Your Competitors Are Serving the Same Ads for 30% Less. The Difference Is the Server.

Ad tech teams optimize the code. They rarely question the server. Here is what that costs—and what one platform changed to fix it.

AT A GLANCE

• Shared cloud servers run other companies’ workloads alongside yours. That competition slows your auctions.
• When traffic jumps 10x in an hour on shared hardware, auctions slow down and the cost per ad served goes up.
• Dedicated bare metal removes the overhead that makes performance unpredictable—without rebuilding everything.
• display.io cut its cost per ad by 30%+ and held auction times under 200ms globally after moving to Servers.com by Nexcess
• Five server decisions separate platforms that win bids reliably from those that lose them at the margin.

The Auction Fires in 200 Milliseconds. Your Servers Don’t Keep Up.

It is Tuesday at 2:47 PM. A user opens a mobile app and scrolls to a monetizable placement. In the next 180 milliseconds, your platform fires a bid request, gets responses from eleven buyers, picks a winner, and returns a filled ad. The user notices nothing. Your advertiser pays for all of it. Your servers spend 15 to 20% of that time on overhead—work that has nothing to do with your auction. That is the cost of sharing hardware with other companies’ software. It does not appear as a line item in your cloud bill. It shows up as bids you were competitive on but still lost, and as traffic peaks that should be your best hours becoming your worst.

“We’re fast, just not reliably fast—and in this business, unreliable fast is the same as slow.”
— CTO, demand-side platform

Most teams that hit this wall have already optimized everything in the code. The problem is not the software. It is the server model underneath it.

Five Server Decisions That Separate Winners From The Rest

What matters most is what happens to a bid request in the 12 milliseconds between arrival and response. Five server choices control that window.

1. Keep auction servers separate. Bid processing and auction logic should run on their own hardware. When those jobs share resources with background data processing or AI model training, your auction speed drops on a schedule you cannot predict. Dedicated servers remove that competition.

2. Plan for sudden traffic surges. Spikes in ad tech are not exceptions—they are the job. A live event, a campaign launch, a breaking news moment. If your servers fall apart under heavy demand, your best revenue windows become your worst. Multiple dedicated machines absorb spikes without slowdowns from shared server space.

3. Put your servers close to your buyers. Sub-200ms auction clearing is impossible from one location when your demand sources span three continents. The bid request travels. The response travels. Every extra millisecond costs you viewability. Servers in the right regions fix that. 

4. Make problems fast to investigate. When viewability dips or you need to track down a specific ad, every minute spent searching is a minute the issue goes unfixed. Tracing a problem through shared cloud infrastructure can take hours. Extended log retention and a support team that responds in minutes—not days—lets you investigate fast and resolve issues before they cost you.

5. Treat cost per ad served as a number you can move. What you pay per ad served comes down to how your servers are set up. Dedicated bare metal removes the overhead layer between hardware and code, putting more server power into your auction. That, plus lower data transfer fees, can cut per-ad cost by 30% or more. The savings scale with you.

What display.io Did—and What You Can Learn From It

display.io fills in-app ad placements for publishers worldwide. At peak, the platform processes 1 million events per second and must return a winning ad in under 200 milliseconds. Before moving to Servers.com by Nexcess, its servers were not built for that scale. The cost per ad kept climbing. Support took a week to respond.

“I was messaging our account manager and getting an answer within a week. Sometimes they just forgot to answer me entirely.”
— Roman Gushel, CTO, display.io

After moving to dedicated bare metal, display.io ran faster processors and spread workloads across multiple machines. Cost per ad dropped by more than 30%. Performance held globally, even when traffic spikes ten times.

KEY RESULTS

• Cost per ad served: Down 30%+; up to 2x on specific server configurations
• Auction speed: Under 200ms across the US, Europe, and Asia
• Traffic spike handling: 10x capacity absorbed in minutes
• Support response time: 7 days by email → under 10 minutes on Slack
• What comes next: An AI layer to rank the most valuable traffic, built on dedicated hardware

Years of software optimization had already done their work. Past that point, the next meaningful speed gain came not from the code but from the hardware under it—removing the overhead layer between the server and the code that was already working. When that layer is gone, auction times stop varying.

“We feel that Servers.com is genuinely interested in us as a customer and as a partner. That’s a drastic difference from what we had before.”
— Roman Gushel, CTO, display.io

What This Looks Like for Your Platform

The display.io results are repeatable. Here is what the same decisions look like by platform type.

• If you sell ad inventory on behalf of publishers

Publishers often run a second auction on their side with latency cutoffs as tight as 10ms—so even 10ms of overhead can mean lost revenue. Beyond latency, track your metrics across many dimensions. Owning your stack is the best way to control the dimensionality and cardinality of your data, giving you the depth for analysis and the speed to act on it.

• If you buy ad inventory on behalf of advertisers

Most exchanges and advertisers will send you as much traffic as you can handle, so your infrastructure has to absorb hundreds of thousands of QPS, scale horizontally, and route each request to the right demand source. Reacting to advertiser behavior in real time is critical to sustainable revenue—and running your machine learning models on specialized hardware makes bid price prediction faster and more competitive.

• If you run ads inside mobile apps

In mobile advertising, your users can be anywhere in the world, so delivering consistent response and ad-serving speed across that footprint is critical. A provider that gives you an abstraction layer for cross-data-center connectivity makes that consistency far easier to achieve. The pattern is the same across all three: revenue is decided at the millisecond, at scale, on data you control, and across a footprint that spans the globe. Start with the workloads where those constraints bite hardest and move them to infrastructure built for ad tech—not borrowed from general-purpose cloud.

Every Bid You Lose at the Margin Is Revenue Your Competitor Captures

Ad tech teams that keep growing will treat server efficiency as a business decision, not just an engineering one. Cost per ad served is a margin number. Auction consistency is a retention number. Software and infrastructure shape both—and once your code is well-tuned, the hardware underneath becomes the largest remaining lever. 

Tuning is always an option. But past a certain point, the biggest gain left is the server itself.

About Servers.com by Nexcess
Servers.com by Nexcess is a bare metal infrastructure platform built for streaming media teams. It operates 28 ISO-certified data centers across the US, EU, UK, Singapore, and Hong Kong. 18,000+ devices deployed. 1,000+ partnerships. Average support response under 15 minutes. Dedicated bare metal, elastic cloud capacity, no hyperscale bill.

Ready to see what your platform can achieve? Talk to Servers.com by Nexcess