NTT's reported ¥1 trillion AI data center: who pays for the power and water?

Good evening. The Owl here.

Osaka may soon get an AI data center whose construction costs could reach ¥1 trillion (roughly $6.3 billion). On October 13, it was reported that NTT Data Group had entered final talks on a facility for the US company Anthropic. Its power-receiving capacity would be up to 500 MW. Nothing has been signed yet, but buildings that run AI are about to reach this scale in Japan, too. [1]

Eleven days earlier, Amazon announced that it would invest an additional $1 billion-plus over the next five years in the US communities that host its data centers. AWS CEO Matt Garman framed AI as a race the United States cannot afford to lose. [2]

Companies want to build fast. The communities that host these facilities, meanwhile, want to know what happens to their electricity bills, their water and their jobs. In North Carolina, with Amazon, Google, Meta and Microsoft taking part, a new arrangement was reached on who pays the cost of connecting data centers to the grid. [3]

Is this a story about some far-off country? According to the International Energy Agency (IEA), annual electricity consumption by Japan’s data centers will rise by about 15 TWh, or roughly 80%, between 2024 and 2030. [4]

The profits AI generates get plenty of attention. But what burdens and what rewards come to the towns where these facilities are built? Let’s look at the construction rush from the ground up.

---

“About 3%” globally, but huge demand locally

In the forecast the IEA updated in 2026, total electricity consumption by data centers worldwide grows from about 485 TWh in 2025 to about 950 TWh in 2030. That is roughly 3% of global electricity demand. Consumption by AI-focused facilities is expected to grow fastest within that. [5]

Seen globally, 3% may not sound like an outrageous share. But data centers are not spread evenly across the map. They need large amounts of electricity, and they need it in specific places.

Getting power to a facility can mean building new substations and transmission equipment, or upgrading existing ones. Even if a company pays for the building itself, local residents could end up affected unless there is a system that makes it pay its fair share of the extra grid costs as well.

Conversely, if large customers fully cover the cost of that equipment, electricity rates will not necessarily go up. The IEA points out that the impact on prices depends on local supply and demand and on how costs are allocated. [5]

So the question to ask is not just “Do data centers use too much electricity?” It is who gets the bill for the equipment they make necessary.

Big Tech buys computing power, and communities ask for terms

In July 2026, Meta announced that it would team up with BlackRock to develop a data center with 1 GW of computing capacity in El Paso, Texas. According to Meta, construction will employ more than 4,000 people at its peak. Once the facility is complete, operating it will take about 300. [6]

Construction jobs are a real benefit. But once the work is done, the same number of people will not keep working there. If a local government is counting on jobs and tax revenue, it needs to separate the effect during construction from what remains afterward.

The questions coming from communities are getting more specific, too. In San Jose, California, residents’ groups are calling for stricter disclosure of data centers’ electricity and water use and their environmental impact. The head of one local group told Reuters that companies have to show why they should use the community’s resources and how they will give back to it. [7]

Some people welcome the tax revenue these facilities bring. If we lump all residents’ concerns together as “opposition to development,” we lose sight of the negotiation over terms that actually matters.

The electricity bill can be changed by contract

The North Carolina case is a step forward in exactly that kind of negotiation.

On October 7, the utility Duke Energy announced an agreement with the state agency that represents consumers, joined by Amazon, Google, Meta, Microsoft and others. Large customers will pay up front, on a non-refundable basis, for equipment such as substations dedicated to data centers, and they will also be asked for deposits or collateral for upgrades to shared transmission equipment. The new arrangement is awaiting approval by the state’s regulators. [3]

If a company simply declares “We won’t raise local electricity bills,” residents have no way to check it. Dedicated equipment, shared equipment, forecasts of future demand: only when the question of who bears each of those costs is written into contracts and rate structures can anyone verify it.

Amazon’s community investment of more than $1 billion is also worth credit. But investing in education and job training is a different matter from paying for the construction of the grid. If both are needed, we should look at the amount and the results of each separately.

“Recirculating water” doesn’t settle the question

Technology is advancing on the water side, too. In October, Meta explained the closed-loop liquid cooling used in its AI facilities, a system that circulates coolant through a sealed circuit. [8]

But circulating water inside a cooling loop is not the same as a facility using zero water overall. Where the water is drawn from and how waste heat is handled differ from site to site. Water is also used in generating the electricity. Lawrence Berkeley National Laboratory in the US notes that indirect water consumption also varies with the mix of power sources that supply a data center. [9]

National average usage figures alone tell you nothing about the drought risk in a particular town. With water, too, we need to look at each facility’s withdrawals together with local conditions.

The question for Japan: where to build

In Japan, the uneven distribution of sites is also an issue. According to the Energy White Paper from the Ministry of Economy, Trade and Industry (METI), as of 2023 about 90% of the country’s data centers, measured by floor area, were concentrated in the Tokyo and Osaka areas. The government is studying how to connect places where power is easy to secure with communications networks, under the banner of “watt-bit collaboration,” linking power (watts) and data (bits). [10]

The NTT Data and Anthropic story is still only a news report. Even so, now that a power-receiving capacity of up to 500 MW has been put on the table, if construction moves ahead the question will be how to secure the power supply and the grid. Note that 500 MW is a power-receiving capacity, not a figure for how much electricity the facility would actually use in a year. [1]

Having AI hubs inside Japan does matter. It would make it easier for Japanese companies and researchers to access computing resources, and related industries could grow.

To share that hope with the community, though, it is not enough to announce the size of the facility. How much electricity and water it will use, who will pay for the additional equipment, and what jobs and tax revenue will remain once it is running all need to be laid out before construction begins.

Closing thoughts

Generative AI may live on our screens, but its foundations sit on the land.

Companies have reasons to want AI up and running quickly. Communities have reasons to want to protect the resources they live on. What connects the two is neither hope alone nor opposition alone, but a concrete agreement on how to share the costs and the benefits.

🦉 The data centers that support the cloud are built not up in the clouds but on the ground of real towns. Just as the Blakiston’s fish owl depends on the great trees it nests in and the rivers full of fish, I think AI’s “nests” can only take root if they protect the everyday lives of the towns they share electricity and water with.

References

[1] Reuters: NTT Data in final talks to build Anthropic’s first data centre in Japan, Nikkei reports

[2] Amazon: The race our nation can’t afford to lose, plus a new commitment from us and a fresh set of community investments

[3] Duke Energy: Duke Energy protects customers from data center costs

[4] IEA: Energy demand from AI

[5] IEA: Key Questions on Energy and AI — Executive summary

[6] Meta: Meta Announces New Strategic Venture With BlackRock to Develop Data Center in El Paso

[7] Reuters: In Silicon Valley, AI data center boom meets local resistance

[8] Meta: Debunking the Biggest Data Center Myths

[9] Lawrence Berkeley National Laboratory: 2024 United States Data Center Energy Usage Report

[10] METI, Agency for Natural Resources and Energy: Energy White Paper 2025, “Energy and industrial policy in light of DX and GX” (in Japanese)