How ad tech’s cloud transformation can keep up with its automation goals

Early in 2025, AdExchanger declared that this year, ad tech would go to the clouds. In many ways, this has proven true: across the industry, platforms are migrating workloads to cloud environments, modernizing their data pipelines, and reengineering their backend systems for scale.

But migration alone doesn’t always mean transformation.

As the industry doubles down on automation (from creative optimisation to AI-driven media buying), a critical gap has emerged between ambition and infrastructure. This is because real-time decisioning at programmatic scale demands cloud-native architecture uniquely designed for low-latency processing, cost efficiency, and intelligent filtering at the impression level.

And right now, many platforms simply aren’t up to the task. 
 

Where legacy infrastructure can turn automation into a tax

Everywhere you look, ad tech is marketing itself as smarter, faster, and more automated: DSPs offer intelligent bidding, SSPs offer curated deals, and AI models are optimsing toward outcomes. 

Unfortunately, many legacy systems weren’t designed to make automated decisions based on large-scale, contextual, and placement-level data. Instead, they batch or approximate. This might work for bulk reporting or audience segmentation, but stops short of filtering impressions dynamically in real time — especially when multiple platforms are stitched together in a non-native cloud environment.

Worse still, many “cloud transformations” boil down to the same old codebase running on virtual machines instead of physical ones. These rebrands might check a box, but they don’t enable the real-time processing, dynamic filtering, or placement-level optimisation that modern programmatic workflows demand.
 

Why real-time decisioning is raising the bar

In today’s media ecosystem, milliseconds matter. Buyers want instantaneous access to curated paths that capture attention and deliver predictable ROI. 

To deliver this, however, the sell side needs to filter billions of bid requests per day using nuanced signals like time in-view, ad refresh rate, ad-to-content ratio, and semantic alignment. Doing so is a massive computational task — and not one that works well on infrastructure that was built for bulk delivery or manual packaging.

To truly support automation, cloud infrastructure needs to support real-time scoring, pre-bid filtering, and campaign-specific logic rather than just pass along impressions to the DSP and hope for the best. But how?
 

How to make cloud-native design the foundation of optimisation

This is where cloud-native infrastructure makes a difference. Systems built from the ground up for distributed processing and real-time traffic shaping can ingest, analyse, and act on bidstream data in milliseconds, with enough headroom to optimise for each buyer’s specific goals.

This is a far cry from the “dumb pipes” many supply platforms still operate on and it’s why curation is evolving from packaging to infrastructure - a shift that only works when your foundation is fast, intelligent, and purpose-built.

If 2025 is the year ad tech finally embraces the cloud, then 2026 needs to be the year it uses it properly. Here are four guidelines for doing just that:

1. Architect for real-time. A lot of cloud moves focus on horizontal scale; however, successfully automating at the impression level depends largely on latency. Always remember that cloud-native systems need to evaluate and respond to bidstream signals in milliseconds, not minutes or post logs.
    
2. Get closer to the bidstream. Because optimisation starts upstream, platforms should be able to process placement-level signals (think GPIDs, time-in-view and ad-to-content ratio) and use them to make informed pre-bid decisions. The closer that logic sits to the request, the more relevant and efficient the outcome.
    
3. Design for infrastructure efficiency. Admittedly, real-time automation is computationally expensive. Infrastructure that prioritises efficient processing over brute-force volume enables more intelligent filtering, faster experimentation, and higher performance with fewer waste cycles.
    
4. Use AI that learns from campaign signals (not just patterns). Plugging in generic ML scoring isn’t the same as aligning AI to buyer-defined outcomes. If you’re looking for better results and not just cleaner delivery, your traffic shaping models have to evolve with live feedback and optimise with an eye on attention, relevance, or conversions.

Keeping these guidelines front of mind will help advertisers lead cloud transformations that actually deliver outcomes instead of racking up costs.

Because automation that doesn’t deliver on the promises of the cloud isn’t progress. It’s just bad marketing.