Powering the next generation of AI datacentres
Gas and nuclear not needed
Exhibit A: https://epoch.ai/blog/trends-in-ai-supercomputers
Power requirements and hardware costs doubled every year. If current trends continue, the largest AI supercomputer in 2030 would cost hundreds of billions of dollars and require 9 GW of power… However, 9 GW of power is the equivalent of 9 nuclear reactors, a scale beyond any existing industrial facilities.
there simply isn’t enough grid capacity to unlock many of the best projects in the region from Balranald to Hay and Buronga … most of the more than 20 GW of projects may never see the light of day, because the south-west zone – as currently created – doesn’t have the capacity to support them.
“Grid capacity” in that sentence means that nobody in that region needs that much power, and we haven’t built the long-range power lines to transfer that power to some other region.
Data communications in the area are great. Both Telstra InfraCo and Vocus have terabit-scale cables running roughly parallel with the Sturt Highway between Sydney and Adelaide, passing within 20–30 km of Hay and Balranald. Leasing a pair of dark fibres (or a handful of 400 Gb/s wavelengths) from those carriers gives you instant multi-terabit head-room suitable for running a big new AI data centre.
So… the AI majors (OpenAI, Anthropic, Google, etc.) need to build some massive new data centres, and are looking for locations. The big challenge is finding a site with good data connectivity and good connectivity to facilities that can generate enough power for them.
Meanwhile, power generation companies in Australia have surplus capacity and wish that they had a customer who was a massive consumer of power.
Am I missing something, or is the answer really this obvious?