INTELLIGENT TECHNOLOGY

DATA CENTRES

The finding, published by TRG Datacenters, not only underscores the extraordinary scale of modern AI systems but also signals growing pressure on already stretched global energy networks.

The report compares the world’ s leading AI clusters by converting their hardware into‘ H100 equivalents,’ a standardised metric based on NVIDIA’ s widely used AI processors. This approach enables systems built on different architectures to be measured on equal footing, highlighting the widening gap between the largest deployments and the rest of the market.

At the top sits xAI’ s Colossus facility in Memphis, delivering an estimated 275,000 H100-equivalent units of compute – nearly three times the capacity of its closest competitors. The system consumes roughly 352 megawatts of electricity, comparable to the daily usage of a mid-sized city, raising concerns among energy providers about long-term sustainability. Despite its immense scale, Colossus is also relatively efficient, requiring less power per unit of compute than many rivals. newer chip architectures are beginning to influence performance benchmarks. Meanwhile, Tesla’ s Cortex system highlights a different approach, focusing on internally driven workloads such as autonomous driving models trained on real-world data at scale.

Geographically, the dominance of the United States is unmistakable. Seventeen of the world’ s top twenty AI supercomputers are located there, with only a handful spread across Europe, including installations in Germany, Norway and Switzerland. This concentration reflects not only access to capital and hardware but also the availability of energy infrastructure capable of sustaining such large-scale deployments.

Energy is emerging as the defining constraint for the sector. According to TRG Datacenters, the top twenty clusters alone require more than 1,200 megawatts of power before accounting for cooling overhead, which can add another 30 to 50 percent. In cities such as Memphis and Phoenix, utilities are already under pressure, prompting faster approvals for new power generation projects tied directly to data centre expansion.

The implications are significant. AI is no longer just a software story – it is increasingly an infrastructure and energy story shaped by supply chains, regulation and national priorities. As organisations push to develop larger models, the ability to secure reliable power is becoming a key differentiator.

In that sense, Colossus represents more than a technical milestone. It signals a shift in the competitive landscape, where computational scale, energy availability and geopolitical positioning are becoming inseparable factors in determining artificial intelligence leadership worldwide. •

Trailing in second place are two clusters tied at 100,000 H100 equivalents: one operated by Meta and the other a joint deployment between OpenAI and Microsoft in Arizona. Both systems consume approximately 142.7 megawatts, reinforcing an industry pattern where increases in compute capacity are closely matched by rising energy demand.

Further down the ranking, Oracle’ s H200- based supercluster demonstrates how

28

INTELLIGENT CIO NORTH AMERICA www. intelligentcio. com