Ambitious plans to scale artificial intelligence are colliding with energy constraints, policy bottlenecks, and geopolitical risk, forcing a difficult question: what kind of AI economy can Taiwan realistically sustain?
Public works programs in Taiwan have historically come in tens. The best-known example remains the Ten Major Construction Projects, launched in 1974 under then-Premier Chiang Ching-kuo.
The program reshaped the island’s economic landscape. It included the construction of Chiang Kai-shek International Airport (now Taoyuan International Airport), National Freeway 1, which runs the length of Taiwan, and the CSBC Corporation shipyard in Kao-hsiung, home to one of the world’s largest drydocks.
Almost 30 years later, another premier, Yu Shyi-kun, announced the New Ten Major Construction Projects. While some of these — including the target of boosting Taiwan’s universities in world rankings and promoting Taiwan as a tourism and culture hub — were broad development goals rather than physical infrastructure projects, they helped stimulate a flagging economy in the early 2000s.
Not to be outdone by his predecessors, Premier Cho Jung-tai put his seal on a new set of initiatives last August. The 10 Major AI Infrastructure Projects were further outlined by President Lai Ching-te during his National Day address in October.
Despite the label, not all of the projects fall neatly within conventional definitions of artificial intelligence. According to Lai’s remarks and media reports, the initiative also encompasses quantum technology, smart robotics, and silicon photonics — fields that intersect with, but are not limited to, AI applications.
Silicon photonics, in particular, points to the program’s broader technological ambitions. By enabling faster, more energy-efficient data transmission, it is seen as a potential solution to bottlenecks in AI data centers, where processing speed and power consumption have become critical constraints.
Among the AI-related infrastructure is the national cloud computer center in Tainan, unveiled by Lai in December. Overseen by Taiwan’s National Center for High-performance Computing (NCHC), the center hinges on the Nano4, a GPU-dense cluster, which ranked 29th on the TOP500 list of the world’s most powerful non-distributed computer systems in November 2025. Powered by NVIDIA’s advanced Blackwell chips, the Nano4 serves as the core AI engine of Crystal26, a data center-sized supercomputer designed to support Taiwan’s expanding AI ambitions.
AI’s power problem
Foremost among the challenges in implementing the projects will be increased energy demand. Taiwan, which imports over 97% of its energy, is watching the war on Iran by the United States and Israel with concern. Disruptions at the Strait of Hormuz, compounded by Tehran’s strikes on Qatar’s Ras Laffan liquefied natural gas (LNG) facility, have underscored the island’s vulnerability.
Roughly half of Taiwan’s energy comes from LNG, with around 30% tied to Ras Laffan, the world’s largest LNG facility. That reliance was further reinforced by a 27-year supply agreement signed last year between state-run CPC Corporation and QatarEnergy. While Taipei has downplayed the risk of an immediate crisis, limited reserves leave little margin for disruption.
As part of a incremental effort to strengthen energy security, Taiwan’s Energy Administration plans to increase LNG stockpiles from 11 days to 14 days by 2027. Following a March 13 Cabinet review of contingency measures, Deputy Director-General Chen Chung-hsien also outlined plans to diversify supply, with U.S. LNG expected to rise from roughly 10% of imports.
“The war in Ukraine already had a huge impact on the process of energy use,” says Eugene Chien, chairman of the Taiwan Institute of Sustainable Energy (TAISE), a non-profit foundation focused on green energies, climate action, and implementation of the United Nations Sustainable Development Goals (SDGs). “Now the situation with Iran is causing further slowdown.”
Having served as Taiwan’s first environment minister from 1987 to 1991, Chien has been at the forefront of the island’s energy transition since its early institutionalization. He sees artificial intelligence as just the latest challenge — a double-edged sword.
“On the one hand, the possibility is for AI to consume huge amounts of energy,” he says. “After all, without electricity, it’s impossible to develop AI.”
However, Chien emphasizes Taiwanese practicality in finding opportunities in game-changing technologies. In conjunction with ICT solutions, he says, AI could greatly reduce carbon emissions by optimizing energy use. Furthermore, AI can create greater efficiency by dramatically streamlining data management processes for environmental, social, and governance (ESG) targets.
