On May 17, Publicis Groupe announced an agreement to acquire LiveRamp for an enterprise value of $2.2 billion. That is a 29.8 percent premium to LiveRamp's prior close, with the deal expected to close by the end of 2026 pending regulatory and shareholder approval.

Read the release carefully and the framing is unusual for an agency M&A. The thesis is not media buying, not creative scale, not client roster. It is "data co-creation for smarter agents," with LiveRamp folding into CoreAI, the Publicis platform that already orchestrates audience and identity work for the holding company's brand clients. This is the first ten-figure acquisition we have seen where agentic commerce is the stated reason on page one.

Publicis did not buy a media network. It bought the data layer that sits between brands and the agents about to spend on their behalf.

What Publicis actually paid for

LiveRamp is, in one sentence, the largest independent identity graph in adtech. RampID, its persistent identifier, ties together deterministic signals (logged-in email, mobile device ID, retail loyalty) and probabilistic ones (CTV exposure, ad surface co-occurrence) across roughly every major US merchant, publisher, and ad network. If a brand wants to know that the user on their site, the viewer of a connected TV ad, and the recipient of an email are the same person, RampID is how that resolution happens.

Set that next to the agentic problem. A large language model can suggest a product to a user. It can route them to a merchant. What it cannot reliably do, today, is tell anyone downstream who that user actually is across the merchant, the network, and the ad surface that originally drove the intent. The model knows the conversation. It does not know the customer.

That is the gap LiveRamp fills. Per the Marketing Dive write-up of the deal, Publicis is explicit that the goal is feeding richer audience data into agent-mediated decisions: better targeting upstream, better attribution downstream, and a brand-owned identity layer in the middle that the agents themselves can read from.

One thing to be clear about: this is not a payments rail. LiveRamp does not move money, hold tokens, or sit in the authorization flow. It is the audience and intent layer one step before the rail. Which is precisely why it matters now.

Two weeks after OpenAI handed checkout back

Timing is the tell. In early May, OpenAI handed checkout back to merchants, retreating from its Instant Checkout product after concluding the model alone could not close the commerce loop. The gap was not technical. The model could parse intent and route a basket. What was missing was the connective tissue between the agent and the merchant's view of the customer: who they are, what they have bought before, what audience segment they sit in, what the brand is willing to pay to acquire them.

Publicis just paid $2.2 billion for that exact piece.

That sequencing is worth holding in your head. The biggest LLM company in the world walked away from owning agentic checkout end-to-end because it did not have the audience graph. Two weeks later, the largest agency holding company in the world bought one. The market is pricing the data layer as the contested middle, not the model and not the rail.

If you are an operator on either side of that middle, you should be doing the same math.

The three-way collision

Three groups now want to own the customer view at agentic checkout, and they have meaningfully different starting points.

The card networks have moved first on credential and trust. Visa's Agent Payments Protocol and Mastercard's Agent Commerce Protocol are both staking out the same claim: the agent needs a verifiable, network-issued credential to transact, and the network is the party that issues, verifies, and indemnifies it. That is identity in the narrow sense of "can this agent pay, on behalf of this person, with this card." It is the rail-adjacent layer.

The LLM platforms want something broader. OpenAI, Anthropic, and Google are each building toward the agent as the front door for commerce intent, with the model holding the conversational context that produced the purchase. Their pitch is that the model is the most natural home for audience and intent data because the model is where the intent originated. The OpenAI Instant Checkout retreat is evidence that holding intent alone is not enough.

Publicis is planting a flag in the gap between those two. Agency holding companies already own the brand budgets. They already negotiate the media buys, hold the first-party data partnerships, and sit in the room when CMOs decide which surface to spend on. Buying LiveRamp gives one of them the largest independent identity graph in adtech, on top of the largest brand client roster in the industry, on top of CoreAI as the orchestration plane. That is a different shape of bet from a network or a model, and the people it threatens know exactly which side of the gap they were not standing on.

What this means for operators

For payments operators, the credential is no longer the only identity that matters at agentic checkout. There is an audience graph competing to sit on top of it. If your roadmap assumes the network-issued agent credential is the canonical identity, it is worth pressure-testing what happens when a brand says, "use RampID, not the network token, to resolve who this user is to our CRM." That is not a hostile move. It is a parallel one, and the two will need to interoperate.

For merchants, a Publicis-owned identity graph changes the negotiation. If you are a Publicis client, the cost of using RampID for agentic flows is bundled into a relationship you already have. If you are not, it is a separate license, and the pricing power has just consolidated. Either way, the question of who owns the identity stitch between the advertising layer that agents see and the merchant's customer record is now answered: an agency holding company does, for the first time.

For adtech, $2.2 billion is the floor, not the ceiling. There is one large independent identity graph left in the market and a handful of smaller ones. WPP, Omnicom-IPG, and Dentsu are now staring at a deal that puts a hard number on a capability they have spent the past two years describing as adjacent and reframes the holding-company thesis. We would expect a bid for at least one of the remaining graphs inside 18 months. If nobody else moves, that is its own signal about how the rest of the industry sees the agentic thesis.

The gap the market just priced

Twelve months ago, the data middleware between brands and AI agents was a curiosity. There were vendors, there were standards conversations, there was no capitalized market. Today there is a $2.2 billion deal in flight with agentic commerce written into the press release, and a holding company betting its CoreAI thesis on owning the layer.

That changes the math on adjacent assets. If LiveRamp is worth $2.2 billion to Publicis as an agent input, what is Publicis itself worth as an agent platform, with CoreAI plus LiveRamp plus the brand roster wired together? The same question applies, with smaller numbers, to every retail media network, every customer data platform, and every clean room sitting between brands and the surfaces their customers actually use.

This is the same gap we flagged in the ad-free AI bet. If the agent surface stays ad-free, the audience graph still has to live somewhere. Publicis just made the case that "somewhere" is the holding company, not the model and not the network.

Sources

If the model is half the product, and the rail is half the product, who owns the half that connects them?

Charlie Major is a Product Development Manager at Mastercard. The views and opinions expressed in Major Matters are his own and do not represent those of Mastercard.