How AI Hype is Fueling Data Center Transformation

TEKsystems’ Ram Palaniappan offers insights on how AI hype is fueling data center transformation. This article originally appeared on Solutions Review’s Insight Jam, an enterprise IT community enabling the human conversation on AI.

Earlier this year, Google Cloud, AWS, and Microsoft Azure eliminated egress fees within three months of each other — a surprising move directly tied to the surging demand for AI solutions.

All of the cloud service providers (CSPs) wanted to win the AI revolution. The elimination of egress fees ended what was essentially a tax on moving data out of an individual CSP, enabling massive data transfers to feed data hungry AI systems. For enterprises, this allows for experimentation with AI solutions without fear of unpredictable data transfer costs.

However, AI-driven transformation isn’t limited to public cloud infrastructure. These same forces are reshaping an element often left out of conversations around hyped-up IT trends: colocation data centers.

The Role of Colocation Data Centers in the AI Revolution

The major CSPs’ elimination of egress fees is an acknowledgment that we now live in a multi-cloud world.

Historically, CSPs could reasonably expect that a customer would host almost all their cloud infrastructure on a single service. But today’s organizations are recognizing the value of a best-of-breed approach, leveraging each cloud for its specific strengths. For example, a company might want to manage their databases on Oracle, but run applications in Azure.

In response, hyperscalers are increasingly specializing in distinct use cases. Google has led this trend, pouring R&D resources into its AI capabilities. In fact, Google Cloud was the first CSP to drop egress fees, enabling customers to use these AI features without committing to Google Cloud for data storage.

As the late entrant in the cloud space, Google had little to lose from this move, and significant market share to gain. The other major players had no choice but to follow suit to remain competitive.

So, is there a place for colocation data centers in this new multi-cloud world? Of course — because there are areas where they are best of breed, as well.

Sensitive data and applications may need to live behind a company’s firewall for security and compliance reasons. And in some cases, applications might demand special server requirements that aren’t available in cloud data centers or make running that application in the cloud prohibitively expensive. In these situations, renting infrastructure at a colocation data center gives enterprises the security and control that sensitive data deployments lack.

However, to run AI workloads in these environments, colocation data centers have to level up — and so do the strategies enterprises use to manage them.

Considerations for Deploying AI in Colocation Data Centers

Running AI workloads in a colocation data center reintroduces some of the complexity cloud computing typically simplifies. Enterprises must consider factors like cooling technologies, energy sources and hardware needs in more granular detail than they would with an AI deployment.

Here are a few considerations to keep in mind:

1. Energy consumption: Data center energy use in the U.S. is projected to double by 2030, fueled largely by compute-heavy AI workloads. In areas with a high density of data centers, such as northern Virginia or Phoenix, AZ. This increase may seriously strain the power grid for domestic consumption. To avoid these impacts, be strategic about where your colocation data centers are distributed, and don’t overcommit to any geographic area.

2. Environmental impact: High energy consumption gives AI-ready data centers an extremely large carbon footprint. In addition, to drive optimum operation parameters, GPUs need liquid cooling technologies that divert another natural resource, water, to service AI workloads. These impacts can be counterproductive to enterprises’ environmental, social and governance (ESG) goals. Mitigate them by looking for colocation centers that emphasize the use of renewable energy or other tactics to reduce environmental impact.

3. Refining AI use cases: We are approaching the peak of the AI hype cycle, with enterprises racing to implement AI models across almost any use case they can. As the hype cycle plateaus, enterprises will begin to strategically determine which tools, technologies and applications need to use GPUs vs. TPUs vs. CPUs. For example, a search with an AI tool like ChatGPT or Gemini uses 6-10x the energy consumption compared to conventional Google search. Enterprises will need to determine which queries are complex enough to merit the richer context AI can give, and which are more efficiently routed through normal channels. Optimizations like this can reduce the burden on AI infrastructure in data centers in the long term.

Beyond the AI Hype Cycle

As AI continues to transform both cloud and data center infrastructure, enterprises should tread carefully. An effective, future-proofed AI strategy needs to consider not just the costs of different options for data storage and compute, but also considerations like sustainability and resource availability.

In the long term, organizations that intelligently optimize workloads and infrastructure for their specific needs will be most likely to reap the gains of the AI revolution.