PubMatic yesterday launched Decision Fabric, a real-time intelligence layer built inside its auction infrastructure that allows DSPs, curators, and algorithm companies to deploy their proprietary AI decisioning models natively within the supply-side platform - not connected from the outside, but operating inside the auction on the full signal at bid speed. The announcement, published June 1, 2026, by Punit Shah, Director of Product Marketing at PubMatic, addresses a structural problem the company argues has quietly cost the industry billions.

The premise is straightforward, even if the technical execution is not. Programmatic advertising was designed with decisioning on the buy side. The DSP holds the logic; the supply path delivers inventory. The flaw in that arrangement, according to PubMatic, is that the richest signals in any programmatic transaction live on the supply side - full auction context, publisher data, content signals, real-time inventory dynamics, and access to inventory before traffic shaping filters what buyers can see. By the time a bid request reaches a DSP, those signals are already compressed, filtered, and delayed.

The signal gap that drove the architecture

Audience qualification at the DSP layer happens against an incomplete picture. Buyers see only a subset of the available opportunity set, and the decisions they make are reduced to binary in-or-out calls that cannot capture the full complexity of what makes a given impression valuable. Wasted spend, missed moments, and performance that has fallen short of its theoretical ceiling are the consequences.

As AI raises the bar for what decisioning models can do, the gap between the data those models need and the data they actually receive gets harder to justify. According to PubMatic, more sophisticated models running on compressed, stale, buy-side-only data do not close that gap. They make it more expensive. The vision Decision Fabric is built around is one where buy-side intelligence and sell-side signals converge at the exact moment of the transaction - qualification happening in real time, on the richest data available, at the point where inventory meets demand.

That premise is not unique to PubMatic. The broader industry has been building toward it for some time. Zillow tested containerized RTB with Chalice and Index Exchange as early as August 2025, embedding buy-side intelligence within supply-side infrastructure to access site-level data that standard bid requests do not carry. IAB Tech Lab released the Agentic RTB Framework version 1.0 for public comment in November 2025, standardizing how containerized agents participate in real-time bidding infrastructure. What PubMatic is doing with Decision Fabric is building its own proprietary implementation of this architectural shift, at exchange scale, on infrastructure it owns outright.

Train, Deploy, Activate: the three-step process

Decision Fabric follows a three-stage process. Each stage addresses a distinct part of the problem.

Train. Partners begin with data. According to PubMatic, the company delivers a privacy-safe sample of exchange-wide bidstream data, drawn from winning impressions across the US market to a partner-specified location. The dataset includes more than 60 OpenRTB fields: content signals, device type, geo, audience identifiers, and contextual attributes across web, mobile app, and CTV inventory. All personal data is hashed before delivery. No raw identifiers leave PubMatic's environment.

The value of this training data lies not just in its volume but in its composition. Partners training models on PubMatic's bidstream are working with a materially larger and more representative pool of inventory and users than they could access elsewhere - specifically because it precedes the traffic shaping that filters out large portions of auction traffic downstream. Supply-side bidstream data, according to PubMatic, captures signals that are compressed or absent by the time a bid request reaches a DSP: full content context, publisher signals, auction dynamics.

Deploy. Partners package their trained model as a Docker container image, using the industry-standard containerization format built on Open Container Initiative specifications. PubMatic hosts that container within its controlled infrastructure. The partner retains full control of what is inside - their decisioning logic, their proprietary IP, their model architecture. Current scope is one container per partner, US ad requests only. Traffic filters give partners control over what their model sees: platform selection across web, mobile web, mobile app, and CTV; publisher allow lists; and per-publisher sampling rates.

Activate. At each qualifying auction, PubMatic routes the bid request through the partner's container. The container receives the request, applies the partner's decisioning logic, and returns a list of segment IDs - the audiences that qualify for that impression, in that moment, against live signals. Those segment IDs flow back into the auction and qualify the impression for the appropriate curated deals. According to PubMatic, total response time is under ten milliseconds.

Because decisioning happens before traffic shaping, partner models see a materially larger pool of users and inventory than any DSP ever processes. PubMatic characterises this as a structural advantage - one built into where Decision Fabric sits in the supply path, not a marginal performance gain achieved through optimisation.

Why the infrastructure itself matters

PubMatic's infrastructure is owned and operated globally, built specifically for the latency and throughput demands of real-time programmatic decisioning. According to PubMatic, there are no cloud egress costs, no third-party dependencies, no latency penalties from infrastructure adapted for this purpose rather than designed for it. The company processes 2.7 trillion advertiser bid requests per day across more than 100,000 streaming channels, apps, and websites on the open internet, including 28 of the top 30 streamers.

That scale matters for model quality. Models trained and activated on PubMatic's bidstream operate on what the company describes as the most representative, diverse signal available anywhere in programmatic. The longer a model runs within the environment, the more data it accumulates. Performance improvements compound over time.

PubMatic launched AgenticOS in January 2026, positioning it as an operating system layer for autonomous advertising execution. Decision Fabric is a distinct but complementary capability within that broader infrastructure. According to PubMatic, partner models entering Decision Fabric do not operate in isolation - they enter an environment where AgenticOS is already running at scale, with more than 1,000 AI-powered deals active and agents operating across inventory discovery, deal management, and campaign optimisation. Partner decisioning models work in concert with that existing layer. The combination, PubMatic argues, produces outcomes that no single-layer system can replicate.

Brand safety scoring and yield optimisation are described by PubMatic as items on the roadmap. Each new capability makes the decisioning layer more sophisticated; each expansion is meant to compound the performance advantage for buyers already running within the environment.

