WPP’s $400m Google Bet: Are Agencies Building Brains—or Renting Them from the Ad Seller?
On October 14, WPP inked a five‑year, $400 million commitment with Google to embed Gemini, Veo and the rest of the Alphabet toy chest into its operating system, WPP Open. It’s the first big swing under new CEO Cindy Rose and it formalises what the industry has been inching toward for years: holding companies striking enterprise‑scale technology compacts that promise speed, savings and “AI everywhere” in return for very large cheques and very public endorsements (WPP press release) (FT) (MediaPost).
The headline - $400 million - is eye‑watering. But the underlying question is bigger: does this kind of deal build durable, agency‑owned capability, or does it lock the industry’s largest “independent” buyer more tightly to one of the biggest sellers of media on earth? Before we get to the risks, it’s worth unpacking how these deals really work, why they get signed, and what—if anything—they’ve delivered when you peer beyond the press release.
How these deals happen (and what they actually include)
The modern holding‑company technology deal is part cloud enterprise agreement, part co‑marketing pact, part talent pipeline and part product roadmap outsourcing. The mechanics are surprisingly similar from one group to the next.
There’s usually a multi‑year spending commitment with a preferred cloud + AI stack. WPP’s is five years with a $400m spend envelope to “advance cloud and AI,” with named access to Veo and Imagen for video and images, deep integration of Gemini across WPP Open, and a promise of early access to new models and research tie‑ins (WPP, 14 Oct 2025). This builds on a 2024 collaboration announced at Google Cloud Next to wire Gemini 1.5 Pro into WPP Open’s creative and performance tooling (WPP, Apr 2024).
There’s always a deliverables list that sounds like a product roadmap. In WPP’s case: bespoke audience models in Open Intelligence on Google Cloud; privacy‑preserving data collaboration (InfoSum “Bunkers” via Marketplace) baked into WPP Open; early access to Veo/Imagen for production; and a training program to mint 1,000 “creative technologists” by 2030 (WPP, Oct 2025). WPP also points to a £300m annual AI budget this year, up from £250m in 2024, with 69,000 monthly internal users on WPP Open—about 85% of client‑facing staff (WPP Interim Results 2025) (Annual Report 2024 “Our Strategy”).
And there are similar plays across the Big Six. Omnicom made Microsoft Azure/OpenAI the engine for Omni Assist and embedded Microsoft Advertising supply directly into its workflow (Omnicom, Jun 2023) (Annalect, CES 2024). Dentsu expanded a Vertex AI partnership to wire Merkle’s GenAI products into Google Cloud (MediaPost) (Dentsu). IPG Mediabrands signed a three‑year Prime Video ads pact with Amazon that granted first looks at formats as Prime Video switched on ads in 2024 (Amazon Ads) (Marketing Dive). Havas announced a €400m, four‑year AI and data program across the group under its Converged.AI strategy (Havas, 2025) (CSA / Havas Media Network).
The “why” is part economics, part signaling. Committing spend can secure better cloud rates and priority access; it can also launder a capability gap into a “transformation narrative” that calms investors and helps win procurement bake‑offs. At Cannes, Vegas and London, “AI‑enabled at scale” plays well in rooms where clients are cutting headcount and want answers faster, not incremental workshops. And for the platforms, commandeering agency operating systems is a strategic wedge into the $1 trillion global ad economy dominated by Google, Meta and Amazon (MarTech) (MAGNA summary).
Do they actually work? The Marcel test
Publicis gives us the cleanest long‑run experiment. In 2018 it unveiled Marcel—the in‑house “platform” to connect and direct a 100,000‑person workforce—built with Microsoft and Azure AI (Microsoft, 2018) (Press kit). Marcel has delivered very visible stunts (100,000 personalised AI thank‑you videos from Arthur Sadoun) and, per Publicis, real internal benefits, with executives crediting it for saving over 2,000 jobs in the pandemic by reallocating talent at speed (LBBOnline) (Campaign Brief).
But the thing that actually changed Publicis’ competitive position is less Marcel’s UI and more the balance sheet moves: spending €12bn over a decade on data and technology, with Epsilon ($4.4bn) and Sapient ($3.7bn) as crown jewels (Reuters, Oct 2025) (BusinessWire 2024/2025 results packs) (Publicis Sapient PR). The result: multiple years of outperformance and raised guidance, with leadership explicitly tying momentum to first‑party data and AI leverage across Epsilon and Sapient—not Marcel alone.
That’s the sober reading for WPP’s Google deal. The OS matters. The integrations matter. But, historically, the durable advantage has come from owning scarce assets (identity, data rights, transaction adjacency, software IP) that compound over time rather than the big cloud deals. Public cloud access and headline model integrations are table stakes. The moat is what you control and own. And I’m not convinced this WPP deal means ownership.
The cultural risk: renting transformation
There is a deeply uncomfortable cultural truth here: when you outsource the scaffolding of your operating model to a supplier, you rarely train the muscles to build your own. Holding companies have spent a decade saying they’re software businesses; then they hire external consultants to wire theirs up temporarily. The immediate effect is seductive - speed, case studies, a shiny demo at Next or Build. The longer‑term effect is capability atrophy.
If you want proof points for the opposite path, look at acquisition compounding. IPG bought Acxiom’s marketing unit in 2018 for $2.3bn and, years later, Acxiom is the technology spine for identity and clean‑room workflows across the group . Publicis bought Epsilon and Sapient and can now credibly claim that 70%+ of its operations are AI‑powered on infrastructure it controls and monetises, with data products that drive client stickiness and margin (Reuters) (Publicis UK).
