Does the cost depend on factors like land prices or regulations?

AWS pricing is influenced by economies of scale. Land, power, and other factors contribute to costs. Larger regions generally have lower costs due to economies of scale, and AWS passes these cost reductions to customers as regions grow. When customers want to move to the cloud, they decide where to place their region. Latency is a factor, but price is usually the main consideration since latency differences, like between Ohio and Virginia, are minimal—maybe one or two milliseconds.

Do those two billion include the land? Typically, land is cheap because it's not in the city center.

In the overall scheme of things, it's nothing. That's why in the news you see that AWS, Microsoft, and others are giving up on land or reselling land they committed to earlier. They're trying to make the case that it's because demand is decreasing, but that's not true. It's just that land is so cheap in the overall scheme of things. In many cases, you can land bank or make commitments to land earlier. In the AI space, it's really a race between those companies, including xAI and others. They're making commitments before doing full due diligence of the site. When you buy land, you drill holes for an environmental assessment and do a test to see how big of a data center can fit there, how to get power, and water. All that takes six months before buying the land. Now, because the market is undersupplied, they're making commitments before full diligence and sometimes drop them if it doesn't make financial sense.

Nvidia would be one million versus 700,000 or something like that.

Yes, probably something close to that. It didn't drastically change. What changed is how fast it's consuming the power for that site and how fast that site needs to grow. The approach of cloud providers has changed drastically with AI. Before, they tried to avoid investing in new regions because of the three things mentioned. They tried to grow and get economies of scale in regions they were already invested in. But now, AI requires so much power, and the demand is so big that there isn't enough power in the regions where they have data centers today. They need to diversify and start building in Minnesota, Iowa, Philadelphia—places that wouldn't have made sense before, but they are doing that now because that's where there's power.