Can Adtech Afford the Infrastructure Behind the AI Boom?

Artificial intelligence is quickly finding its way into nearly every part of the advertising ecosystem. It is helping platforms build more sophisticated audience models, optimize campaigns in real time, generate and adapt creative, and make sense of the enormous volumes of data flowing through digital advertising. For an industry that has always competed on speed, intelligence and precision, the appeal is obvious.

But every new AI capability has an infrastructure cost behind it. That matters particularly in adtech, where the underlying technology was already doing an extraordinary amount of work before the current AI boom. Platforms routinely process enormous volumes of data in real time, scale around unpredictable traffic patterns and make decisions fast enough to participate in transactions completed in fractions of a second.

The infrastructure supporting those systems has historically required significant I/O, compute, storage and networking capacity. Real-time analytics, machine learning and programmatic advertising were already driving those demands before generative AI entered the picture. AI adds more intensive workloads to that environment and, as those workloads move from experimentation into production, makes the economics of running them increasingly difficult to separate from the economics of the products themselves.

For adtech companies, this is becoming one of the more consequential aspects of AI adoption. The industry is understandably focused on what AI can make possible. Increasingly, companies will also have to determine whether those capabilities can be delivered at the speed, scale and cost their business models require.

Marketing Technology News: MarTech Interview with Mark Listes, CEO @ Pendulum Intelligence

AI at Scale Changes the Math

During the experimentation phase, infrastructure costs can be relatively easy to overlook. A company might test generative AI for creative development, introduce an AI assistant or run a model against a limited dataset. Even an inefficient architecture can appear manageable when usage is small.

Production changes the equation. An audience intelligence capability that works well in a controlled pilot may eventually need to analyze billions of signals. A creative tool might go from producing a handful of assets to generating thousands of variations across campaigns, audiences and markets. Models used to optimize campaigns may run continuously as bidding conditions, inventory and consumer behavior change.

Recent research suggests those workloads will only become more pervasive. IDC predicts that by 2030, autonomous AI agents will manage a significant share of enterprise advertising operations, with agents increasingly making real-time decisions around segmentation, creative variation, budget allocation and channel mix. IDC argues that competitive advantage in this environment will depend in part on “architectural readiness,” as advertising organizations structure data for machines and build workflows and governance for increasingly autonomous systems. For adtech companies, that readiness also has an infrastructure dimension: as AI systems make more decisions continuously and at scale, the compute, storage and network resources underpinning those operations take on greater importance.

Each of those activities consumes resources, and they do not necessarily require the same infrastructure. Certain workloads benefit from GPUs, while others can run efficiently on CPUs. Some are memory intensive, some require substantial storage, and others are particularly sensitive to network performance. The infrastructure challenge is increasingly about matching resources to workloads and making sure expensive capacity is being used where it creates meaningful value.

That becomes particularly important in a high-volume business. Digital media companies have already experienced versions of this problem with other compute-intensive workloads: customer activity grows, infrastructure consumption grows with it, and operating expenses can begin to track revenue more closely than anticipated. AI potentially intensifies that relationship because inference can introduce recurring compute requirements into interactions that previously required far less processing.

At sufficient scale, those infrastructure choices can determine the margin profile of the AI product itself. What looks economical in a pilot can become considerably less so when the same capability is invoked millions of times across customers, campaigns and transactions.

Cloud Economics Become Product Economics

Adtech has dealt with infrastructure volatility for years. Campaign launches, elections, major sporting events and holiday shopping periods can create enormous swings in traffic, forcing platforms to accommodate sudden peaks without permanently provisioning for their highest possible level of demand.

The challenge is familiar across digital media. Some platforms have seen traffic increase several times over within minutes, making it impractical to treat peak demand as the operating baseline. The lesson for AI-era adtech is important: scalability is not simply about having access to more capacity. It is about understanding what capacity is consistently required, what is needed only during bursts and how economically the infrastructure can move between the two.

AI introduces another variable because the cost of serving an interaction can change depending on the model, processing requirements and underlying hardware. When an AI capability becomes part of a commercial product, those infrastructure costs begin to resemble product costs. Compute utilization, storage, networking and data movement can ultimately affect the cost of serving an individual customer, campaign or transaction.

That makes visibility and predictability more valuable. Companies need to understand what individual workloads consume, how infrastructure costs change as usage grows and whether expensive resources are sitting idle during quieter periods. A product can look attractive at one level of adoption and have very different economics when usage increases tenfold.

It also argues against assuming that every workload should be treated identically. Predictable, consistently utilized workloads may have very different infrastructure requirements from highly variable ones. Some applications may benefit from dedicated resources, while others are well suited to elastic public cloud capacity. In many cases, companies will use a combination of environments and adjust that mix as workloads mature. As workloads mature, that mix is likely to shift. Infrastructure decisions that made sense during development may need to change as usage patterns become clearer, giving companies an opportunity to optimize around actual demand rather than assumptions made at launch.

