Fitch says power, not chips, will decide Taiwan’s AI data centre growth

Fitch says power, not chips, will decide Taiwan’s AI data centre growth

The island builds most of the world's AI servers, but its grid is becoming the real bottleneck for hosting them at home

Taiwan has spent years becoming the factory floor of the AI boom. Now it is finding out that building the machines and plugging them in are two very different problems.

Fitch Ratings says power availability will drive the growth of Taiwan’s AI data centres.

A booming economy with a power problem

Fitch affirmed Taiwan’s sovereign credit rating at ‘AA’ with a Stable outlook in July 2026.

The agency pointed to strong economic growth powered by AI-related exports. GDP growth is projected to reach 9.4% for 2026.

Taiwanese manufacturers account for nearly 90% of the global supply of AI servers.

Fitch cautioned about energy security risks tied to Taiwan’s high dependence on imported energy sources.

Advertisement

The grid is the gatekeeper

As of November 2025, there were 79 applications for AI data centre power, totaling approximately 4,758 MW.

Only 40 of those applications, covering about 3,033 MW, had been approved.

Since 2024, northern Taiwan has had a freeze on new power applications above 5 MW because of grid constraints, particularly around Taoyuan.

High-density AI loads have exceeded traditional industrial demand, which complicates where data centres can realistically go.

Operational AI data centres include Google’s Changhua site at around 300 MW, with total tracked capacity of approximately 660 MW across five facilities.

Taipower’s very large to-do list

Taipower holds a 67% share of power generation and a 98% share of distribution. Fitch rates the company ‘AA’/AAA(twn) with a Stable outlook.

Taipower’s projected capital expenditure runs between TWD 261-442 billion annually from 2026 to 2029, aimed at expanding gas-fired capacity and integrating more renewable energy.

Comments from Taipower and Ministry of Economic Affairs officials in July 2026 suggest restrictions in the north could potentially ease, contingent on new power sources coming online.

What this means

Projects that already hold approved power allocations look meaningfully better positioned than those still sitting in the queue.

With the northern freeze in place since 2024, sites outside the Taoyuan area, like Google’s Changhua facility, show where capacity has actually been achievable.

If new gas-fired and renewable capacity arrives on schedule, it could unlock the northern restrictions and clear part of the application backlog.

The energy security warning adds a longer-term layer of risk. A grid built heavily on imported fuel is exposed to supply disruptions that no amount of local engineering can fully offset.

Taiwan supplies the hardware that powers AI data centres around the world, yet its own ability to host them is capped by the wiring at home.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.
Fitch says power, not chips, will decide Taiwan’s AI data centre growth
Fitch says power, not chips, will decide Taiwan’s AI data centre growth

The island builds most of the world's AI servers, but its grid is becoming the real bottleneck for hosting them at home

Taiwan has spent years becoming the factory floor of the AI boom. Now it is finding out that building the machines and plugging them in are two very different problems.

Fitch Ratings says power availability will drive the growth of Taiwan’s AI data centres.

A booming economy with a power problem

Fitch affirmed Taiwan’s sovereign credit rating at ‘AA’ with a Stable outlook in July 2026.

The agency pointed to strong economic growth powered by AI-related exports. GDP growth is projected to reach 9.4% for 2026.

Taiwanese manufacturers account for nearly 90% of the global supply of AI servers.

Fitch cautioned about energy security risks tied to Taiwan’s high dependence on imported energy sources.

Advertisement

The grid is the gatekeeper

As of November 2025, there were 79 applications for AI data centre power, totaling approximately 4,758 MW.

Only 40 of those applications, covering about 3,033 MW, had been approved.

Since 2024, northern Taiwan has had a freeze on new power applications above 5 MW because of grid constraints, particularly around Taoyuan.

High-density AI loads have exceeded traditional industrial demand, which complicates where data centres can realistically go.

Operational AI data centres include Google’s Changhua site at around 300 MW, with total tracked capacity of approximately 660 MW across five facilities.

Taipower’s very large to-do list

Taipower holds a 67% share of power generation and a 98% share of distribution. Fitch rates the company ‘AA’/AAA(twn) with a Stable outlook.

Taipower’s projected capital expenditure runs between TWD 261-442 billion annually from 2026 to 2029, aimed at expanding gas-fired capacity and integrating more renewable energy.

Comments from Taipower and Ministry of Economic Affairs officials in July 2026 suggest restrictions in the north could potentially ease, contingent on new power sources coming online.

What this means

Projects that already hold approved power allocations look meaningfully better positioned than those still sitting in the queue.

With the northern freeze in place since 2024, sites outside the Taoyuan area, like Google’s Changhua facility, show where capacity has actually been achievable.

If new gas-fired and renewable capacity arrives on schedule, it could unlock the northern restrictions and clear part of the application backlog.

The energy security warning adds a longer-term layer of risk. A grid built heavily on imported fuel is exposed to supply disruptions that no amount of local engineering can fully offset.

Taiwan supplies the hardware that powers AI data centres around the world, yet its own ability to host them is capped by the wiring at home.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.