More AI workloads are moving back into enterprise and colocation environments. But many of these environments were built long before AI was in use; their power and cooling profiles can’t handle the density. Now what?.
For many, AI started in the cloud, but it isn’t always staying there. Data center and infrastructure teams are realizing the latency, cost and control tradeoffs that result when all AI processing is pushed to hyperscale facilities.. Accordingly, more AI workloads are moving back into enterprise and colocation environments.
But there’s one problem with this shift: These environments were built long before AI was in use, and their power and cooling profiles can’t handle this level of density.. Not all organizations have the capital (or the desire) to construct new data center spaces every time AI demand spikes.
And, especially within enterprises, their teams don’t have the expertise or time to design new high-density solutions for each new project.. Factors to consider when deciding where AI will live. When enterprises started to deploy AI, it initially made sense to lean heavily on cloud and hyperscale providers.
Capacity was available, resources could be scaled quickly and experimenting with pilot projects was fast and low risk.. As data center and IT leaders decide which AI workloads should run where, three factors need to be considered: how quickly AI services need to respond, what kind of data they interact with and how predictable their long-term cost and capacity needs must be..
How quickly does AI need to respond?. – Getty Images. Many AI applications are latency-sensitive, which means they depend on fast, consistent response times to be effective. This includes applications like:.
Fraud detection in financial transactions. Smart building controls that adjust heating, cooling and lighting based on occupancy and environmental conditions. Real-time personalization in customer-facing digital experiences.
Smart campus operations that coordinate energy use, space utilization and services across buildings. In these instances, sending data to a remote data center and waiting for a response can add delays that operations can’t tolerate.. Bringing a portion of the AI stack closer to where critical data is generated and consumed helps reduce latency so performance and user experience can be maintained..
Where does it make the most financial sense for AI to run?. Cost is another factor bringing AI back to enterprise and colocation facilities. High-density GPU clusters that run 24/7 can be expensive to host in the cloud..
Many organizations reach a point where it makes financial sense to keep certain work in hyperscale environments while relocating mission-critical workloads into infrastructure they control. The reason: more predictable cost and capacity planning. Teams can better forecast spend across hardware lifecycles, avoid surprise consumption spikes and manage budgets and upgrades..
What types of data will AI need to use?. AI relies on training and inference to turn raw data into decisions that matter to the business. These activities involve large, sensitive datasets that enterprises don’t always want to move to external environments due to regulations, internal policies or even customer expectations..
Running AI in enterprise or colocation data centers helps organizations protect critical data by keeping it close, simplifying compliance and reducing the operational overhead associated with moving data back and forth.. Adapting existing data center spaces for AI. Bringing AI in-house means asking for spaces that have fixed power, cooling and space constraints to support a new set of requirements..
While some data centers may have room for more racks, these racks are now filled with GPUs, which draw several times more power per rack than the space was built for. Finding another 20 kW to 40 kW and beyond, and removing the associated heat at the racks is a different story. Traditional cooling sized for legacy loads often struggles to keep high-density AI clusters within safe operating ranges without overcooling the rest of the room or creating hot spots.. “As densities climb, taking a rack-level approach to power and cooling is critical” Brian Kennedy, Belden.
Bringing cooling to the rack supports AI. As densities climb, taking a rack-level approach to power and cooling is critical; AI clusters concentrate more power into fewer racks. Instead of asking the whole room to absorb the heat load, rack-level cooling moves this handoff point to the rack..
Rear-door heat exchangers attack the problem at the source. A cooled rear door captures and removes the heat as it leaves the rack instead of letting hot exhaust spill into the room.. For retrofit projects especially, this is a big deal.
Rear-door and other rack-level solutions can often be added to existing rows with minimal disruption inside a live environment. You can increase density exactly where and when you need it to support critical workloads.. How Belden enable AI-ready infrastructure.
Belden enables AI-ready infrastructure by delivering the robust connectivity foundation required for high-bandwidth data flows, low-latency communications, and reliable operations across industrial sites, buildings, and data-centric facilities.. An AI workload is only as good as the data feeding it.
Belden helps organizations capture and move data securely from machines, sensors, and control systems to compute resources at the edge or in the cloud, where models can be trained and deployed. With industrial networking and connectivity solutions designed for harsh environments, Belden supports resilient architectures that keep critical applications running, even when conditions are demanding..
AI also increases the need for deterministic performance and strong cybersecurity. Belden’s portfolio supports segmentation, secure remote access, and network visibility, helping teams reduce risk while maintaining uptime. At the same time, scalable physical infrastructure – structured cabling, connectors, and network components – helps customers expand capacity as AI use cases grow..
By combining trusted connectivity, industrial-grade reliability, and secure network design, Belden empowers customers to build infrastructures that are ready to operationalize AI, today and at scale.. More from Belden. 20 Dec 2024
