Amazon DSP Adds ChatGPT as The Trade Desk Shrinks: How to Choose a DSP Now

If you buy programmatic media, your demand-side platform has probably felt like plumbing: chosen once, renegotiated at renewal, rarely questioned in between. Two announcements a week apart in early September give you a reason to question it now. Amazon's DSP began piloting ads inside ChatGPT, and The Trade Desk said it would cut about 15% of its workforce.

What happened between September 3 and September 10

On September 3, The Trade Desk announced a plan it calls an organizational realignment. Its 8-K filing, published the next day, puts the reduction at approximately 15% of the total workforce, with severance and benefits charges of about $39 million to $51 million. CFO Dive, citing the 10-K, reports 3,843 full-time employees at the end of 2025. In an email to employees, CEO Jeff Green called the company "very healthy," with about $1.5 billion in cash and no debt, and said he wants teams organized into smaller, more focused groups.

On September 4, S&P Dow Jones Indices said The Trade Desk will move from the S&P 500 to the S&P SmallCap 600, effective September 21, as part of its quarterly rebalance. PPC Land reports second-quarter revenue of $715 million, up 3% year on year, and third-quarter guidance of at least $650 million, roughly 12% below the same quarter of 2025.

On September 10, Amazon Ads announced that select U.S. advertisers are testing ChatGPT Ads through it in a pilot. According to Search Engine Land, the inventory is sold through Amazon DSP as a managed service and bought on a cost-per-click or CPM basis. Ads appear as sponsored text and image units beneath organic ChatGPT responses, alongside product feed ads built from advertisers' catalogs. OpenAI continues to control delivery and placement through its own systems. Marketing Dive notes that ChatGPT runs ads on its free and Go tiers, and that Delta Vacations is among the early participants.

Digiday frames the deal as the latest in a supply-chasing streak. Over roughly 18 months, it reports, Amazon's ads business has secured access to inventory from Netflix, Roku, Spotify, SiriusXM, Disney, Hulu and ESPN. Digiday also reports that OpenAI has plugged in Criteo and StackAdapt and cut deals with the major agency holding companies, so Amazon is not the only route in.

Why this is a supply and data story

It is tempting to read the two events as cause and effect. Neither the filing nor Green's email, as reported, attributes the cut to Amazon. What the pair does show is how DSPs now compete. The independent DSP's case has rested on neutrality: it owns no media, so it has no reason to favor any seller. Amazon's case is the opposite. It brings its own shopping and streaming signals, a lengthening list of premium publishers it deals with directly, and low fees.

PPC Land, citing Guideline data, reports that Amazon moved from under 10% to just under 20% of measured programmatic spend in roughly fifteen months, while The Trade Desk grew at about the market rate. If that trend holds, more of the inventory buyers compete for will be sold through deals a particular platform holds, rather than through an open auction any DSP can bid into.

ChatGPT makes the point sharper. Search Engine Land describes Amazon's role as the buying and campaign-management layer, with OpenAI deciding how ads are served. On that description, your DSP is not choosing placements or deciding delivery there. What it adds is access and campaign management.

The fee figures, and why they are not like-for-like

Two sets of fee figures are circulating, from different reports, and they measure slightly different things.

These are reported figures, not rate cards, and we have not verified them. Green's objection has a mechanism behind it: when a platform also sells or packages the media, part of its margin can sit inside the media price rather than in a visible platform fee. A headline fee near zero tells you little until you know what the media cost and who kept the difference.

What to do before your next DSP review

Choosing a DSP

• List the publishers and channels your plan depends on, then ask each DSP which it reaches through a direct deal, which through a reseller and which not at all.
• Separate the data from the pipes. Ask which signals can be applied to each piece of inventory, not to the platform in general.
• Keep a second DSP live on a modest budget. Running the same deal or audience through two platforms for a fixed period lets you compare cost per outcome instead of claimed fees.

Negotiating fees

Planning conversational inventory

Treat ChatGPT as a test line, sized so you can learn from it without needing it to work. The reporting described so far is aggregated: impressions, clicks, cost per result, CPM and CPC. That is enough to manage pacing but not to prove incrementality, so set up measurement outside the platform first: consistent UTM tagging, conversions recorded in your own analytics, and a holdout or geographic split where volume allows.

Product feed ads build their assets from your catalog, so feed quality becomes creative quality. For advertisers in the EU, UK and Gulf, there is nothing to buy through this route yet. Search Engine Land lists expansion beyond the initial U.S. pilot as an open question.

The questions still open

• Targeting data. Amazon's announcement describes its browsing, shopping and streaming signals as a general Amazon Ads capability and does not say they target ads inside ChatGPT. Delta Vacations' president talks about using consumer insights to shape how and when the brand appears, which suggests a role for data but is not a specification.
• Measurement. Search Engine Land and Marketing Dive both describe aggregated performance data. Neither mentions log-level data, user-level conversion paths or third-party verification.
• Control. With OpenAI deciding delivery and placement, frequency and which conversations an ad appears beside appear to sit with OpenAI, not with your DSP settings.
• Maturity. OpenAI's ad operation is still being built. One analyst quoted by Digiday put it bluntly:

None of this argues for moving budget in September. It argues for asking your DSP different questions: which supply do you hold, which data can you apply to it, and what does the media cost before your fee. The wider paid media mix is being judged the same way: who owns the audience, and who owns the data about it.