The Great AI Vendor Squeeze - Where AI Actually Lands Inside Agencies

A year ago, the “AI solution stack” framing was comforting in the way org charts are comforting: layered, rational, tidy enough to argue over in a steering committee. You could point to “creative generation” over here, “workflow automation” over there, “measurement” somewhere downstream, and pretend the whole system behaved like enterprise software usually does.

Then 2025 happened. Not in the cinematic sense — no singular “AI moment” where everything flipped overnight. Instead, it happened the way large systems always change: quietly, then suddenly, and finally all at once when procurement asks why you’re paying seventeen vendors to do what one suite now claims it can do.

The most important change is this: what used to look like a stack now behaves like an operating system plus an ecosystem of apps, connectors, and specialized “skills.” The holding companies are racing to build that OS layer inside their own walls — and the mega-vendors are racing to make sure you never need to leave theirs.

That collision is where the future of “small, single-problem AI solutions” gets decided.

The platform era is being forced by physics, not hype

If you want the non-sexy reason the stack is converging, it’s not “innovation.” It’s volume.

Adobe has been blunt about the pressure: it describes the content supply chain as “a complex web” and notes that marketers expect content demand to quintuple between 2024 and 2026.  That kind of scale doesn’t just punish manual production — it punishes fragmented tooling, broken handoffs, and “creative ops by spreadsheet.”

When content velocity becomes a business model requirement, the center of gravity shifts from “best creative tool” to “end-to-end throughput.” That’s why suites are winning mindshare: they sell coordination, not just capability.

Adobe’s GenStudio positioning is a perfect artifact of this shift: it isn’t pitching one product; it’s pitching the ability to plan, create, manage, activate, and measure — and to unify workflows across Creative Cloud and Experience Cloud. 

And once a vendor can credibly claim “end-to-end,” every point solution immediately gets re-labeled as one of three things: a feature, a plugin, or a procurement problem.

Holding companies are building internal “AI OS” layers on top of vendors

If vendors are building suites, the holding companies are building platform organizations — not only because it’s strategically attractive, but because it’s a survival requirement when every client asks, “So where is AI in your delivery model, exactly?”

Publicis formalized this direction with CoreAI, positioning it as a unifying layer connecting enterprise knowledge, proprietary data, and applications across the group.  In a 2025 Cannes interview, Arthur Sadoun tied the strategy to years of investment, saying Publicis has been “investing over $10bn in data, AI, and technology,” highlighting Sapient and Epsilon as key foundations. 

WPP has been telling a similar story with WPP Open. In a 2025 interview, Mark Read described broad internal adoption: “50,000 people every month” using WPP Open for everything from insights to audience testing to producing work and planning media.  Reuters also framed WPP Open as part of WPP’s effort to consolidate brands and integrate services around an AI-enabled platform. 

Omnicom’s posture, especially post-IPG combination messaging, leans into data + platform scale. In a 2025 interview, John Wren described a “platforms group powered by generative AI” and argued that it would become hard to match without being “one of the big six tech companies.”  That’s not just bravado — it’s a signal: platform gravity now competes with agency craft as the differentiator.

This is the new reality: the “stack” decision is no longer just which vendors you buy. It’s which platform layer you trust to arbitrate identity, permissions, workflow, and quality across everything else.

The stack view still works — but only if you invert it

Here’s the twist: the stack concept is still useful, but it needs to be flipped from “tools” to “control points.” Instead of asking, “Which tools cover which functions?” the sharper question is, “Where does control live?”

Control in 2025 mostly lives in four places.

First, it lives in identity and data foundation — because personalization, measurement, and safety are downstream of knowing who is who, what they’re allowed to see, and what you’re allowed to do with their data.

Second, it lives in workflow orchestration — the unglamorous layer that decides whether your brilliant model output becomes a shipped asset, a compliance incident, or an abandoned experiment.

Third, it lives in asset governance — rights, provenance, brand rules, reuse, and the audit trail that suddenly matters when clients ask uncomfortable questions.

Fourth, it lives in distribution and performance feedback — because once systems can generate at scale, the bottleneck becomes not “creation” but “what is working, for whom, where, and why.”

Suites are racing to own all four. Holding-company platforms are racing to intermediate all four. The winners are whoever can turn those control points into a repeatable operating model — not a demo.

What Adobe and ElevenLabs are really signaling with “expansion.”

Adobe’s GenStudio story is explicitly about unifying the content supply chain and embedding generative capabilities into workflows, including activation partnerships with Microsoft, Google, and LinkedIn.  It’s also about scaling production via APIs — Firefly Services is positioned as a set of generative and creative APIs for workflow automation. 

This matters to holding companies because APIs are the new procurement hack: you can “standardize” on a vendor without forcing every team into the same UI, and you can embed capabilities into your own platform layer while keeping governance centralized.

ElevenLabs is a different flavor of expansion: it’s the “single capability” vendor moving upstream and downstream. It’s no longer just text-to-speech; it’s positioning itself around podcast production and even “GenFM,” where content becomes generated conversations from documents.  And in 2025, it pushed into music generation with commercial-use positioning in the market conversation. 

Put bluntly: vendors are learning that “one feature” is fragile. A platform story is defensible. That defensiveness is precisely what compresses the space for early-stage point solutions — unless they evolve the way the market now rewards.

So… do point solutions still have a chance?

