We’re witnessing the global adoption of artificial intelligence (AI) right before our eyes, but it’s not without its challenges. While AI can potentially improve how we do things, it requires the technological capacity that enterprises cannot yet provide. The current infrastructure is far from where it needs to be to leverage the true power of AI.
(Data source: Omdia)
The Global Tech Trends Survey by Equinix, a global colocation data center leader, revealed some grounding findings on enterprise infrastructure and AI adoption. The gist of the survey that questioned IT decision-makers was that while more and more companies are adopting AI, they lack the infrastructure readiness to handle the power and data-hungry workloads.
This article discusses some of the key findings from the survey and potential solutions to address the infrastructure issues highlighted.
The 2023 Global Tech Trends Survey by Equinix, now in its fifth iteration, asks IT executives and leaders worldwide their opinions on new trends and strategies. This survey questioned 2,900 participants from all three major business regions (the Americas, Asia-Pacific, and EMEA).
The general takeaway from this survey was that IT teams are under immense pressure to adopt new technologies while maintaining tight budgets. Even though innovation in enterprise solutions, cybersecurity, and AI is going strong, IT leaders face less-than-ideal socioeconomic situations. Therefore, any overhauls of technology must be made with the financial implications in mind.
As AI has emerged as the top trend in technology in 2023, the survey also focused on the advantages and challenges of its adoption.
Here are some of the key findings of the Global Tech Trends Survey 2023:
4 out of 10 IT leaders believe their infrastructure can’t handle AI needs. The same survey shows that 85% of leaders already used AI. Clearly, leaders and their teams want AI for their business applications, but they’re not as confident in their existing infrastructure.
AI workloads, especially training workloads, have a massive power draw. On average, a traditional data center rack would have a power draw of up to 10 kW per rack. However, if it supports AI operations, the same rack may have a draw of over 30 kW.
Besides power issues, AI workloads also require rethinking cooling. More power usage for more processing results in heat. Existing cooling infrastructure can’t keep up with the heat generated by servers with multiple CPUs and GPUs.
For racks drawing power over 30 kW, liquid cooling is necessary, another expense companies may need to account for in addition to the large power bills. More importantly, many data centers are not designed for liquid cooling.
Besides the lack of infrastructure readiness, the survey revealed that tech leaders consider the lack of talent a serious roadblock for AI adoption. The tech talent gap seems to be widening over the years, and the emergence of new technologies is only exacerbating the situation. Companies need the right talent to leverage advanced technologies, but as some tools are pretty new, it’s hard to find people completely adept with them.
AI-as-a-Service can be a viable solution for enterprises that lack the infrastructure to support their ambitious objectives. It eliminates the need to invest heavily in infrastructure while still being able to experiment with AI using public cloud platforms.
While deploying such solutions on out-of-the-box platforms is easy, their cost-effectiveness is limited. Therefore, if saving money is the only reason you’re considering large cloud platforms for setting up AI operations, you might want to analyze costs beforehand. The cost can quickly increase as you deal with large amounts of data read/writes and extractions.
Nevertheless, AI-as-a-Service is a good option to test the waters and train AI models. If those models generate new channels of revenue or improve your company’s productivity, you may consider investing in your own AI-ready infrastructure.
Moreover, different cloud providers may be better suited for specific types of model training. Learning which provider best suits your business needs can ensure your efforts pay off.
Enterprises with on-premise data centers that have the financial capacity to improve their hardware and adopt AI will need to update key parts of their infrastructure.
HPE, Dell, and several other manufacturers have released AI-ready servers powered by multiple GPUs that efficiently handle training and inference. Investing in these servers can enable enterprises to run AI workloads without compromising performance.
Besides refreshing some or all the servers, enterprise data centers will also need to rethink power and cooling. To keep energy consumption in check and meet any compliance requirements, adopt strategies that prevent the wastage of energy.
One viable solution is to segment workloads so that only the most efficient servers and network equipment are dedicated to AI training. The rest of the assets can be used normally.
Any future refreshes should be considered in the light of AI handling.
If you’re looking to refresh your data center and embrace AI, you have to start with the right equipment. It’s an expensive move, but given the proven efficiency AI can bring to operations, it may just be worth it.
PivIT can help you procure the latest AI servers from major brands to kickstart your AI strategy. Whether you want to improve data security or make the management of resources intelligent, AI can enable a wide range of applications that can transform data center operations.
Keep in mind that advancements in AI can only be made with strong hardware in the background that can support the complex computations and movement of massive amounts of data. PivIT can be your partner as you make the leap and embrace the future!