Retail media is becoming one of the largest parts of the global advertising market. WARC forecasts that global retail media spend will reach $196.7 billion in 2026, overtaking combined linear and connected TV advertising. Its newer Future of Commerce Media 2026 report expects the figure to rise to $223.4 billion in 2027. For retailers, that growth puts more pressure on the systems behind sponsored search results and ad auctions.
Retail media is becoming a high-load systems problem
That is harder than simply adding an advertising service to an ecommerce platform. Sponsored products often appear inside search results or catalogs, so advertising logic has to run while the platform is already retrieving and ranking products.
Stepan Lyakhov, a backend engineer at Magnit OMNI, works with high-load retail systems, including marketplace search for a catalog of more than three million products and advertising infrastructure handling over 1,000 requests per second. He says an advertising failure should never become a shopping failure.
“If the ad system slows down, the user should still get the page they came for,” Stepan says. An ad request may depend on campaign settings and an auction before the platform can decide what to display, while the customer is still waiting for the underlying search results. If one component takes too long, waiting for it can push the whole request beyond the platform’s latency target. In that situation, skipping the ad may be better than delaying the page.
Ad auctions have to fit inside the customer’s latency budget
This makes latency limits part of the advertising architecture itself. Frequently requested information can be cached, while operations that do not need to happen during a live customer request can run asynchronously. “The architecture needs to either show an ad safely or skip it if the SLA won’t hold,” Stepan says. Campaign validation or data preparation, for example, can happen outside the path the customer is waiting on.
Search infrastructure becomes part of the advertising stack
Search creates another constraint because retail advertising often depends on the same infrastructure used to retrieve organic products. Before a search engine can rank the catalog, the system may need to identify a likely category or convert parts of the query into structured filters.
“The request has to be processed in 20–30ms: extracting stop words, guessing a likely category, or spotting a color in the query and turning it into a structured filter,” Stepan says. Moving some of this work outside full-text search can reduce the amount of processing left for the search engine.
The trade-off becomes more visible as retailers add search features. Typo correction and prefix matching can improve results, but each operation adds compute time. “More features mean more compute, and at some point the 200ms SLA stops holding,” Stepan says. If advertising has to make a decision within the same request, the available latency budget becomes even smaller.
The market’s scale makes these constraints more important. WARC reported in February that retail media is expected to account for around 30% of US digital advertising spend in 2026, reaching $107.6 billion. Search remains the dominant retail media format, while WARC’s April analysis found that display formats were still trailing search. That keeps much of retail advertising closely tied to search infrastructure.
Indexing and campaign updates must stay off the live path
Catalog updates create another source of load. Product prices, descriptions and availability change constantly, so search indexes may need to be rebuilt while the platform is still handling live traffic. Stepan says running heavy indexing work on the same index serving customer requests can hurt performance. One option is to build and validate a replacement index separately, then switch traffic after it is ready.
Batching updates can also reduce network calls when millions of product records are involved. At that scale, operations that would barely register in a smaller application can start to affect performance. The goal is to keep background processing away from the customer-facing path. Advertising campaign updates create a similar problem when they have to coexist with live auctions.
Ad relevance matters as much as delivery speed
There is also a quality issue that infrastructure alone cannot solve. An ad can be delivered within the required latency and still be damaging if it is irrelevant to the shopper’s query. “You have to keep monitoring ad quality constantly,” Stepan says. “An irrelevant ad can seriously damage metrics, both for the ad system and for search results.”
As retail media takes a larger share of advertising budgets, retailers are asking these systems to do more during the same customer interaction. WARC’s latest figures suggest that growth will continue, even as the market matures.
For engineering teams, retail advertising is becoming another high-load production system whose failures are immediately visible to customers.
The challenge is to scale monetization without making the shopping experience slower or less relevant.
