Amazon takes its near-zero DSP fee into OpenAI's ChatGPT ad pilot

The interesting term is the unpublished one, which is what Amazon keeps on the way through. Its DSP has been landing near zero on programmatic guaranteed deals while The Trade Desk historically took 15 to 20 per cent and Google took something close to it ^[7], so on a $10m guaranteed buy the difference is $1.5m to $2m that reaches inventory instead of a broker ^[11]. Kelly MacLean, Amazon's vp of ads, spent this year arguing exactly that, that the best supply costs less and targets better on Amazon's rails ^[8], and every deal since has been offered as evidence for the claim rather than a test of it.

ChatGPT is the eighth premium name Amazon has attached in roughly 18 months, after Netflix, Roku, Spotify, SiriusXM, Disney, Hulu and ESPN ^[6][12], and Chris Conetta of the Amazon DSP calls conversational ads the fastest growing engagement opportunity for brands ^[3]. The mechanics differ from streaming in a way that framing skips. Ad breaks in a Netflix show are countable and rationed, while chat impressions are produced on demand, and OpenAI has already wired in Criteo, StackAdapt and the major agency holding companies ^[9], which leaves Amazon with non-exclusive access to supply nobody has to ration. That is a distribution win for OpenAI before it is a supply win for Amazon.

OpenAI's side of this is demand it cannot yet sell on its own. The arithmetic behind that 2030 number is unforgiving: tripling every year for five years is a factor of 243, which implies a base of roughly $412m today ^[13][17]. That is a believable size for a business that only started testing ads in February ^[4], and it is also the yardstick, or rather the only yardstick on offer, for integrations like this one, because deals with buying platforms are how several hundred million becomes several billion, and one named travel advertiser in a US pilot says nothing about the two orders of magnitude that follow.

Buyer tolerance is the one thing I would call settled, and Digiday puts it plainly, that advertisers have shown again and again that a platform offering performance inventory can get away with a great deal ^[16]. My read on the rest is narrower than the announcement invites, since Amazon has taken a cheap option on a channel of unproven scale and is spending nothing scarce to hold it. Two findings would move me: an exclusivity or first-look term on ChatGPT supply, which would mean Amazon paid something real and expects volume worth defending, and evidence that conversational impressions clear near streaming prices, which would make this supply worth rationing and the fee argument only part of the story.