Taipower planning for AI power demand

DedicateD LINES: Taipower is planning on running power lines directly from power plants to high-draw data center sites, skipping the grid entirely for resilience reasons

• By Meryl Kao / Staff Reporter

• • Add TT as Preferred Source

  • 

  • 

  • 

Taiwan Power Co (Taipower, 台電) plans to supply electricity to artificial intelligence (AI) data centers through direct lines, as rising AI demand puts pressure on the grid, Taipower chairman Tseng Wen-sheng (曾文生) said on Sunday.

The utility plans to allow AI data center operators to manage those dedicated power lines themselves, with their length, costs and timing determined by market mechanisms, Tseng said during an interview.

“The most stable way to supply electricity is to locate as close to the power source as possible,” Tseng said.

Taipower chairman Tseng Wen-sheng speaks at a company event on April 30 in Taipei.

Photo: Peter Lo, Taipei Times

Electricity is a “sovereign industry,” and Taiwan needs to reinforce the resilience of its power supply amid geopolitical risks, he said.

Taipower has discussed future power demand with the Taiwan Semiconductor Industry Association to get a clear picture of the industry’s electricity consumption, Tseng said, to ensure there would be sufficient power to meet the industry’s demand.

Following a major power outage on March 3, 2022, Taipower reviewed its grid and introduced measures to increase resilience, including plans to connect power plants with science parks, Tseng said.

Rather than routing electricity through the main north-south transmission grid, Taipower is developing dedicated lines to ease congestion and diversify risks, he said.

Examples include links from the Datan Power Plant (大潭電廠) in Taoyuan’s Guanyin District (觀音) to the Taoyuan Aerotropolis, and from the Tunghsiao Power Plant (通霄電廠) in Miaoli County to the Hsinchu Science Park (新竹科學園區), he said.

Grid resilience has become a key part of supplying the semiconductor industry amid the AI boom, he said.

Tseng said Taipower has shifted from focusing primarily on overall electricity supply to matching major power users with available generation and grid capacity, tracking demand at individual factories and science parks to plan supply more precisely.

AI data centers across the world face challenges such as location controls, pricing mechanisms, capacity thresholds and requirements for self-generated power, he said.

Growing power demand requires upgrades to the grid, including higher voltage levels, greater cable capacity and underground infrastructure overhaul, he added.