GIL 2026 | Frost & Sullivan Georgia Edell: The Impact of Artificial Intelligence on Public Utilities and Broader Fields: Regulatory Fragmentation and Market Disruption Risks

Georgia noted that artificial intelligence exists not only in algorithms and models but also relies on data centers, electricity, and cooling water. In 2025, global data centers consumed approximately 448 trillion watt-hours of electricity, resulting in about 208 million tons of carbon dioxide emissions and around 1.2 trillion gallons of water used for power generation. The “physical footprint” of artificial intelligence is nearly equivalent to that of a country. As the number of data centers continues to increase, power supply, water usage, and environmental costs are gradually turning from technical issues into public concerns.

A recent Gallup poll showed that about 70% of Americans oppose the construction of data centers near their homes, with water usage being one of the main concerns. In the first quarter of 2026 alone, at least 75 data center projects worth approximately $130 billion were affected by local opposition activities. Meanwhile, countries are developing “sovereign artificial intelligence”—capabilities that are controlled by the state. This means more countries around the world are building more data centers.

It is expected that global investment in sovereign artificial intelligence will exceed $100 billion by 2026, with various countries actively participating and increasing investment in computing resources. However, those countries that cannot ensure large-scale resource deployment must pay higher costs and face the risk of falling behind in actual model capabilities:RAND Europe(RAND Europe) found that early 2026, total operating computing capacity in Europe was approximately 123,000 H100 equivalent chips, while in the United States it was about 1.4 million. Even with continuous investment, the gap between the two regions could remain around six times by 2030. Thus, resource acquisition is not only related to operating costs but may also determine whether a country can develop competitive models.

Georgia believes that whether due to the needs of large-scale enterprise development or based on their own artificial intelligence capabilities, countries are trying to address the energy issues behind artificial intelligence data centers. These approaches can be summarized into five categories:

First is “coordination”, where governments manage the grid and data center layout. Japan plans to use “Watt-Bit” to restrict low-efficiency projects, the EU promotes mandatory disclosure of energy information and unified rating systems, and China requires new data centers in national computing centers to use at least 80% renewable energy;

Second is “pricing”. For example, U.S. energy regulators require major grid operators to explain the basis for charging large power users such as data centers;

Third is “power bypassing”, allowing enterprises to build on-site or dedicated power sources, but different regions have varying attitudes regarding reliability, cost allocation, and regulatory authority.

Some countries can benefit from natural geographical conditions, such as Brazil’s long-standing hydropower structure. Others impose “additional conditions” on market access, requiring data centers to prove their contribution to energy transition and water security.

Georgia said that these five approaches reflect policy logic related to planning, pricing, private power supply, resource endowment, and access review. The choices of different countries vary, and the construction cycle, energy costs, and compliance requirements for data centers in different markets also create differences.

Georgia believes that as countries deal with resource constraints, this situation brings several distinct risks to enterprises:

The first risk is capital and resource access. Waiting times for grid access in major markets have reached five to eight years, and key equipment such as transformers is in shortage. Large cloud service providers who can lock in power procurement, prepay equipment, and build supporting power sources many years in advance are better able to obtain scarce resources, while ordinary enterprises may face higher electricity prices, indirectly bearing the cost of grid upgrades, and also being affected by equipment competition and rising cloud computing prices.

The second risk is regulatory differentiation. AI governance rules at international, national, and industry levels often fail to be consistent. According to the OECD,OECD, 80 jurisdictions and organizations have over 2,000 AI policies. As rules continue to increase but are not well coordinated, enterprises face duplicate, overlapping, or conflicting compliance requirements. Large multinational companies can allocate professional teams to handle this, but small and start-up companies find it difficult to bear the costs. Therefore, uneven distribution of regulatory complexity may further solidify the monopoly advantages of existing leading enterprises.

The third risk is market fragmentation. Although sovereign artificial intelligence can increase local computing power,localization of data, cross-border transmission restrictions, and regional infrastructure isolation may also create new market barriers. At least 34 countries have introduced or strengthened localization requirements, limiting where artificial intelligence processing can take place. If enterprises are restricted to using artificial intelligence infrastructure within specific regions, they must bear compliance costs, and their computing power will be limited to the level affordable in that region, rather than the best globally available level.

Finally, Georgia emphasized that power resource access, capital barriers, regulatory differences, and market fragmentation are collectively changing business conditions. “Who Gets the Power” refers both to real energy supply and industry influence in the era of artificial intelligence. In the future, those enterprises that can prepare and plan thoroughly for these risks earlier will gain the initiative.