The Vogtle reactors, Units 3 and 4, are the 2 newest units in the US. They were marred by horrible cost overruns (including $3.7B paid to the original construction contractor to **walk away** from the project), totaling a little over $34B for 2,234MW of electricity.
Of course, that's first of a kind rollout. So let's say you could reduce those costs by 50%. You're still looking at an installation cost of $7.6M per megawatt of electricity.
The site permit was first applied for in 2006, and Unit 3 was connected to the grid on April 1, 2023 - seventeen years from permit application to juice-on-the-grid. Even if you reduce THAT by 2/3 (which is laughably optimistic), you're still looking at 6 years out. I think that's way beyond the time horizon that OpenAI can even think about - for 1 GW of electricity, let alone 5.