In the Taiwan Strait, roughly 35 miles off the coast of Changhua County, 111 wind turbines now spin above some of the strongest sustained winds in the western Pacific. Every megawatt-hour they generate flows to a single customer: Taiwan Semiconductor Manufacturing Co., the company that fabricates the advanced chips powering nearly every major AI system on the planet.
The arrangement dates to July 2020, when TSMC and Danish energy developer Ørsted signed what was then the largest corporate renewable-energy power purchase agreement ever recorded. Under the deal, TSMC committed to buying the entire output of Ørsted’s Greater Changhua 2b and 4 offshore wind farm, a project rated at 920 megawatts, according to Ørsted’s official announcement. Not a partial slice. The whole farm.
Five years later, with the wind farm now fully operational and AI chip orders straining Taiwan’s electricity infrastructure to its limits, that bet looks like one of the shrewdest energy moves in semiconductor history.
Why a chipmaker needs its own wind farm
TSMC’s most advanced fabrication plants run 24 hours a day, seven days a week. A single extreme ultraviolet lithography tool, the kind used to print transistors at the 3-nanometer node, draws roughly one megawatt of power on its own. Multiply that across hundreds of tools in multiple fabs, and the numbers become staggering. TSMC consumed approximately 23.6 terawatt-hours of electricity in 2023, according to the company’s ESG disclosures, making it by far the single largest industrial power consumer in Taiwan.
That consumption is growing fast. Orders for AI accelerators from customers including Nvidia, Apple, and AMD have pushed TSMC’s advanced-node fabs toward full utilization. Each new generation of chip technology demands more fabrication steps, more tools, and more electricity per wafer. By some government estimates, the semiconductor sector now accounts for roughly 8 to 9 percent of Taiwan’s total electricity demand, a share that has climbed steadily and shows no sign of leveling off.
Taiwan’s grid, meanwhile, is under visible pressure. Taipower, the state-owned utility, has issued tight supply warnings during summer peaks in recent years, and reserve margins have dipped below comfort levels on multiple occasions. The island still relies heavily on natural gas and coal for baseload generation, and new capacity additions have struggled to keep pace with industrial growth. For a company whose customers depend on uninterrupted chip supply, that gap between demand and reliable generation is not an abstract risk. It is an operational threat.
The structure of the Ørsted deal
The PPA with Ørsted was designed to address that threat directly. Rather than buying renewable energy certificates or contracting for a fraction of a project’s output, TSMC underwrote an entire offshore wind development with a single offtake commitment. The deal documentation confirms the full-output structure, which gives TSMC long-term price visibility and supply certainty that spot-market grid purchases cannot match.
Ørsted’s Greater Changhua complex sits in one of the best offshore wind corridors in Asia. Taiwan’s government has promoted the area aggressively as part of its broader energy transition, offering development rights to international operators willing to build at scale. The 2b and 4 phases that TSMC contracted represent a significant portion of that buildout. Ørsted completed construction and brought the farm to full commercial operation in 2024, meaning the turbines are now actively delivering power into the grid on TSMC’s behalf.
The exact pricing terms of the PPA have not been disclosed publicly. Without knowing the tariff structure, escalation clauses, or contract duration beyond broad references to “long-term supply,” it is difficult to calculate precisely how the deal compares to Taiwan’s regulated grid rates or how much of a hedge it provides against fossil fuel price swings. What is clear is that TSMC locked in a fixed-cost power source years before the AI-driven surge in electricity demand made such contracts far more valuable and harder to secure.
A strategy that predates the AI boom
One detail worth underscoring: TSMC signed this agreement in mid-2020, well before ChatGPT launched in late 2022 and triggered the current wave of AI infrastructure spending. The company joined the RE100 initiative the same year, committing to source 100 percent of its electricity from renewables by 2050. That timeline suggests TSMC’s leadership was already modeling a future in which electricity supply would become a competitive bottleneck, independent of any specific demand catalyst.
The AI boom has since validated that assumption and accelerated the timeline. Nvidia’s data center revenue alone grew from $15 billion in fiscal 2023 to over $115 billion in fiscal 2025, and virtually all of those GPUs were fabricated at TSMC. Each surge in AI chip orders translates directly into higher utilization at TSMC’s most power-intensive facilities. The company that anticipated rising electricity needs before the boom began now finds itself at the center of it.
Whether TSMC has signed additional wind or solar PPAs since the Ørsted deal is less clear. Industry reporting has referenced further renewable energy contracts, but no official filings or press releases confirming exact volumes, counterparties, or pricing have surfaced in primary documentation as of June 2026. TSMC’s annual sustainability reports disclose aggregate renewable energy procurement but do not break out individual agreements in detail.
What this signals for the semiconductor industry
TSMC’s approach to energy procurement has implications well beyond its own balance sheet. The Ørsted deal demonstrated that a single industrial buyer can underwrite an entire large-scale renewable project, a model that had previously been associated with tech giants like Google and Amazon buying wind and solar to power data centers. For other heavy manufacturers facing similar grid constraints, TSMC’s playbook offers a template: contract directly for dedicated generation capacity rather than relying on the grid alone.
Competitors are watching. Samsung and Intel both operate power-hungry fabs and face their own energy sourcing challenges as they ramp advanced-node production. Neither has publicly matched the scale of TSMC’s offshore wind commitment, though both have expanded renewable procurement in recent years. If Taiwan’s grid pressures intensify, or if other semiconductor hubs face similar constraints, the TSMC-Ørsted model could become standard practice rather than an outlier.
For Taiwan itself, the dynamic creates a tension. The island’s most economically critical industry is growing faster than its power infrastructure can comfortably support. TSMC’s willingness to secure its own supply through private contracts eases some of that pressure, but it also highlights the gap between what the grid can deliver and what advanced manufacturing requires. Government planners at the Ministry of Economic Affairs have acknowledged the need for accelerated grid investment, but building new generation and transmission capacity takes years, and AI-driven demand is growing now.
Where the evidence stops
The verified core of this story is narrow but solid: TSMC signed a full-output PPA for a 920 MW offshore wind farm, the project is now operational, and the company has been building its renewable energy strategy for at least six years. Those are documented corporate actions, not forecasts or speculation.
The broader narrative connecting that strategy to Taiwan’s grid strain and AI chip demand is supported by strong circumstantial evidence but not by a single authoritative source tying all three threads together. TSMC has not publicly stated that it pursued wind power specifically because of AI-driven electricity concerns. The grid pressures are real and widely reported, but precise figures on how much of the strain is attributable to semiconductor fabrication versus other industrial growth remain difficult to pin down from institutional data.
What the record does show is a company that treated electricity as a strategic input long before the rest of the industry caught up. The 920 megawatts spinning in the Taiwan Strait are not a branding exercise. They are infrastructure, built to keep the world’s most important chip factories running when the grid cannot guarantee it will.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
