The "electricity hunger" of American tech giants: When Amazon and others encounter Buffett's power supply crisis

Amazon recently filed a complaint with the Oregon regulatory authority, accusing Pacific Power, a subsidiary of Berkshire Hathaway, of failing to fulfill its contract to provide the promised electricity for its four new data centers. Among them, one is short of electricity, another is not powered at all, and the remaining two have not even completed the contract process. What seems to be a commercial dispute actually reflects the increasingly acute contradiction in the digital age: AI and cloud computing are expanding rapidly, while the most basic energy supply is struggling to keep up.

This dispute originated from the power supply agreement between the two sides in 2021. According to the plan, Amazon's data centers were supposed to receive stable electricity to support the growth of its cloud business. However, the project has currently come to a standstill. More symbolically, although Pacific Power is affiliated with energy giant Warren Buffett, it is also difficult to meet the seemingly reasonable demands of technology companies, highlighting the obvious disconnection between modern energy infrastructure and the expansion speed of the digital economy.

This reflects a core paradox: digital technologies, which are regarded as enhancing efficiency, actually rely on rapidly expanding energy consumption. Today's large data centers consume as much electricity as medium-sized cities and require a large amount of industrial-grade water resources for cooling. Meanwhile, ordinary households and small businesses may bear the rising energy prices due to power shortages. Microsoft CEO Satya Nadella pointed out that the key bottleneck of the AI industry is no longer chips, but the lack of a "computer room shell" that can continuously supply power and dissipate heat.

Amazon's predicament is not an isolated case but a global trend. In Ireland, data centers consume approximately one fifth of the country's electricity. In Chile, they compete with communities for water resources; In a certain place in Mexico, a Microsoft data center has even triggered a public health crisis. Technology companies often build these facilities under the name of their subsidiaries, deliberately downplaying their presence and rarely disclosing the real consumption of water and electricity, making it extremely difficult for the public to supervise and for the government to manage.

The "hunger" of AI is not only reflected in electricity but also in water resources. Many data centers are built in areas where land is cheap, approval is fast but the climate is arid. Once the heatwave arrives, its cooling system has to switch to a more power-consuming mode, bringing the already tight power grid close to instability. In addition, the industry prefers to build super-large industrial parks, but cross-regional redundancy is limited. Once extreme weather conditions are combined with restricted power transmission, it may cause chain failures and bring systemic risks.

To make the AI economy "anti-fragile", it is necessary to change the existing model. Computing tasks should be capable of being dynamically scheduled in response to time, electricity prices, climate and other conditions, and training work should be transferred to regions with abundant renewable energy or low power load. At the same time, a system for disclosing energy consumption and water usage in data centers should be established to enable regulatory authorities and the public to accurately understand the status of resource consumption. In the long run, the industry needs to enhance its "self-powered" capacity, such as developing geothermal energy in areas with suitable geothermal conditions, building cross-sky energy storage systems, and integrating the waste heat from data centers into the energy supply network.

Reality is forcing tech companies to confront a cruel truth: what is truly scarce is not computing power, but electricity. When "the end of AI is energy", and the energy system needs AI optimization, a new cycle of interdependence is taking shape. The next time we smoothly invoke cloud AI services, perhaps we should realize that what supports it is not the invisible "cloud", but the real, tense and highly competitive energy flow on the ground.