“For a lot of ESG work, we have too much data,” he says. Tracking emissions across supply chains, particularly across different scopes, is highly data-intensive work. “Integrating with AI is an important part of [handling] that.”
Navigating this cost-benefit tension will be crucial to the government’s AI strategy, says Tarcy Sih-Ting Jhou, a senior researcher at the Asia-Pacific Energy Research Centre (APERC) in Tokyo. “AI tends to be discussed primarily as a source of new electricity demand, but it also has a role to play on the supply management side,” she says.
As a project director from 2023 to 2025 at Green Energy and Environment Research Laboratories (GEL), a research division under Taiwan’s Industrial Technology Research Institute (ITRI), Jhou focused on low-carbon energy and storage technology in Taiwan.
“Better load forecasting, more efficient dispatch, and stronger demand response programs are areas where AI-driven tools can add real value,” she says. “As Taiwan brings more offshore wind and solar onto the grid, managing the variability that comes with those sources becomes increasingly important, and that is precisely where these capabilities can help.”
She notes that the state-owned Taiwan Power Co. (Taipower) and independent power producers are developing new AI control systems to improve the management of their grids.
Projections from the Energy Administration’s 2024 National Electricity Supply and Demand Report show AI-based electricity increasing eight-fold between 2023 and 2028, though such forecasts are inherently shaky. Citing figures from the 2025 APEC Energy Demand and Supply Outlook, 9th Edition, Jhou says AI is likely to account for approximately 7% of electricity consumption in Taiwan’s building sector.
“While this is a significant share for a single subsector, it remains manageable at a total system level,” says Jhou.
Rather than systemic strain, the pressing issue is the clustering of demand growth in high-technology industrial zones. “This makes local and regional grid reinforcement and coordinated capacity planning the central priority,” she says.
As AI developments have occurred far faster than anticipated, Taiwan’s energy technocrats must factor in exponential growth. “That pace adds uncertainty to any forward-looking estimates,” Jhou says, “and planners are working to account for it.”
Missing the mark
Inevitably, the role of renewables has come under scrutiny. After missing its initial target of a 20% share of the energy mix for renewables by 2025, the government now risks falling short of a revised deadline of November 2026, as Minister of Economic Affairs Kung Ming-hsin acknowledged in February.
Depending on the dataset, low-carbon sources currently account for between 12% and 16% of Taiwan’s electricity generation. Further out, the 2030 target of renewables supplying 30% of the energy mix appears increasingly precarious.
Progress has been slowed by a range of obstacles, according to experts at the Taiwan Renewable Energy Alliance (TRENA), a nongovernmental organization that advocates for the green energy transition, policy reform, and community sustainability. Regulatory barriers and social opposition are among the key challenges, says Raoul Kubitschek, the group’s chairman and a 20-year veteran of Taiwan’s renewables sector.
He points to amendments to environmental impact assessment procedures introduced by the opposition Kuomintang and Taiwan People’s Party and passed by Taiwan’s legislature in November. The changes, which target solar installations, have made it more difficult to get new projects off the ground.
“Bringing in these regulations will slow things down, especially with ground-mounted projects,” he says. With most suitable rooftop space “more or less taken,” Kubitschek points to a potential bright spot: new rules requiring photovoltaic installations on new or renovated buildings with more than 1,000 m2 of ground-level floor space. Under the policy, solar systems must provide one kilowatt of capacity for every 20 m2.
Even so, he says, Taiwan is likely to fall short of its target of 10.9 gigawatts (GW) of solar capacity by 2030.
Prospects are somewhat stronger for offshore wind. The original target of 5.7 GW by 2025 now appears more achievable by 2027. Still, as Taiwan enters round 3.3 of its offshore wind auctions, questions remain about the viability of some existing projects. “They’re also not really coming online,” says Kubitschek. “There’s only about one gigawatt out of five that has the path to survival right now.”
