For most of the history of real estate, location meant land, people and roads. Was there a site? Could you hire there? Could trucks get in and out?
For anything built for AI, there’s a new first question: can you get the power?
A rule nobody announced
In August 2024, the chairman of Taipower, Taiwan’s state-owned utility, said publicly that it would no longer approve power supply for new data centres of 5 megawatts or more in Taoyuan and the area north of it. The restriction had quietly been in place since late 2023. The next day the economics ministry backed it, explaining that Taiwan as a whole was not short of power but the north uses more electricity than it generates and depends on supply sent up from central and southern Taiwan.
Five megawatts is not a hyperscale campus. The Taipei Times put it in local terms: roughly what the Shin Kong Life Tower next to Taipei Main Station draws.
The map is being redrawn by substations
In July 2026 the picture shifted. The economics minister said relaxing the cap should not be a problem, and Taipower said it could lift it on conditions. Local governments would propose data-centre zones next to power plants, such as Datan in Taoyuan or Hsieh-ho in Keelung, and Taipower would first check whether nearby substations and feeder lines could take the load. Taipei and New Taipei, with no new power sources, did not qualify.
Read that again as a real estate person. The site is chosen by the power plant, and the zoning follows the substation.
The numbers make the point on their own. That same month, Taoyuan’s city government counted more than ten data-centre investment cases needing about 624 megawatts on roughly 50 hectares of land. The land was the easy part.
It isn’t only Taiwan
The International Energy Agency estimates that data centres used about 415 terawatt-hours of electricity in 2024, around 1.5% of the world’s demand, and expects that to more than double to about 945 TWh by 2030. That’s slightly more than all of Japan uses today. The agency also warns that around one in five planned data-centre projects could be delayed by grid constraints, and that waiting times for transformers and cables have doubled in three years.
Globally, there’s no shortage of electricity. Locally, there’s a shortage of connections. That’s a real estate problem.
The cheapest megawatt is the one you don’t use
If grid capacity is the scarce resource, then efficiency in existing buildings stops being a line in the sustainability report. It becomes capacity.
Heating, cooling and ventilation use roughly half the energy in US commercial buildings, according to the most recent federal survey, and in a climate like Taiwan’s, cooling dominates. Getting those systems to waste less doesn’t require tearing them out. Lawrence Berkeley National Laboratory followed about 6,500 buildings that adopted analytics software and found median energy savings of 9% from fault detection alone, with a typical payback of about two years. That’s software laid over equipment that was already there.
I think of this as turning power into capacity. Every megawatt an existing building stops wasting is a megawatt the grid can give to something new.
What I’m watching
Site memos that start with the grid. The first page will be about the substation and the connection date, not the land price.
Night as the new peak. AI loads run around the clock, and solar doesn’t. Taiwan’s own forecasts show night-time reserve margins below the 15% benchmark through 2029.
Buildings valued for their data. A building that can prove and control its energy use will be worth more than one that can’t, because it’s the one you can trust with scarce power.
I’ve put these on the record as dated calls on the Think page, so I can be checked against them.
Sources
- Taipower and economics ministry statements, August 2024, reported by Business Today, TechNews and Bloomberg via Data Center Knowledge
- July 2026 changes: Taipei Times and UDN
- Night-time reserve margins: Taipei Times editorial, August 2026
- IEA, Energy and AI (2025)
- US EIA, Commercial Buildings Energy Consumption Survey 2018
- Lawrence Berkeley National Laboratory, Proving the Business Case for Building Analytics (2020)
These are my personal views, based on public information.
Written in Taipei. Last revised 9 October 2026 (revision 1).