The first live partners

Decision Fabric is live today with inPowered as its first active integration. Chalice AI is currently in onboarding, with additional partners described as being in the pipeline through the end of 2026. To participate, a partner needs a decisioning model, the ability to package it as a Docker container, and a focus on the US market for the initial integration.

Chalice AI is not a new entrant to this architectural territory. Adam Heimlich, Founder and CEO of Chalice AI, noted in a LinkedIn post today that the IAB Tech Lab's Agentic Real-Time Framework (ARTF) is currently live in three SSPs, with three more under development, and that several DSPs are also working with it. The IAB working group for ARTF now has over 100 members. Heimlich said Chalice has been active in this space for over two years.

"Chalice's modeling is per-advertiser, with advertiser-specific data modeled on the same advertiser's unique goals," Heimlich wrote in a comment responding to industry questions about AI alignment. "Our process is human-in-the-loop, never unsupervised. We've been doing this since 2020. We are correcting for the misalignment inherent in one-size-fits-all platforms."

That framing - per-advertiser models correcting for platform-level generalism - is precisely the use case Decision Fabric is designed to support. The container approach isolates each partner's model and logic from other participants in the environment, preserving proprietary IP while giving that model access to infrastructure and data it could not build or acquire independently.

Containerised decisioning as an industry pattern

Decision Fabric sits within a broader pattern that has accelerated significantly in 2025 and 2026. IAB Tech Lab formally named its umbrella agentic initiative AAMP - Agentic Advertising Management Protocols - on February 26, 2026, clarifying the relationship between ARTF and other components of the agentic advertising stack. ARTF, which IAB Tech Lab says cuts latency by 90% and improves agentic real-time decisioning for quality and trust, provides the technical foundation on which containerised deployments are built.

Index Exchange's Index Cloud is built in alignment with ARTF. Bedrock Platform became the first DSP to run its bidder inside an exchange - on Index Cloud infrastructure - announced on April 21, 2026. Amazon donated its Dynamic Traffic Engine to IAB Tech Lab as an open-source project on April 15, 2026, enabling DSPs to signal bidding priorities to SSPs and reduce queries-per-second waste - a different layer of the same efficiency problem. OpenX launched OpenXBuild in January 2026, reporting 70% cost-per-conversion reductions through APIs that let advertisers deploy custom logic directly into supply-side infrastructure.

The common thread is directionality: intelligence is moving toward the supply side, toward the data, toward the moment of the transaction. PubMatic has been building its own infrastructure in that direction since at least early 2026, through AgenticOS. Decision Fabric adds a new layer - one specifically designed to bring external partner models into that environment rather than only deploying PubMatic's own agents.

The question of who controls the decisioning layer is not trivial. As PPC Land documented in February 2026, PubMatic's Activate buying platform creates a direct path to agentic revenue that does not require dominant DSPs to adopt the Advertising Context Protocol - which Google, Amazon, and The Trade Desk have notably avoided. Decision Fabric extends that logic: if partner models run inside PubMatic's infrastructure, PubMatic becomes not just an SSP but the compute and data environment within which AI-driven programmatic decisioning happens.

What it means for buyers

For buyers, the immediate implication is a different kind of audience qualification. Real-time qualification against live signals - rather than last week's data compressed into a bid request - means campaigns reach the intended audience with greater precision. The impression pool partner models see within Decision Fabric is larger than what any DSP processes, because it precedes the traffic shaping that narrows what buyers can bid on downstream.

Whether that precision translates into measurable performance improvement at scale depends on the quality of the models partners bring to the environment, how those models are trained on PubMatic's bidstream data, and how effectively they integrate with the curated deals infrastructure through which segment IDs flow back into the auction. Chalice has approximately a dozen case studies available from its work in the containerised decisioning space over the past two years, according to Heimlich, which provides some basis for assessing what outcomes are achievable.

The onboarding process starts with a discovery conversation with PubMatic's product and partnerships team to confirm fit and walk through the data access and deployment framework.

Timeline

Summary

Who: PubMatic (Nasdaq: PUBM), a supply-side advertising technology company, announced Decision Fabric. The first live partner is inPowered. Chalice AI, founded by Adam Heimlich, is currently in the onboarding process. The announcement was authored by Punit Shah, Director of Product Marketing at PubMatic.

What: Decision Fabric is a real-time intelligence layer built inside PubMatic's auction infrastructure. It allows DSPs, curators, and algorithm companies to deploy their proprietary AI decisioning models as Docker containers hosted within PubMatic's infrastructure. Those containers receive full bid requests at auction time, apply the partner's decisioning logic, and return segment IDs in under ten milliseconds. The training dataset covers more than 60 OpenRTB fields across web, mobile, and CTV inventory in the US market.

When: Decision Fabric was announced and published on June 1, 2026. inPowered is already live on the platform. Chalice AI is in onboarding. Additional partners are expected through the end of 2026.

Where: Decision Fabric operates within PubMatic's owned and operated global infrastructure. The initial scope covers US ad requests only, with partner containers hosted within PubMatic's controlled data center environment.

Why: The structural argument is that the richest signals in programmatic transactions reside on the supply side, but decisioning has historically happened on the buy side, by which point signals have been compressed, filtered, and delayed. Decision Fabric moves partner AI decisioning to where the data actually lives - inside the auction, before traffic shaping, on the full signal. The approach addresses a long-standing inefficiency in how audience qualification works in programmatic advertising, where buyers have historically evaluated inventory against an incomplete picture of the available opportunity set.