When we introduced Decision Fabric in June, we described a new capability: partners deploying proprietary decisioning models inside PubMatic’s auction, running inference on live bid stream signals at the exact moment inventory is available. Curation was the first use case. Containers were the mechanism. But the market response told us we had touched something bigger.
The idea underneath Decision Fabric was never about only containers. It was about where intelligence belongs in programmatic advertising.
The Distributed Problem
The industry has spent two decades distributing decisioning across the supply chain—a hop here for audience qualification, there for traffic shaping, another for curation, another for optimization. Trace a single impression from budget to auction, and it crosses ten, twenty, even twenty-five decision points. Every decision pays a tax in latency and compute, running on a smaller, later version of the truth than the one before it.
That architecture made sense when compute was cheap and models were simple. Neither is true anymore. So we expanded the definition to match the ambition.
The Expanded Decision Fabric
Decision Fabric now names and unifies all decisioning already happening inside PubMatic’s platform—far more than any single product describes. Audience qualification, traffic shaping, signal filtering, inventory quality, price discovery, yield optimization, spanning every channel and customer type we serve.
It is PubMatic’s intelligence layer: every decisioning model running on our platform, across every use case, operating where data is dense and latency is low.
That intelligence comes from two sources.
Partner Models. Decisioning logic from DSPs, curators, agencies, and algorithm companies, deployed as containers inside PubMatic’s infrastructure. Curation and traffic shaping containers are currently live. Partners bring the model; their logic and IP stay sealed inside the container, applied at the supply source instead of after a network round trip. This capability keeps growing.
PubMatic Models. Native intelligence we build from a vantage point no one else has: we see every auction, every bid request, unfiltered, before traffic shaping narrows downstream visibility. Our models deliver audience reach and performance, infrastructure efficiency, and price discovery and sell-side intelligence—through Inventory Quality Sync, Audience Fastlane,
The flow runs in both directions. Any participant can consume intelligence, contribute it, or both. A DSP might consume curation intelligence to buy smarter while contributing its own traffic shaping model back into the platform.
Partner Models plus PubMatic Models.
Why This Moment
Three forces converged.
- Enterprise AI budgets jumped more than a third in the past year; cloud spending has passed $400 billion and is climbing. Every wasted impression—fraud, bad placement, off-target ads—is no longer hidden waste. It directly hits compute bills and margins. Buyers and DSPs demand filtering before cost impact, requiring decisioning closer to inventory, where signal is richest.
- Container technology is now enterprise standard. Mature orchestration runs in most large organizations. This changes specialization economics: algorithm companies no longer need entire ad tech stacks. They package models, deploy into our infrastructure, and instantly inherit our inventory scale, data density, and latency advantage.
- For years, decisioning spread thin across systems built for other jobs. The AI era rewards depth: companies excelling at their core competency. Curators own curation logic. Algorithm companies own the competitive edge that uniquely defines them—fraud detection, optimization, whatever they choose. Data companies own signal. The market wants distributed decisioning on one platform—where each specialist stays specialized, their intelligence executing alongside everyone else’s.
Our Advantage
Three things make PubMatic different.
- Our compute is owned, operated, and purpose-built for ad tech inference at scale.
- We operate across CTV, mobile app, web, and online video simultaneously with inventory from 2,000+ premium publishers and 28 of the top 30 streamers.
- We sit at the intersection of supply and demand, seeing every auction, every bid request. Identity, commerce signals, content context, and auction dynamics converge at the moment of decision.
Work With Us
If you have decisioning logic of your own—or you’re thinking about where decisions should run—we want to talk.