WPP, to its credit, hasn’t been passive. It has partnered with NVIDIA to rewire creative production via Omniverse and invested in Stability AI, signalling a desire to own parts of the content engine rather than just rent APIs (NVIDIA) (WPP x Stability AI) (FT Stability AI). The question is whether a Google‑anchored deal accelerates or crowds out the internal build/buy cadence that creates agency‑owned IP. If it’s the latter, shareholders are financing Google’s P&L while WPP’s core technology leverage remains thin.
Broadly, I think the future here needs to be buying technology companies and capability to build product-led orgs within holding companies. To me it’s imperative that the holding companies, to become software companies, need better product organisations to function. And they need products that their end users - clients - interact with willingly and frequently.
The model risk: you’re embedding a black box you don’t control
The second risk is structural. When you wrap your operating system around someone else’s LLM, you are embedding a moving target into your decision factory. Models update constantly. Guardrails tighten. Context windows expand. Reasoning modes change names. Training data shifts. You will wake up one morning and the tool you use for performance prediction or creative scoring will behave differently because a vendor rolled a new update overnight.
We’ve already seen how quickly things can swing. Google paused Gemini’s people image generation after “unacceptable” outputs, then pushed fixes; subsequent releases have aimed at addressing regressions in reasoning and output stability (Ars Technica). This is normal in frontier software; it’s untenable if your governance depends on managing billions in capital decision making (which holding companies absolutely do!). Enterprises call it model drift; regulators will soon call it your problem (IBM on drift) (Microsoft on drift).
The solution isn’t to avoid models. It’s to put them behind a measured framework: version‑pinned inference for production decisions, and likely some degree of orchestration or governance control. Open source models have a really good framework here as well. But the reality is: if you don’t know how the model is changing underneath you, you may introduce huge risk you cannot see.
The independence risk: agencies buying intelligence from the sellers of media
Here’s the conflict no one wants to say out loud: the world’s largest ad sellers are also the world’s most aggressive AI vendors. Google and Meta dominate global digital ad revenue; Amazon is the fastest riser; Microsoft is back above $20bn in annual ad revenue across Bing, LinkedIn and gaming (MarTech) (BestMediaInfo/MAGNA) (PPC Land on Microsoft). These same companies are the providers of the “AI brains” agencies are now wiring into planning, creative and buying.
It’s as if a ratings agency licensed its credit models from the investment banks it’s meant to be policing. WPP’s own announcement says the Google partnership includes preferred access to models “integrated directly into WPP Open” and explicitly references activation across Google ad platforms (WPP, Oct 2025). If the brain that guides budget and creative is trained, tuned and shipped by the seller of the inventory, the onus is on the agency to prove independence- not just assert it - and to play an active role managing that conflict responsibly. Regulators are already circling Google’s market power in the UK and EU; independence isn’t just a virtue signal, it’s a hedge against future remedies that could land on your stack (AP—UK CMA “strategic market status”) (EU adtech conflicts debate).
There’s a practical angle too. Platform automation is swallowing chunks of the planning and buying craft. Agencies report mounting pressure from Google reps to push Performance Max and AI‑driven tools, with fewer levers left for human optimisation (Gradient Group). Meta is marching in the same direction. The more your OS depends on their intelligence, the less room you have to differentiate when the defaults are already set. It’s incredibly risky for advertisers to have a legion of agencies building their intelligence stack completely dependent on the world’s largest sellers of advertising inventory.
Why holding companies all sign these deals anyway
Because clients want speed and certainty—and boards want a transformation story that shows up in a quarter, not a decade. A Google, Microsoft or Amazon logo buys permission. A joint press release buys time. AAnd sometimes the benefits are tangible: there are real production and time‑to‑market gains from Veo‑grade video models, Omniverse‑based pipelines, and centralised data collaboration when it’s done well (WPP–NVIDIA) (WPP–Google 2024/2025).
But it’s also true that what looks like strategy can sometimes be procurement theatre. The discipline is not signing the deal; it’s constraining the blast radius.
What this means for clients—and why they should care
Clients hear “AI everywhere” and mostly want faster answers and less internal bureaucracy. They should ask different questions. Who holds the model governance? Are the predictions reproducible and controlled? How many models are in the loop and where do they play? What happens when Google ships a breaking change the week before Black Friday? Does the agency’s OS nudge spend in ways that systematically preference platform inventory and how do they manage that risk? If the answer is “trust us,” don’t.
There’s another wrinkle: the LLM vendors are not neutral utilities. They are, directly or indirectly, ad businesses. Google and Meta already are. Microsoft is. OpenAI isn’t—yet—but retail partnerships and shopping features show where the gravity points, regardless of today’s statements about “no ads” in product (Reuters—OpenAI shopping features, no ads). If the intelligence layer becomes an ad network in its own right, agencies hard‑wired to that layer will find independence an expensive retrofit.
The verdict
Cindy Rose’s $400m Google bet is bold, fast and—on the surface—rational. WPP was always going to go big on someone. Better to nail a five‑year roadmap with early access than drift platform‑agnostic while rivals sprint. If the deal helps WPP turn weeks into days for asset production and gives Open Intelligence real teeth, that’s tangible value for clients—and a story the street will reward (FT).
But the market has seen this movie. Technology partnerships are accelerants, not strategy. The lesson from Publicis isn’t that Marcel transformed a culture; it’s that owning Epsilon and Sapient transformed cash flows. The lesson from the platforms isn’t that AI is magical; it’s that the sellers of inventory intend to be the sellers of intelligence too. If agencies want to be taken seriously as independent stewards of client spend, they need an OS that stands on its own feet independent of that world of conflict.
So celebrate the speed. Bank the training. Use the Veo demos to win pitches. Then take the cash you saved on compute and buy the assets that make you harder to copy and keeps intelligence independent. Otherwise you’re just renting the future from the very companies you’re supposed to negotiate with on behalf of your clients.