Yes — but not by behaving like it’s still 2023. A point solution wins in the platform era when it fits at least one of these realities.

It becomes a specialized capability that the platforms don’t want to build because it’s too niche, too messy, or too liability-heavy. In practice, this often means compliance automation, rights management, regulated-industry workflows, or hard-to-maintain edge cases inside production.

It becomes a performance unlock that is measurable enough to survive the “suite rationalization” wave. If your value can’t be expressed as time saved, errors reduced, rework prevented, or throughput increased — in numbers a CFO won’t laugh at — you are a feature request, not a vendor.

It becomes a governance unlock. Platforms are powerful, but they also create a single point of failure. The more generative output scales, the more executives care about auditability, permissions, model choices, and policy enforcement. If you solve a governance pain that suits hand-waving, you can earn a seat even in a consolidated stack.

It becomes a connector — not in the basic “integration” sense, but in the “makes the operating model actually work across fragmented realities” sense. In 2025, many organizations don’t have a tooling problem; they have a system-coordination problem. The point solution that insists on being “the destination” struggles. The point solution that becomes a reliable component thrives.

Are WPP, Publicis, Dentsu, Omnicom “one or two vendor” shops now?

They want to be. They can’t be. And that’s the opportunity. They want fewer vendors because vendor sprawl is operational drag: fragmented permissions, inconsistent data handling, overlapping costs, and endless enablement. But they can’t be pure “one or two vendor” ecosystems for three reasons.

The first is client heterogeneity. A holding company platform has to serve wildly different client stacks, regulatory contexts, and maturity levels, sometimes inside the same quarter.

The second is competitive differentiation. If every holding company standardizes on the same two mega-vendors, differentiation collapses into pricing and talent branding — which is exactly what everyone claims they’re escaping.

The third is risk. Putting all value creation inside one vendor suite is strategically dangerous. If the vendor changes terms, shifts the roadmap, breaks compatibility, or becomes the headline for the wrong reasons, you inherit the blast radius.

That is why most giants aim for a “core + ecosystem” model: a few anchor platforms for identity/workflow/asset management, plus a controlled set of specialized capabilities.

Dentsu’s integration of Adobe GenStudio with its Merkury data/identity platform is a good example of this “core + ecosystem” posture: it’s explicitly a combined offering rather than a single-vendor surrender. 

Are they still interested in stealth-stage AI?

Yes — but the definition of “interesting” has changed. In the earlier stack era, a stealth tool could be interesting because it was clever. In the platform era, it becomes interesting because it is adoptable inside a governed system.

The large holding companies are not short on demos. They are short on deployable capabilities that survive real operations: permissions, procurement, security review, model risk review, integration into existing workflows, training, and support. Stealth-stage companies that understand this are rare — which is exactly why they remain attractive.

This is also why many holding-company leaders keep emphasizing adoption and upskilling as a core constraint.  The question isn’t “Can the model do it?” It’s “Can the organization do it repeatedly without breaking itself?” Stealth is still desirable when it produces a capability that can be operationalized as part of the platform layer — especially when it closes a gap that a suite won’t prioritize.

The most under-discussed shift: agencies are becoming system integrators again

There’s an irony in all of this. For years, agencies tried to escape the perception of being “implementers.” They wanted to be strategic partners, creative leaders, growth advisors. Fair. But the AI platform era is dragging agencies back into a familiar role: integration and orchestration — not of media plans, but of systems. The biggest holding companies are effectively building internal “enterprise software companies” that happen to sell marketing outcomes.

Publicis literally framed itself as becoming an “Intelligent System company” with CoreAI sitting centrally.  WPP’s messaging around WPP Open is similar: an internal platform used at scale across the organization. 

This redefines what “best-in-class” means. It’s less about the best single model and more about the best operationalization of many models across many workflows under governance. Which also means: the point solutions that win are the ones that respect the integration reality instead of pretending it doesn’t exist.

A practical way to think about your stack in late 2025

If you run operations, production, or creative enablement inside a large group, you can treat the stack question as three overlapping negotiations. You’re negotiating with your holding-company platform team about what gets standardized, what gets embedded, and what gets left to local choice. You’re negotiating with mega-vendors about how much of your workflow they will own, and how portable your assets, metadata, and learning will remain if you ever need to move.

And you’re negotiating with smaller vendors about whether they are building a product or building a component. The ones building components will talk about APIs, governance, deployment models, audit trails, and how they fit inside your platform. The ones building products will talk about features. Features are easy. Fit is rare.

The contrarian conclusion: consolidation makes niches more valuable, not less

Most people look at suite expansion and conclude, “This kills startups.” Often, it does. But inside large organizations, consolidation creates a predictable side effect: it forces standardization at the center, and then pushes complexity to the edges. The edge is where real work happens — local regulations, client-specific processes, legacy systems, unique brand constraints, unusual formats, weird supply chains, and the occasional executive request that begins with, “This should be simple…”

That edge complexity is where specialized solutions still win — if they’re built to be consumed by platforms, not to compete with them head-on. So, the “solution stack” didn’t vanish. It matured into something more political: a battle over who owns the operating system, who governs the workflows, and who gets to define what “quality” means at scale.

If you’re a holding company, the stack is now a strategy. If you’re a mega-vendor, the stack is now a moat. If you’re a startup, the stack is now a choice: become a component, or become a cautionary tale.