Inside 152 Media: Building technology that puts publishers first

Server Costs

Since 2021, the rapid expansion of AI workloads has fundamentally reshaped the economics of cloud computing. According to McKinsey & Company, the widespread adoption of generative AI and advanced machine-learning models has shifted cloud infrastructure toward GPU-intensive, energy-dense systems, significantly increasing baseline infrastructure costs.

In parallel, Scope3 highlights how these cost pressures are amplified in programmatic advertising due to structural inefficiencies. A single ad impression can trigger multiple redundant, AI-driven evaluations across the supply chain, multiplying compute usage without a proportional increase in value. As a result, AI raises the baseline cost of cloud infrastructure, while ad-tech inefficiencies magnify that cost at scale—directly increasing operating expenses for platforms, DSPs, and publishers.

Optimizing Publisher Revenue in This New Paradigm

The practice of routing every bid request across multiple supply paths has given rise to bid throttling—a mechanism that restricts the volume of bid requests sent to DSPs, especially when several paths point to the same impression and risk duplicate bids.

When DSP-side throttling occurs, valuable requests can be filtered out simply because they are bundled with low-performing traffic. This negatively impacts demand discovery and revenue. To avoid this, request selection must occur earlier in the supply chain.

A supply-side throttling approach allows publishers to offload unnecessary compute from DSPs while better aligning inventory with demand signals. This is the motivation behind BidLift, a Prebid RTD module that enriches bid requests with additional signals, corrects missing or invalid fields (such as supply chain errors), and makes real-time decisions on whether a request should be sent, routed through a specific supply path, or discarded. These decisions are driven by both user-level (individual) and publisher-level (global) performance signals.

Common Dynamics in Header Bidding

Header bidding is a programmatic advertising technique in which publishers offer each ad impression to multiple exchanges simultaneously. The most widely used library for this process is Prebid.js.

In a typical header bidding flow, bid requests are sent to multiple bidders, who may respond with bids or choose not to bid. The highest bid is then passed to the ad server, where it competes against the ad server’s own demand sources and decision rules to determine whether it is ultimately rendered.

The following framework illustrates the possible outcomes for a single bidder based on two variables: CPM (cost per thousand impressions) and auctions won (the number of header bidding auctions that result in bids forwarded to the ad server). The product of these two variables represents the bidder’s potential revenue. This creates multiple outcome regions, where higher CPM and higher win rates correspond to greater revenue potential.

Revenue Levers

There are multiple strategies to increase revenue, but the two fundamental levers are:

1. Increasing CPM
2. Increasing the number of auctions won

While CPM and auctions won are correlated—higher CPM generally increases the likelihood of winning—this relationship is not linear or guaranteed. Bidders do not participate in every auction; selective bidding and no-bid responses are common behaviors across many auction types. As a result, improving revenue requires not only higher bids, but also smarter request selection and routing to ensure high-value opportunities are delivered.

Forces That Limit Revenue Growth

Case (A): Forces That Resist Increases in CPM

Several structural forces limit a bidder’s ability to simply increase CPM:

● Bid throttling: Responding only to a small subset of bid requests with very high CPMs can result in a low impressions-per-bid-request ratio.

● Demand constraints: DSPs are often unable or unwilling to raise CPMs beyond certain limits due to campaign objectives, budget pacing, or bidding strategies tied to performance goals.

As a result, CPM increases alone do not guarantee higher revenue and may reduce overall auction participation.

Case (B): Forces That Resist Increases in Auctions Won

Similarly, increasing the number of auctions won faces its own constraints:

● Auction dynamics: Winning auctions is inherently competitive and highly dependent on CPM relative to other bidders.

● DSP optimization: Maintaining a very high response rate with a consistently low win rate can negatively affect DSP bidding algorithms. Over time, this may lead to reduced participation as DSPs deprioritize inventory perceived as inefficient or uncompetitive.

Thus, maximizing auction wins without improving bid quality can also be counterproductive.

Real Data: Observed Market Behavior

In the following figure, we observe a common pattern in a heterogeneous inventory where nine Prebid bidders participate. Most bidders fall below the diagonal:

From this distribution, we can identify three strategic paths for an individual bidder:

1. Case (A): Increase CPM while keeping auctions won low This reflects a highly selective bidding strategy: low response rates paired with high CPMs that occasionally clear the auction. While this can produce results, it limits scale and overall revenue potential. It is susceptible to bid throttling.
2. Case (B): Increase auctions won while keeping CPM low This strategy relies on a high response rate with low CPM bids, clearing auctions primarily when no higher-value bidders are present. While it can increase win counts, it often undervalues impressions and typically results in lower revenue for publishers.
3. Case (C): Increase both CPM and auctions won This is the most interesting outcome and the one that maximizes revenue. Achieving it requires a combination of improved bid valuation and sustained bidder participation. This is the area of focus at 152Media: increasing bid value while maintaining a healthy response rate by reducing the causes of bid throttling that prevent DSPs from bidding on inventory. Inventory is consistently presented with higher perceived value per impression.

Conclusion

In an AI-driven, high-cost cloud environment, scale without efficiency is no longer viable. CPM and auction wins cannot be optimized independently—demand constraints limit price increases, while auction dynamics restrict win rates. The only sustainable path forward is upstream intelligence. By filtering, enriching, and routing bid requests before they reach DSPs, solutions like BidLift reduce wasted compute, mitigate bid throttling, and align inventory with true demand. The result is a market where higher CPMs and higher win rates coexist—unlocking durable revenue growth for publishers.

Andrés Ajras – 152 Media Developer