo podcast last week for the third time. Eric titled the episode "Understanding Amazon's Advertising Advantages." Below is my attempt to distil the key themes into something useful for anyone building in or buying through the Amazon ad ecosystem.

The $100 Billion Question

Amazon's advertising business generated just shy of $70 billion in 2025, growing at greater than 20% year over year. Forecast that out and you get to somewhere north of $80 billion in 2026. There is a clear path for Amazon to hit $100b in advertising revenue. But how will it get there?

The two primary components of Amazon's ad business are Sponsored Ads and the Amazon DSP. Sponsored Ads are pay-per-click units that run on search results and product detail pages on Amazon.com. The DSP buys display, video, connected TV, and audio inventory across Amazon's owned properties like Prime Video, as well as third-party publishers like Disney and Netflix.

Sponsored Ads are largely tapped out. CPCs have flatlined because sellers and brands can't pay more than their margin thresholds allow, and the inventory is already fully monetised. Every product page on Amazon is already an ad placement. There's nowhere left to grow.

The DSP is the growth engine. And the DSP's growth is anchored to one thing: CTV.

The Identity Spine

At Amazon's upfront last month, the company announced that its authenticated graph now deterministically reaches 90% of US households with verified, logged-in, non-modelled data. To put that number in context: it's the result of decades of Prime penetration. Prime is the most ubiquitous loyalty programme in the US. Every household that signed up, every Prime Video subscription, every Alexa device in a living room, built the foundation of an identity graph that Amazon has been quietly assembling for years. The Roku partnership announced last summer enhanced that spine by closing the remaining gaps. Roku is the largest OEM in connected TV hardware, and combining the two data sets gave Amazon deterministic reach into households it couldn't identify on its own.

To understand why this matters, you have to think about how large TV media buyers plan their campaigns. The two variables they optimise for are reach and frequency. Specifically, deduplicated reach and frequency across every TV buy they make. To manage that effectively across linear, streaming, and CTV inventory, you need a consistent identity layer that tells you when you've reached the same household twice.

Amazon's pitch to a brand or agency is simple: centralise your TV buys on the Amazon DSP because the identity layer here is the highest-fidelity available anywhere. You get cleaner reach and frequency management than you'd get on The Trade Desk, DV360, or any other enterprise DSP. And the Amazon DSP is the only place to buy Prime Video ads. If Prime Video is in your plan, you're coming to Amazon regardless.

The Unfair Data Advantage, Extended

Amazon's identity advantage compounds because Amazon keeps manufacturing new data signals in categories it didn't originally own.

Take auto. Auto is one of the largest CTV advertising categories in the world. Traditionally, Amazon had only indirect signals here: household demographics, durable goods purchasing patterns. Then Amazon built Amazon Autos, digital showrooms for OEMs where consumers can compare vehicles and in some cases complete a purchase. Amazon has now inserted itself into the path to purchase for auto in a way that no other DSP can replicate. The data that flows from Amazon Autos becomes a proprietary targeting signal that Amazon can use to make DSP buys in the auto category more effective.

The same logic could apply to pharma. Amazon has made significant investments in the pharma stack through One Medical and its acquisition of PillPack. Those data sets haven't been opened to advertisers. If they ever are, Amazon will have effectively created an endemic category out of pharma, another massive non-endemic vertical where Amazon holds a data advantage that nobody else can touch.

This is the broader pattern: Amazon identifies high-value advertising categories where it lacks proprietary first-party data, builds or acquires retail touchpoints in those categories, and converts the resulting data into targeting signals. It's a flywheel that gets harder to compete with every year.

The Trade Desk Problem

Three years ago, if you asked most people who Amazon's primary competitor in programmatic media was, they would have said The Trade Desk. The competitive landscape has shifted considerably since then.

Amazon and Google have spent the last few years methodically dismantling the Trade Desk's strategic rationale. Amazon's argument to every major agency is: centralise your TV buys here because we have better identity data, exclusive Prime Video inventory, and the strongest outcome measurement in the industry. Google makes a similar argument around YouTube and DV360. The Trade Desk, by contrast, doesn't offer anything to marketers that Google and Amazon don't already offer, and it owns no proprietary inventory or first-party data of its own.

As Hemingway wrote about going bankrupt: gradually, then all at once. For years, Amazon and Google were slowly chipping away at the Trade Desk's position, arming their sales teams with reasons why ad budgets should centralise to their platforms. Now we're in the "all at once" phase. CMOs and agency decision makers are increasingly asking why they're routing any dollars through the Trade Desk at all.

To their credit, The Trade Desk has built a real business. Billions in ad spend flow through it. It's still growing. But its growth rate has decelerated significantly and its CEO's response, taking public shots at Amazon with claims that are categorically false, hasn't helped. Claiming that Amazon's DSP is not a priority internally, or that Amazon pushes its own inventory over third-party supply, doesn't hold up. Amazon's advertising growth story is almost entirely a DSP story, and Amazon's sales teams have actually been incentivised to sell non-Amazon inventory because the bigger opportunity is centralising all of a client's TV budget, not just driving Prime Video placements.

The Trade Desk will survive. Log-level data access is a genuine advantage in categories like pharma, where advertisers need to match specific impressions to healthcare professional outcomes. But the trajectory is one direction, and it's been that way for two years.

The ChatGPT Variable

The most interesting unresolved question in Amazon's advertising future is a potential looming partnership with ChatGPT.

Amazon currently sits on >50% of US e-commerce volume. Any AI platform with commerce ambitions needs Amazon's catalogue, pricing data, and logistics infrastructure to offer a credible shopping experience. ChatGPT's Instant Checkout experiment, which partnered with Shopify, Walmart, and Etsy, was shut down after six months, with Walmart subsequently pulling out as well. It was never going to work without Amazon. The infrastructure problem isn't about merchant selection, it's about the trust and seamlessness that Amazon’s ecosystem provides, and nothing else at that scale can replicate it.

ChatGPT needs Amazon.

Amazon, for its part, is navigating a disruption dilemma: the same AI-native shopping behaviours that ChatGPT enables could compress the browsing and clicking patterns that fund Amazon's $70 billion Sponsored Ads business. Fewer page views mean fewer ad impressions, preseting material downside risk to the foundation of Amazon’s advertising business.

The most likely resolution is an advertising partnership on Amazon's terms. Amazon extends its DSP demand and identity spine onto ChatGPT's inventory. ChatGPT gets access to Amazon's catalogue and monetisation infrastructure without having to build it from scratch. Amazon gets a new ad surface that offsets potential declining on-site revenue. Eric put it well on the podcast: Amazon missed social. Twitch was never the answer. The chatbot layer may be the surface area where Amazon finally gets a foothold in attention it doesn't control natively. This seems inevitable and mutually beneficial.