Retail media is projected to hit $300 billion in global ad spend by 2030, with over 80% of top U.S. retailers already operating an ad network. While that growth is exciting, most of it happened on the marketing side of the business. The infrastructure supporting it is struggling to keep up with the unique technological needs of an ad delivery business.
A real-time bidding system needs to run on top of a live storefront and make decisions in milliseconds without slowing anything else down. It’s not just another page on the site.
Let’s look at what makes that kind of infrastructure different, what high-performing networks do differently, and what it takes to run one at adtech-grade performance.
What is retail media?
Retail media is advertising sold on a retailer's own digital properties: the search results, the product pages, the app, and sometimes the checkout flow. A brand pays to get a sponsored placement in front of a shopper who's already there to buy something, and the retailer sells that placement the way a publisher sells an ad slot.
The simple definition, however, belies the complexities of the adtech infrastructure that needs to run underneath.
Most retail media doesn't run on a separate media property with its own traffic patterns. It runs directly on top of the storefront that's already serving customers, which means the ad decisioning system and the ecommerce platform are now competing for the same compute, network, and uptime budget.
Early retail media pioneers like Amazon, Walmart, Instacart, Target, and Kroger have spent years building the custom architecture for traffic and first-party data to make it work. For retailers that want to get into the retail media space and don't have the same infrastructure experience, there are a few things you need to know.
How retail media ads are served
A shopper opens a search results page. Before that page finishes loading, an ad auction runs: eligible advertisers bid for the placement, a winner gets selected, and the ad gets inserted into the page the shopper is about to see.
All of that has to happen in under 100 milliseconds, because the ad request can't be allowed to hold up the page it's supposed to appear on.
That's the onsite case, and it's the hardest one from a performance standpoint because it's happening live, in the critical path of a page load. Offsite retargeting, which follows a shopper to another site or app with an ad based on retailer data, and in-store digital placements run on different timing and different infrastructure, but they still draw on the same retailer data and the same ad-serving backend.
Teams often plan for one of these cases and get surprised by the load the others add later.
Why retail media is a harder infrastructure problem than it looks
Two things make this workload unusual: the performance bar and the cost curve.
- The performance bar: A retail media auction has a hard latency budget, and every hop from shopper, to ad server, to exchange, back to the page, eats into it. A checkout page that loads half a second slower is annoying. An ad decision that arrives late doesn't get shown at all, and that's lost revenue for every ad that is not shown.
- The cost curve: Retail media spikes around flash sales, holiday shopping, and product launches, exactly when ad revenue is highest and the cost of an outage is worst. Nearly two-thirds of CPG marketers plan to increase retail media spend, which means the traffic these systems need to handle is still climbing. Infrastructure built for an average day gets tested on the worst day, and that's usually the day it matters most.
What high performing networks do differently
They give ad decisioning dedicated capacity.
Split the ad serving functionality onto server compute infrastructure that's isolated from the storefront, so a surge in shopping traffic and a surge in bid requests don't compete for the same resources.
They plan capacity around peak days.
Size infrastructure for the traffic you expect to see during flash sales and holidays. Use auto-scaling to size up for peak and back down for normal traffic days.
They set a latency budget and monitor every step to it.
Treat the sub-100 millisecond bid response as a firm target, not an aspiration, and track every hop between the shopper's request and the returned ad. That end-to-end view is what catches a slow route to an exchange before it shows up as lost fill rate. To meet sub-100 millisecond times, deploy your ad servers across multiple regions to mirror your highest traffic area’s and get the servers closer to the user, minimizing hops. Use high-performance load balancing and route user requests to the nearest edge location or point-of-presence. Cache key data (like fraud lists, active ad inventory, etc.) directly in RAM rather than on a hard or solid-state drive.
They assemble the stack holistically.
Consolidate infrastructure and reporting for each placement type behind a single operational layer. Adding a new channel then means extending that layer, not building and maintaining a parallel system for it.
Build it, buy it, or something in between?
Retailers choose from three paths.
| Approach | Speed to launch | Performance control | Cost predictability | Engineering lift |
|---|---|---|---|---|
| Off-the-shelf retail media platform | Fast | Low | Low | Low |
| Fully in-house build | Slow | High | High | Very high |
| Hybrid: your own infrastructure, a licensed platform layer | Moderate | Max | High | Moderate |
The out-of-the-shelf option gets an ad-system up and running quickly, but you're renting someone else's infrastructure decisions along with the software. Building fully in-house gives you complete control, but requires time and a large engineering team to support it.
The hybrid path may be the sweet spot. You license or write the auction logic, but control the underlying hardware stack.
Retail media has real challenges too
Most of the talk about retail media focuses on the upside: the market is growing, more retailers are launching networks, more brands are spending. All of that is true. But it’s important to be clear about the demands too.
Building a network pulls engineering time away from the storefront and checkout roadmap, and that doesn't end once the network launches. Keeping it fast and reliable takes ongoing engineering time that's easy to leave out of the original budget.
That doesn’t mean you should skip retail media, but it will help to plan for this trade-off from the start.
The non-negotiables for "adtech-grade" infrastructure
Whichever path a retailer takes, the underlying infrastructure needs a few specific things:
- A, ad decisioning process that comfortably executes inside 100 milliseconds, end to end
- Compute that scales up for a flash sale and back down the next day, without a manual provisioning cycle in between
- Dedicated resources for ad serving, so a traffic spike on the storefront doesn't starve the ad decisioning system (and vice versa)
- Network presence close to the exchanges and DSPs the network talks to, since distance adds latency that no amount of server tuning can remove
- Cost that scales with traffic in a way finance can actually forecast
FAQs
What is a retail media network?
A retail media network is an advertising system built on a retailer's own digital properties, like search results and product pages, that lets brands pay for sponsored placement in front of shoppers who are already there to buy.
How is retail media infrastructure different from standard ecommerce hosting?
Ecommerce hosting serves pages. Retail media infrastructure has to run a live auction and insert a decision into that page before it loads, inside a latency budget usually well under 100 milliseconds.
How much latency can a retail media platform tolerate before it affects revenue?
A retail media platform can tolerate very little latency before feeling the revenue impact. Bids typically need to be requested, returned, and placed in under 100 milliseconds. Past that, the ad slot either shows a default or gets skipped, and that's lost revenue on that request.
Do retailers have to build their own infrastructure to run retail media?
No. Off-the-shelf platforms exist and can get a network live in days. What you give up is control over performance and cost, since you're working within someone else's infrastructure decisions.
What usually causes a retail media platform to slow down or go over budget?
Shared compute between the storefront and the ad server, capacity planned for average traffic instead of peak events, too many network hops between the server and the user and infrastructure assembled piecemeal across vendors with no single owner for performance or cost.
Getting started with retail media
Retail media only works if the ad shows up on time, every time, even when traffic spikes.
If dedicated, scalable infrastructure is the gap, that's exactly what servers.com's adtech solutions are built for. By running ad-serving logic on single-tenant bare metal, you gain total hardware isolation, low latency connections to major exchanges, and predictable hosting costs that scale cleanly with your business.
PS: If you're at DMEXCO this year, retail media infrastructure is one of the conversations we're having on the ground. Come find us at booth C028 in hall 06.1 if you want to talk through what your network's infrastructure needs to look like before your next peak event.
About the author
