Site icon Connected World

AI at the Data Center

Data centers are poised for big disruptions, as new innovations become available. We are beginning to see new power and cooling capabilities, as well as the rise of AI (artificial intelligence) that is set to transform everything.

Recent research from Dell’Oro Group suggests AI infrastructure spending will propel data center capex to more than half a trillion dollars by 2027, with worldwide data center capex forecast to grow 15% by 2027.

Baron Fung, senior research director for Data Center Capex, Dell’Oro Group, suggests the biggest takeaway from this research is the increased projections for AI-related spending and that AI-related infrastructure will account for a large portion of data center spending in five years.

“When we talk about AI-related infrastructure spending, servers have the largest contribution of overall data center spending,” Fung says. “These are servers that are related with accelerators such as GPUs, FPGAs, and other custom chips that are optimized for AI-related workloads. By optimized, I mean these accelerated servers are more efficient at processing AI workloads than traditional general-purpose solutions could. These accelerated servers are just part of the AI solution.”

Fung goes on to explain that specially design networks and physical infrastructure (such as power and cooling) are also part of the overall AI solution. Within those solutions, we also have components such as accelerators, memory, storage, and optics.

“We anticipate that the hyperscalers—both U.S. and China—will lead the way to AI-related investments, with other opportunities in enterprise and government sectors,” says Fung.

Further, the organization estimates these accelerated servers could account for nearly 30% of total data center capex by 2027. Fung explains in addition to the higher adoption rate of accelerated platforms by 2027, these accelerated systems also have premium cost over traditional general-purpose infrastructure. “The 30% number is just our estimate for servers alone and does not include complementary infrastructure such as networking and physical infrastructure.”

Maikel Bouricius, CCO, Asperitas, says with AI, the applications are now arriving, and the demand increased and materialized almost overnight, it seems, and even now a lot of industry players are still preparing their investment plans for this trend.

Short-Term Slow Down

With all this growth also comes some slowdown predictions. Dell’Oro Group suggests near-term cloud and enterprise capex growth to decelerate as the market undergoes all of this digestion.

“However, we expect this slowdown or digestion is not long-term, and another expansion cycle is expected for some cloud service providers in 2024,” says Fung. “These hyperscalers have undergone a cadence of expansion and digestion cycle as they optimize their capacity and deploy new technologies. However, recent events such as the pandemic and supply chain issues have disrupted the timing of these cycles. We also project enterprise spending to slow in 2023. There was a bit of pent-up demand that was satisfied in 2022.”

Another prediction here is that the edge computing forecast was trimmed as the ecosystem and compelling use cases have been slow to materialize.

“Edge deployments tend to be complex and require alignment and standardization of the ecosystem, involving system vendors, telcos operators, cloud service providers, system integrators. So far, we see that this ecosystem is still quite fragmented. There are also regional differences. We also see some of the hyperscale cloud service providers deploying their own edge solutions,” says Fung. “Customers still need time to understand the ROI of these use cases and develop applications. Furthermore, edge network deployments as reported by our telecom analysts on our team, and from MEC system shipments of OEMs have been limited to date. Thus, we had to reduce our edge computing forecast.”

Bouricius of Asperitas chimes in and says the most surprising thing to see here is edge computing has slowed down.

“I think the big question for everyone will be: where are my workloads going to run and a lot of it will need to be outside the large data centers hubs as we know them, closer to the users, so my expectation is that we will see edge sites materialize in the next two years,” says Bouricius.

Bouricius suggests from an immersion cooling provider perspective, the company has watched as the edge space has been a trend for years, which didn’t materialize or was visible in project pipelines.

“Nowadays, we don’t speak of edge anymore as much, but we talk about decentralized data centers,” says Bouricius. “The latter can be an add-on for micro sites for high performance computing next to centralized data centers, it can be connectivity-based assets on a network, or closer to end-users of the applications. There can also be even more innovative approaches closer to the users of heat to utilize that value stream coming out of data centers or compute hubs. We see this trend going up across the multiple use cases and also many new entrants in this space. We’ll need to see how that’ll develop further. Immersion cooling is part of the ‘enabling’ solution for this kind of use case, as it allows for standardized efficiency across the board, in any location and makes it easy to manage from an operator point of view.”

Looking to the Future

With all this in mind, what next steps should companies take as we move forward in a new era of innovation? Fung says we are entering a new era in which efficient computing will be a key focus, which means deploying more purpose-built solutions with accelerators that are optimized for certain workloads, particularly AI.

“These are the workloads that could deliver the most value,” says Fung. “However, budget is always limited for these companies. So, they need to figure out where they can trim cost. They can optimize their general-purpose infrastructure through the use of newer servers, denser storage, and faster networks to minimize their footprint and power consumption and increase the number of virtual machines per physical machine. Or they can trim costs by exploring other models, such as deploying certain workloads to the public cloud and utilize a multi-cloud strategy. However, one key challenge in making these technology shifts is for companies be able to accurate assess the ROI.”

Bouricius adds companies should start bringing innovation on board throughout the full data center stack, including power and cooling. “It only makes sense when you’re investing heavily on compute power to consider the best options to utilize and protect that investment. Then immersion cooling comes into the picture as one technology to support that. The best way to start is to talk to specialized leaders in that field, start sharing your requirements and ambitions, pilot, and optimize from there.” It’s clear the future is changing fast, and organizations need to be thinking about where they are headed both internally and externally to keep up with the pace of change.

Want to tweet about this article? Use hashtags #IoT #sustainability #AI #5G #cloud #edge #futureofwork #digitaltransformation #green #ecosystem #environmental #circularworld #datacenter

Exit mobile version