Amazon Ads AI Platform Claims 18% Lower CPMs in Beta Tests

Full analysis

Amazon is merging its ad console and its DSP (the software advertisers use to buy ads programmatically) into one AI-driven platform, and claiming beta advertisers got 18% lower CPMs, 16% lower cost-per-acquisition, and 65% better ad delivery. The pitch, from Amazon Ads VP Kelly MacLean, is "programmatic for everyone": launch a cross-channel campaign in four clicks, powered by Amazon's identity graph that she says reaches 90% of US households.

This is hard to undo for the people it threatens. Amazon isn't deciding anything here. The decision lands on everyone else: agencies with managed-service teams, independent SSPs and DSPs, and identity vendors who sell the thing Amazon now gives away native. What's actually being decided is where performance budget clears, and whether the open programmatic stack keeps its cut. No hard deadline, but general availability sets the clock, and holdco budget reviews heading into the 2027 upfront are when the rate pressure gets real.

The Market Analyst. Amazon Ads is already a roughly $56 billion business growing faster than Google's ad network. A tool that visibly lowers advertiser costs while improving delivery pulls budget that would otherwise clear through independent SSPs (the sell-side software that connects publisher inventory to buyers). That squeezes Magnite, PubMatic, and Index Exchange on the open web, and it squeezes The Trade Desk, the largest independent DSP, on the buy side. In plain terms: the toll booth on the open internet gets less traffic as Amazon paves its own road. The catch is scope. Amazon's edge is purchase data on retail and CPG. For brand and awareness money, where nobody is buying soap off a click, that graph does far less work.

The Skeptic. Eighteen percent lower CPMs from a beta of self-selected advertisers is selection bias reported as proof. The buyers who opt into a four-click tool are small and mid-tier accounts that were already badly optimized. Of course their numbers improved. That tells you nothing about the sophisticated multi-DSP buyer whose campaigns are the real benchmark. Real complexity, audience suppression, frequency caps across walled gardens, incremental reach, does not collapse into four clicks without trade-offs that never show up in a launch deck. And "90% of US households authenticated" is a claim that needs a methodology, not a headline. For a general reader: Amazon measured the easiest cases and reported the win.

The Operator. The four-click flow goes straight at managed-service headcount. First to feel it: the junior trader and campaign manager doing bid strategy, pacing, and cross-channel sequencing by hand. Within 90 days of general availability, agencies lose the complexity story that justifies their fee on Amazon line items. Clients will see "four clicks" and ask why they're paying a managed rate. Planners either move up into strategy, creative, and measurement, or they eat the rate pressure. The practitioner question for Tuesday morning: is your Amazon desk billing for work the console now does for free? Finance will notice the margin before strategy admits it.

The Customer / End User. Here the advertiser is the customer, and the trade is real. You hand Amazon the "why" and the "what," Amazon runs the "how," and in return you give up the ability to see and tune the machine. Lower CPMs are nice. But the advertiser also loses the leverage that comes from being able to walk across DSPs and play them off each other. You optimize against Amazon's purchase signals, inside Amazon's graph, measured by Amazon's attribution. The CPG brand already living on Amazon shrugs and takes the savings. The brand marketer who needs cross-platform reach should read the exclusivity of the setup before celebrating the rate.

The CFO. Free is never free. Amazon's platform is buying something with that 18%: deeper dependence and your campaign data flowing through its closed loop. The line-item savings are real and immediate. The cost is structural and shows up later, when Amazon is your planning layer, your buying layer, and your measurement layer, and the take rate on all three is whatever Amazon decides next year. The payback math looks great in the pilot. At scale, the bargaining power sits entirely on one side of the table, and it isn't yours.

Where they disagree. The Skeptic says the 18% won't survive contact with real campaigns, so there's no structural shift, just a good demo. The Market Analyst and the Operator say it doesn't matter whether 18% holds exactly: even a smaller, real improvement, attached to a four-click flow, moves budget and headcount. The second split is about reach. The Market Analyst thinks the damage is confined to retail and CPG performance money, where Amazon's purchase data actually helps. The Strategist view in the room says Amazon absorbs performance budget wholesale and exiles the open web to brand spend. Both can't be fully right.

What it hinges on. Two things. First, does the CPM improvement generalize beyond underoptimized Amazon-native accounts to the buyers who matter, or does it fade as campaign complexity rises? Second, how far does the purchase-signal graph travel past retail and CPG? If it stays mostly a commerce tool, the open programmatic stack loses a slice and survives. If performance budget broadly decides that Amazon's closed loop beats stitching together a DSP, an SSP, and an identity vendor, the independent stack retreats to brand and awareness, and the identity middlemen get squeezed from both sides. The council leans toward real but bounded: budget moves where Amazon's commerce data is relevant, and that's enough to pressure the independents without ending them. Before anyone reprices their Amazon desk on the 18% headline, match the beta cohort against your own campaign mix and demand the identity methodology.

Prediction: PubMatic's full-year 2026 revenue, reported in February 2027, will be lower than its 2025 revenue.

Confidence: Medium — Amazon pulling performance budget off the open web is one of several pressures, but it points the same direction.

Why: Amazon merging its console and DSP into a four-click tool with lower CPMs is built to keep retail and CPG performance money inside its own graph, which is exactly the budget that otherwise clears through independent sell-side platforms like PubMatic. PubMatic monetizes open-web impressions; when Amazon captures the buyer and the inventory in one closed loop, those impressions never reach PubMatic's exchange. The opposite case, that PubMatic grows, would need open-web programmatic to hold its share of performance spend at the same moment Amazon is explicitly engineering a cheaper, simpler alternative, which runs against where the easy budget is flowing. The signal here is Amazon targeting the exact "no deep expertise" buyer PubMatic depends on volume from.

Revisit by 2027-02-28: We're right if PubMatic's reported full-year 2026 revenue comes in below its 2025 full-year revenue. We're wrong if it comes in at or above 2025.