The Middle Cannot Hold

The classic lumascape of the programmatic supply chain—advertiser, agency, DSP, exchange, SSP, publisher—has always been a polite fiction. In early 2026, it's completely unraveling. Buy-side platforms are plugging directly into publishers. Sell-side platforms are courting buyers directly. Agentic AI is actively gutting the manual trading steps that kept these middlemen employed for the last decade.

Hovering above all of this, vacuuming up the vast majority of digital ad dollars, the walled gardens are largely ignoring the drama. They never needed this convoluted plumbing to begin with. The independent ad tech ecosystem is fracturing under the pressure. The middle layers aren't just facing margin compression; their fundamental purpose is under fire. If the open internet cannot align on how to route, measure, and automate this new reality, it risks losing the commercial gravity required to survive.

Market Context · 2026

• ~$1.19T — Global ad spend 2025 (WARC)
• 58% — Meta, Alphabet & Amazon share of non-China spend
• ~$650B — Big Four AI capex commitments 2026

Sources: WARC Global Ad Spend Outlook; company filings; Magna Global. Open ecosystem estimated at ~42% of ex-China global spend and declining.

SPO enters the agentic era

Every major demand-side platform has a clear financial imperative: reduce the SSP margin. For years, they wielded Supply Path Optimization (SPO) via curated allowlists. Today, SPO is increasingly handled by AI agents that allocate spend autonomously, slashing any hop in the chain that doesn't definitively prove its value.

The Trade Desk (TTD) forced the issue in late 2025 when it reclassified SSPs as "resellers" within its Kokai platform. Under programmatic bidding logic, this wasn't just a semantic tweak; it was an algorithmic penalty against sell-side auction priority. TTD was sending a blaring signal to publishers: Our direct-to-publisher OpenPath is the favoured pipe. The financial hit was tangible. PubMatic, for example, saw its full-year 2025 revenue decline by ~3%, repeatedly attributing the headwinds in earnings calls to "a legacy DSP" altering its inventory valuation model. The message was clear: if DSPs can go direct, SSPs are redundant.

The Core Tension: Buy-Side vs. Sell-Side Strategic Moves

Amazon is squeezing the middle from a different angle. Amazon Publisher Services (APS) has aggressively expanded its direct publisher footprint, especially in connected TV (CTV). This setup feeds Amazon DSP buyers exclusive, first-party retail data. When an Amazon DSP bid carries proprietary purchase intent that an open-web SSP physically cannot access, the SSP isn't just an added cost—it's an inferior pipe.

The sell-side pivot: Upstream execution over plumbing

Cornered, the sell side is attempting a massive pivot. The old SSP playbook (take a bid request, run an auction, pass the impression) is completely commoditised. Now, companies like Index Exchange, Equativ, Magnite, and PubMatic are sprinting to rebrand themselves around what Index CEO Andrew Casale has termed "sell-side decisioning." They have to prove they add enough analytical and agentic value at the impression level to justify their existence.

The solution taking hold in early 2026 is "upstream execution"; moving AI decisioning to the exchange level before the bidstream even reaches a DSP. By baking AI models directly into their infrastructure, SSPs are positioning themselves as intelligent, proactive filters.

Consider the flurry of recent sell-side product launches:

• Equativ’s Maestro: In late 2025, Equativ rolled out an "Agentic AI Programmatic Suite" embedded directly into its platform, featuring an AI Media Planning Agent that interprets natural language agency briefs and automatically generates Private Marketplaces (PMPs) without human traders.
• Index Exchange & Chalice AI: Index Exchange is aggressively pushing "sell-side decisioning." Through a major partnership with Dentsu and Chalice AI, Index allows custom AI bidding agents to run directly on top of the exchange, evaluating rich publisher signals upstream to eliminate the network latency of external API calls.
• Magnite’s ClearLine & Demand Manager: Magnite acquired streamr.ai to inject generative AI directly into its workflow, while simultaneously deploying machine learning within Demand Manager to continuously run A/B tests and autonomously optimize Prebid wrapper configurations for publishers.
• Scope3's Brand Agents: Operating across SSPs like Index Exchange and Equativ, Scope3 deployed LLM-powered "Brand Agents" that evaluate every impression in real-time for safety, suitability, and carbon emissions before a DSP even makes a decision.
• PubMatic’s AgenticOS: Acting as an orchestration layer, an AI agent fed a natural language brief (via Claude or GPT-4) can plan and execute campaigns autonomously on PubMatic's rails.

