Steven Carlini, Chief Advocate, Data Centers and AI, Energy Management Business Unit, Schneider Electric.
After a loss, many athletes leave the field saying their performance was unacceptable, that they need to “do better.” As a sports fan, I agree with these assessments and believe the same guidance can be applied to the data center industry, which finds itself front and center in national and global debates.
The world is evolving with AI, and the data centers that power it need to adapt to the changing world. In my opinion, the data center playbook from last decade will not be successful going forward. The industry must “do better.”
Here are my suggestions for the new data center industry playbook.
Data Centers Must Be Transparent And Give Back To The Community
Today, utility power is a main headwind for data center projects. While utilities are eager and willing to expand, they have to mitigate risks of overbuilding like they did for the dot-com era. Yes, we have seen this movie before at the turn of the century. Data center construction completely stopped for a few years, waiting for the internet applications and demand to catch up. Utilities were left with overcapacity for years and a bitter taste for data centers.
For data center operators to move faster today, they need better transparency (no more non-disclosure agreements), accuracy (only request power they really intend to use) and direct funding for power projects or preset use contracts where they are committed to pay for power use for a predetermined period of time.
In essence, data center operators need to give to the local communities and not be a burden. Keeping power prices low in the areas they build while keeping away from the local water supply will help accomplish this goal.
Physical Infrastructure That Doesn’t Come At The Expense Of Local Resources
Many of the new gigascale data center projects are “bringing their own power,” whether that is the majority or all of it because of the long wait for utility power. Grid studies or “impact analysis” can take 10 years. When you request to add a large load to the grid, the utilities have to conduct studies to determine the effect if part of the grid goes down. Data center operators can cut the time to power dramatically by bringing their own power, which consists of mainly natural gas turbines, but in the future will be fuel cell and some form of nuclear (traditional or SMR). In addition, many data center operators are also oversizing power systems with the intention to provide the incremental power to the local utilities and residents. The benefits are less grid stress potential and lower utility prices for residents.
Water use is different. For many new projects, AI data centers are ditching evaporative cooling and are minimizing the use of water in the latest designs. If they are using water, they are relying less and less on potable town water and are purifying waste water to use for data center cooling.
Available Excess Power In The Grids Must Be Used
Renewable wind and solar power only operate when the wind is blowing or the sun is bright. Believe it or not, there are times of excess and a lot of what’s called power curtailing is happening, which means low carbon power production is turned off or wasted. I believe the answer is to store that energy in large batteries with capacity for hours, either on the grid (grid scale) or in the data center. Operators can strategically discharge these batteries during times of grid stress or when only carbon intensive power is available.
Next-Gen Data Center Cooling Must Reduce Water Use
It sounds paradoxical, but liquid cooling can use much less water than traditional data center air cooling. Closed-loop liquid cooling systems for data centers only have to be filled once—like the cooling in your car—while air cooling traditionally evaporates water in its cycle.
In legacy data centers, cooling can consume up to 40% of a data center’s operating budget. Moving to liquid cooling—specifically direct-to-chip liquid cooling, cuts cooling energy by 30% to 60% compared to the air-cooled configurations. Liquid cooling is 3,000 times more efficient at removing heat than air.
Why was liquid cooling not implemented before? The reason is simple: It was very complex to retrofit air-cooled servers. But now operators have no choice because the newest accelerated compute AI servers are natively liquid cooled and come with input and output liquid connections. Plus, many newer GPUs can operate at very high temperatures. The liquid circulating through the cold plates can reach 45 degrees Celsius (113 degrees Fahrenheit)—very warm. This means the heat rejection doesn’t have to work hard, resulting in free cooling in most climates with dry cooler heat rejection, which is essentially large passive radiators. Shifting from air cooling to liquid cooling with optimized temperatures 45 C reduces water consumption by at least 50%. Our analysis finds that raising TCS (IT supply temperature) temperature from 32 C to 45 C increases WUE, or water usage effectiveness, by roughly 66% in Dallas and 82% in Paris.
Addressing Resident Concerns To ‘Do Better’
Residents have legitimate concerns about electricity and water use as they may have had negative experiences with legacy data centers. Developers can address those concerns by being transparent about potential impacts, engaging communities earlier and investing in solutions that reflect local priorities.
When a data center “does better,” the community where it’s located will have excess capabilities including power, fiber and road infrastructure that serve the residents and keep water supplies intact.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?