Google's power-first data centres - building the grid alongside the server farm

[SuperPosition]

The Wall Street Journal has covered Google's distinctive approach to data centre construction in 2026: instead of connecting to the existing power grid and hoping for capacity, Google is now building power generation directly alongside its data centres. The Meitner Energy Center in the Texas Panhandle is the clearest example - a co-located campus pairing Google's AI infrastructure with over 1 gigawatt of wind, solar, battery storage and on-site gas-fired backup generation.

Google acquired clean energy developer Intersect Power in March 2026 for this purpose, having previously formed a partnership with them and TPG Rise Climate. The strategy directly addresses the central constraint on AI infrastructure expansion in 2026: power availability, not compute availability, is now the binding limit on how fast hyperscalers can grow. Transmission upgrades take years. Grid interconnection queues in some states run to 7+ years.

https://www.wsj.com/tech/ai/googles-unique-approach-to-getting-data-centers-built-2cfae652

https://www.wsj.com/tech/ai/googles-unique-approach-to-getting-data-centers-built-2cfae652

[Cobra]

This is the most important infrastructure story in AI and it gets a fraction of the attention of the model benchmarks. Every major AI company is now limited not by their ability to build data centres or buy GPUs but by their ability to access reliable power at the scale they need. Google internalising the energy development function is the logical response

[BankHolidayBlues87]

The power-first framing is a genuine strategic insight. Building compute and then trying to connect it to the grid creates a dependency on infrastructure you do not control. Building power and compute simultaneously means both are online at the same time and the grid sees a smaller incremental load increase