More Intelligence Still Has to Arrive on Time

Cost is only one constraint. Advertising technology operates under a performance requirement that makes the infrastructure conversation even more consequential: many decisions have to happen almost instantaneously.

Supply-side platforms can process millions of ad requests per second, while demand-side platforms analyze vast amounts of information to inform bidding and placement decisions. Adding AI to those workflows can make the decision more intelligent, but a more sophisticated answer has little value if it arrives too late to influence the transaction.

That creates a practical tension. Larger or more computationally demanding models can offer greater capabilities, but they can also require more processing time and infrastructure. The most useful model for an adtech application will not necessarily be the largest one available. It will be the one capable of delivering the necessary intelligence within the performance and economic constraints of the application.

Infrastructure architecture plays an important role in making that possible. Network performance, the distance between applications and data, storage performance and access to appropriate compute resources can all influence how quickly an AI-enabled application responds. A few milliseconds may seem insignificant in many enterprise applications. In an ecosystem built around real-time transactions, they can matter considerably.

The industry has encountered the same principle in digital content and media environments, where infrastructure must absorb dramatic traffic spikes without allowing the user experience to deteriorate. AI adds more processing into that chain, making architecture and workload placement even more consequential.

Model strategy and infrastructure strategy will therefore become increasingly difficult to separate. The sophistication of an algorithm matters, but so does the environment required to run it reliably at commercial scale.

Data Has a Geography

AI also adds another dimension to a challenge adtech companies already know well: data does not exist in a regulatory vacuum.

Advertising businesses frequently operate across countries and regions while managing large datasets used for targeting, measurement, analytics and optimization. AI increases the amount of data being processed and can create additional questions about where that processing occurs, particularly as models interact with datasets across multiple applications and environments.

As AI capabilities expand internationally, companies need a clear understanding of where their data is stored, where it is processed and how it moves between systems. Regulatory requirements may dictate that certain information remains within a particular jurisdiction. Enterprise customers may impose their own data residency requirements, while latency can make it advantageous to process information closer to particular users or datasets.

That means a single infrastructure model may not make sense for every workload or geography. Companies may need the ability to place applications and data according to a combination of regulatory requirements, customer expectations, performance and cost. The appropriate architecture could include public cloud, private environments, dedicated infrastructure or multiple providers depending on the business.

Flexibility is important because these requirements are unlikely to become simpler. As AI becomes more embedded in advertising products, infrastructure decisions made for performance or economic reasons will increasingly overlap with decisions about data governance and sovereignty.

Build for an AI Market That Will Keep Changing

Perhaps the biggest risk for adtech companies today is designing infrastructure around the assumption that the current AI landscape is settled. Models are evolving quickly and hardware is changing. Companies are experimenting with smaller and more specialized models alongside large general-purpose systems. New approaches are improving inference efficiency, while increasingly ambitious applications are creating new demands for compute, memory and networking.

The infrastructure that makes sense for an AI workload today may not be the best fit several years from now. Companies should understand how easily applications can move between environments, how dependent they are on particular infrastructure services and what it would take to shift resources if the economics or technology changes.

That does not mean every adtech company needs a sprawling multicloud architecture. Complexity has its own cost, and adding infrastructure environments without a clear reason can make operations harder rather than more resilient. The goal is to preserve options where those options have business value and avoid making today’s AI decisions unnecessarily difficult to revisit tomorrow.

Adtech has spent years becoming exceptionally good at optimization. Platforms continuously evaluate audiences, inventory, pricing and performance because even small improvements can have a meaningful financial impact when multiplied across enormous transaction volumes. The infrastructure supporting AI now deserves that same level of scrutiny.

There will continue to be intense competition around who can introduce better models, smarter optimization and more sophisticated AI-powered advertising products. But access to AI itself is unlikely to remain a durable differentiator as models and capabilities become more widely available.

The harder question will be who can make the economics work. Adtech companies that understand the relationship between workload, infrastructure and cost will be better positioned to turn AI experimentation into sustainable products. In the next phase of adtech, competitive advantage may come less from simply having access to powerful AI and more from the ability to run it efficiently, reliably and economically at scale.

About The Author Of This Article

Jeffrey Gregor, is General Manager at OVHcloud US

About OVHcloud

OVHcloud is a global, cloud provider that offers businesses industry-leading performance and value. Vertically integrated with its own fiber optic network, water-cooled server technology, and green data centers, OVHcloud operates more cost effectively and passes these benefits on to customers. Controlling each step in the value chain allows OVHcloud to provide the best performance, price, security, and customer service in the industry. OVHcloud is a subsidiary of OVH Group SAS, the largest European cloud provider, with more than one million customers across 138 countries and four continents.

Marketing Technology News: How MarTech Is Enabling Autonomous Brand Engagement Across Channels?

source https://martechseries.com/mts-insights/guest-authors/can-adtech-afford-the-infrastructure-behind-the-ai-boom/