"Moving AI intelligence upstream makes technical sense. But selling 'the SSP as an operating system' is a massive uphill battle in a market trained to see them as interchangeable, expensive pipes."
— The dilemma facing independent sell-side platforms in 2026

Commercially, this defense strategy is logical. If an SSP can generate revenue through pre-bid intelligence, direct AI workflows, and proprietary curation, it creates a necessary buffer against buy-side margin compression.

AI agents show no loyalty

Agentic AI doesn't resolve the buyer/seller standoff; it throws fuel on it. Human media buyers make SPO decisions sporadically, possibly hindered by incomplete data or legacy platform loyalties. AI agents optimise continuously, at microsecond resolution, with brutal cost transparency. The fee compression that manual SPO ground out over the last five years is about to be replicated in months.

Early pilots from late 2025 are posting highly marketable metrics. Independent agency Butler/Till ran a fully autonomous campaign for Geloso Beverage Group (Clubtails) using Claude to brief PubMatic’s agents. Beyond the claimed 87% drop in setup time, the kicker was the cost efficiency: the automated path reduced buy-side supply chain costs by 5.5x while hitting a 98% video completion rate. If those numbers hold at scale, no human trader can justify their own workflow.

But the sword cuts both ways. The same ruthless AI logic that kills low-value hops will inherently route money toward SSPs that genuinely boost yield, provide better audience signals, or secure premium supply for less cash. Since TTD has explicitly avoided building yield optimisation tools for publishers, independent SSPs still have a technical foothold—provided their intelligence layer actually works.

The Walled Garden Reality

A market splitting into walled gardens and everything else

This entire supply-chain conflict is ultimately a fight over a shrinking pie. The open programmatic ecosystem isn't losing ad spend to rogue DSPs or SSPs; it’s losing them to closed platforms that operate without a traditional supply chain.

Meta, Alphabet, and Amazon control roughly 58% of non-China global digital ad spend in 2026, per WARC. That dominance is accelerating. All three deploy agentic AI within closed-loop systems, vastly outperforming open-web alternatives while marking their own homework.

Estimated global digital ad revenue share (ex-China) · 2026:

• ~26% — Alphabet (Google/YT)
• ~22% — Meta (Facebook/Insta)
• ~10% — Amazon Ads
• ~42% — Open ecosystem (all independent publishers, ad tech vendors, CTV, etc.)

The remaining 42% is fought over by every independent publisher, CTV app, news outlet, retail media network, DSP, SSP, and ad tech vendor combined. It's a hyper-competitive space battling for a fraction of what Google rakes in by itself.

The driver of this concentration is closed-loop signal quality. Meta’s Advantage+ campaigns drive better returns because Meta perfectly tracks post-ad conversions on its own properties. Google’s Performance Max allocates spend in a black box that defies third-party auditing. Amazon wields proprietary purchase intent data. As these platforms deploy AI agents, the loops tighten. An AI agent inside Meta is highly effective precisely because it never has to leave.

The uncomfortable reality is that the independent ad tech arms race is happening on a battleground the walled gardens have already won. Arguing over whether The Trade Desk, Index Exchange, or PubMatic controls the pipe is highly consequential for those companies, but at a macro level, it resembles a fight over the governance of a declining asset.

The Widening Gulf

The ad market has permanently bifurcated. On one side: walled gardens with closed data loops, deterministic attribution, and massive scale. On the other: an open ecosystem surviving on reach, contextual relevance, and fragmented inventory.

The open web's advantages are real but fragile. Premium CTV remains fiercely independent, with players like Paramount, Warner Bros. Discovery, and major FAST channels operating outside the gardens. Premium journalism and off-Amazon retail media networks (Walmart, Kroger, Target) also provide unique, brand-safe inventory.

Yet, these holdouts are vulnerable to the exact AI dynamics the SSPs are touting. If a brand’s AI agent optimises purely for real-time ROAS, it will consistently shift budget to the walled gardens because their closed-loop attribution is mathematically superior. The gravity well is inescapable: money follows measured performance, and the platforms own the measurement.

Scenarios for the Open Ecosystem

Most Likely Trajectory: A tiered open ecosystem with AI-defined routing
Premium CTV, quality journalism, and retail media outside Amazon retain meaningful open-ecosystem share. Agentic buyers route programmatic spend through a consolidated set of SSPs with genuine AI infrastructure (PubMatic, Magnite, Index Exchange, Equativ). Mid-tier and long-tail open-web inventory commoditizes further. The open ecosystem's share stabilises at ~35–38% of ex-China spend by 2028 — smaller, but concentrated in higher-value environments.