Projects that fell by the wayside in earlier rounds due to failure to meet local consent standards or secure Corporate Power Purchase Agreements (CPPA) on time may now have to be revisited.
Striking a balance between community and engagement will be crucial, complicated by the need to press ahead with projects for fear of falling further behind.
“You used to have the luxury of compensating local communities, while at the same time procuring a lot of components locally,” says Y.D. Chang, vice chairwoman at TRENA. “But now the only goal is to get these projects commissioned on time.”
A notable recent breakthrough was the first ever approval of what is known as “Free, Prior and Informed Consent” under Taiwan’s Indigenous Peoples Basic Law for a geothermal development project in November. The developer, Baseload Power Taiwan, attributed the success to “public briefings, neighborhood consultations, household visits, and open dialogue” with the Indigenous community at Hongye Village in mountainous Hualien County.
Kubitschek notes that although geothermal has seen positive movements, it is unlikely to meet its targets (reported at 200 GW by 2030) “because there are still regulatory issues in these Indigenous communities.”
A particular conundrum for the AI industry is the disconnect between the semiconductor industry and renewable developers.
“The confusing thing is that the industry says, ‘We want renewable energy by 2030’ and the developer goes, ‘That’s great. We want to sell it to you,’ but they’re not reaching agreement on deals,” Kubitschek says. Besides differences over price, other hurdles are “policy, stakeholder management, finding the sites, and grid connection,” he says.
David Chiang, program lead for the Energy Collaborative in Taiwan under the sustainability framework of global microelectronics industry association SEMI, echoes this point, emphasizing that renewable energy development must be supported by clear incentives on both sides of the market.
“We have one [party] that needs to buy and another that will produce and develop if there is a buyer,” says Chiang. “So these two sides are obviously intertwined.”
Out of sync
While obstacles remain, the conditions are ripe for a “three-in-one convergence” among the semiconductor industry, ICT and high-tech — including AI applications — and renewable energy. “There’s no way of separating these elements anymore,” says Chiang.
However, turning Taiwan into a data hub without sufficient energy supply is increasingly viewed as unrealistic. “To start with, you cannot build in north Taiwan anymore, as there’s just not enough grid,” says Kubitschek. He cites Taipower’s announcement in 2024 that it will no longer approve data center projects larger than 5 MW in areas north of Taoyuan.
For TAISE Chairman Chien, it is a question of playing to Taiwan’s strengths. He references the “five-layer cake” ecosystem of AI proposed by NVIDIA founder and CEO Jensen Huang — energy, chips, infrastructure, models, and applications — pointing to the final layer as most important for Taiwan.
“We cannot compete with the United States and China for data centers or robotics,” Chien says. “So we have to pick our advantages from among those layers.”
Nuclear power, long a political third rail, has returned to the policy debate as concerns over electricity supply intensify. Speaking at a business forum in March, President Lai said that Taiwan’s second and third nuclear power plants — Kuosheng and Maanshan — “meet the conditions” for potential restart, with Taipower expected to submit formal plans by the end of the month.
Lai has stopped short of a full reversal of Taiwan’s nuclear phaseout policy. In subsequent remarks, he emphasized that the island could maintain a stable power supply without nuclear energy through at least the early 2030s, even as authorities evaluate restart options and explore next-generation technologies such as small modular reactors.
Chien says stark realities have begun to outweigh safety concerns among the public and even among previously skeptical legislators. “If you asked people on the street, 70% would now agree” with restarting nuclear power, he says. “Without energy, there’s no economic development, so the mentality has changed.”
Just as the original 10 projects were shaped in part by the 1973 oil crisis — and later reinforced by the upheaval in energy markets following the Iranian Revolution — ongoing instability in the Middle East may yet influence the trajectory of Taiwan’s latest AI-driven initiatives.
With Beijing increasingly linking Taiwan’s energy vulnerability to the question of “reunification,” strengthening energy security is likely to become a central policy priority, says Chien.
“Whether renewable or not, indigenous energy is most important,” he says. “And you can’t just concentrate on the one you like, because you need all options to survive.”
