- Gartner research suggests that AI data center power requirements will grow by 26% by 2026
- This is a 13% increase over a previous forecast, which limited growth to 500 TWh.
- AI data centers currently account for 31% of total data center power consumption, but are expected to exceed conventional server power demand by 2027
In the last few years, the demand for AI chip has skyrocketed, with all major players in the industry investing in infrastructure, training and inference hardware to build their own data centers and clouds for computing.
The assumption was that better, faster chips were the key to unlocking both Artificial General Intelligence (AGI) and AI-infused efficiencies as the world shifts its focus from AI agents to AI operators.
The bottleneck that many saw coming, but was undoubtedly downplayed, is now back in focus: power constraints could limit future growth in data centers globally.
Not a chip problem, but an energy conundrum in 2030?
A recent report from Gartner indicates that AI servers may not have a chip supply problem, but power limitations that could be critical to future expansion of the data center, bringing it to a standstill by 2030 if not addressed.
Gartner estimates that while current data center power needs are limited to 132 GW, they could reach 290 GW by 2030, indicating that energy constraints will undoubtedly determine the root of future AI data center planning.
“Increasing demand for compute-intensive AI workloads is driving unprecedented growth in data center power, while AI capabilities are now constrained by power availability, making data center power security the new battleground for scaling and protecting margins in the global AI race,” said Linglan Wang, director analyst at Gartner.
The current estimate makes even the most extreme case painted by electrical infrastructure provider Schneider Electric look tame.
This is why Nvidia CEO Jensen Huang has already started touting power efficiency as the reason its chips are superior to the competition.
In a recent interview with BloombergHuang said that both data centers and enterprise consumers will want the highest number of “tokens per watt” to achieve maximum value in a power-constrained future.
Scaling power generation or upgrading the grid may arguably be a more complex or time-consuming endeavor than simply building the AI data center, with Goldman Sachs estimating that as much as $720 billion in grid spending may be needed by the end of the decade to account for the additional load that AI data centers will bring to the table.
Whether this plays out exactly as projected by Gartner remains to be seen; However, with every industry player indicating their intention to increase spending on AI infrastructure, the projection that sees current power demand (565TWh) more than double (1200TWh) by 2030 is a very possible scenario, and industry focus may shift to delivering both power and efficiency versus raw data processing over time to account for the change.
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds.