Adverse Scenario: Walled garden gravity reaches escape velocity
Agentic AI optimization systematically redirects budgets to closed-loop environments where measurable performance consistently exceeds open-web equivalents. Independent SSPs fail to prove incremental ROI versus DSP-direct or publisher-direct models. The open ecosystem's share falls below 30% of ex-China spend, investment in open-web infrastructure collapses, and the market consolidates into three to four vertically integrated walled gardens plus a rump open-web layer with inadequate measurement and declining brand appeal.

Establishing a baseline for survival

The open ecosystem doesn't need to defeat the walled gardens; it just needs to remain viable enough to attract sustainable investment. Achieving that requires clearing a few massive hurdles.

First, the attribution gap must close. Clean rooms and identity graphs are band-aids. The discrepancy between tracking an open-web impression and an Instagram conversion is still too wide. Brands need standard ways to prove that open-web inventory actually drives business outcomes.

Second, the industry needs a truce on protocol standardisation. The back half of 2025 birthed two competing frameworks. The Ad Context Protocol (AdCP) backed by Scope3, PubMatic, and others—acts as an open communication layer allowing buyer and seller LLMs to negotiate intent flexibly. Conversely, the IAB Tech Lab’s Agentic Advertising Management Protocols (AAMP) aims to containerise AI within existing programmatic infrastructure to prevent auction latency and "AI hallucinations." Until the independent ecosystem universally adopts a single standard, cross-platform AI buying will require endless bespoke integrations. A fragmented standard is a tax on the open web—one the walled gardens don't pay.

Finally, SSPs must prove that their "agentic" layers actually drive better outcomes than a direct DSP integration. Early pilots posting 5.5x reductions in supply chain costs are phenomenal case studies, but the market needs hard proof that routing through an AI-enhanced SSP consistently yields higher publisher CPMs and lower buyer CPAs across thousands of simultaneous executions.

"The fight between DSPs and SSPs is essentially a turf war over a shrinking pie. Unless independent ad tech fixes its measurement gaps and agrees on an AI standard, Big Tech will keep eating the market."

Agencies try to keep control

Agencies occupy a precarious position in this transition. The obvious read is that agentic AI massacres execution roles. If an SSP tool like Equativ's Maestro or PubMatic's AgenticOS can cut campaign setup by 80% and resolve pacing issues autonomously, thousands of trading desk jobs become obsolete. This breaks traditional agency billing models.

However, holding companies aren't fighting the automation; they're trying to orchestrate it. Dentsu’s partnership with Index Exchange and Chalice AI is a clear example of agencies building proprietary, transparent AI bidding agents tailored for their clients. Similarly, WPP Media was a marquee launch partner for PubMatic’s AgenticOS to enhance its WPP Open platform. This isn’t about playing nice with vendors, it’s a bid to white-label the technology. The pitch evolves from "we execute campaigns for you" to "we own the master AI that routes your media."

Independent agencies are far more exposed. Agile independents are leaning into AI for raw efficiency, but survival means shifting billable hours away from manual trafficking and toward prompt engineering, guardrail design, and creative strategy. Agencies clinging to execution margins will simply be priced out.

Conclusion: Time is running out

The DSP vs. SSP war will keep generating fun headlines—new direct connections, containerised exchange-level AI, protocol disputes, and earning-call snipes about "resellers." These tactical skirmishes matter deeply to the vendors involved.

But the existential threat is the walled gardens' compound advantage in signal quality and AI optimisation. Every year the open web fails to close the attribution gap, its share of ad spend shrinks. Advertisers aren't leaving the open web because it lacks good content; they're leaving because it lacks flawless, closed-loop measurement.

The companies tackling this (e.g. Magnite, Index Exchange, PubMatic, Equativ) are solving real problems. But their window is closing. AI agents are deploying at scale today. Once these algorithmic buyers establish their routing preferences, changing their behavior will be nearly impossible.

The open programmatic middle can survive, but only if it delivers interoperable agentic infrastructure and undeniable measurement. The next eighteen months will determine if the current wave of innovation is enough to stop the bleeding, or just a sophisticated delay of the inevitable.

March 2026 · All market share figures are estimates based on WARC, Magna Global, and company filings.