Why AI workloads may not transform the data center industry

At first glance, the AI ​​boom might seem like a huge boon to the data center industry. The more companies invest in AI, the more data center capacity they need, right?

AI will certainly drive demand for data centers, but I believe that the impact of AI on the industry will ultimately be Proof is limited. Here’s why.

How AI Impacts Data Centers

To increase the demand for AI in data centers, the reason is very simple: Building and deploying AI workloads requires a lot of IT infrastructure - in many cases, including dedicated infrastructure, such as servers equipped with GPUs. The core of data centers is obviously where the infrastructure is hosted, as they not only provide space to host servers, but also provide the physical security controls, efficient energy systems, cooling solutions and other resources that enterprises need to protect their investments in AI infrastructure.

So, over time, as more businesses seek to build or deploy AI models, they will turn to data centers to host the servers needed to achieve their goals—at least that’s what popular intelligence tends to think of.

Will AI really change the data center?

Over the next few years, the number of servers in data centers will gradually increase, dedicated to AI workloads. In some cases, companies are even building new data centers specifically for AI.

However, in general, it is not a given that AI will completely disrupt the industry, or that AI workloads will surpass other types of applications (such as web hosting) and become a critical path in the data center.

Here are four reasons why the AI ​​boom may not be as big an impact on data centers as it seems.

Temporary need for AI infrastructure

First, allowing AI use cases does not require enterprises to permanently own AI infrastructure. If you need to train a model, you will need a lot of computing power during training, but then you won't be able to use that server capacity until you train the next model.

Therefore, for most enterprises interested in AI, it is more financially feasible to use an IaaS solution for their AI infrastructure needs rather than purchasing their own servers and deploying them in a data center. Significant. Unlike other types of workloads, AI requires intermittent, large-scale infrastructure.

Idle AI infrastructure is already plentiful

Buying AI infrastructure and data center space to host is even harder to justify given the vast amounts of cheap infrastructure capacity already available from IaaS providers. reasonable.

For example, Spot VM instances are available at significant discounts compared to standard public cloud servers and are a great way to perform AI training. The main disadvantage of Spot instances - cloud providers can shut down instances without any warning, potentially disrupting any workloads hosted on them - isn't too much of an issue for AI training, since in many cases the training can Pause and resume on different instances.

In short, enterprises are unlikely to expand their data center footprint to support AI workloads when they can use ultra-cheap existing IaaS offerings for the same purpose.

Few businesses build their own AI models

No matter what infrastructure you use, developing, training, and deploying an AI model from scratch is hard work—so hard , so that few companies are likely to do so. Most people will probably choose third-party AI services from companies such as Microsoft or Google.

These services are provided by vendors who build and train their own models, so customers using these models do not need to purchase their own AI infrastructure.

The AI ​​craze will eventually subside

Currently, GenAI is a hot topic, and there is increasing pressure on enterprises to invest in AI solutions, but in five or ten years, most enterprises’ AI The strategy may have matured and they will move to new technology trends.

What this means for data centers is that any rise in demand caused by AI is likely to be mostly temporary - it would be unwise to significantly expand data center capacity, only to find that it is no longer needed in the medium term.

Conclusion

Bottom line: Except for those that specialize in developing AI software, few companies have a good reason to invest in data centers to support AI workloads. Expect the hype around AI to drive some growth in data center capacity, especially in the next few years, but don't expect AI to cause a significant rise in demand for data center space - as existing space will likely be enough to meet the needs of most enterprises.