虎嗅

LSE Research Paper: Are AI-generated electricity bills making everyone pay the price?

原文:LSE研究文章:AI的电费账单,正在让全民买单?

Summary of Key Points

The electricity demand from AI data centers is experiencing explosive growth, and the resulting costs associated with grid expansion, peak power generation, and delayed decarbonization are being quietly passed on to ordinary households through electricity prices. The UK has attempted to mask this structural issue by temporarily reducing the VAT on electricity bills to 0, while China’s “East Data West Computing” initiative aims to alleviate the pressure through spatial distribution. However, the increase in electricity prices in the eastern regions indicates that the demand-side challenges have not been fully resolved. Global responses to this issue are fragmented, and the core debate revolves around one question: Who should actually pay for the electricity bills incurred by AI?

Why Do AI Data Center Bills End Up on Your Account?

AI data centers are power-consuming giants, consuming approximately 415 terawatt-hours of electricity annually (equivalent to 1.5% of the global electricity demand), and this figure is expected to double by 2030. When these large consumers rapidly connect to the grid, they affect ordinary households in three main ways:

1. The “hidden tax” of shared infrastructure: The costs of building substations and transmission lines for data centers are shared by all users, effectively meaning you are paying for the infrastructure development on behalf of tech companies.

2. Peak power consumption driving up prices: Data centers operate at full capacity 24/7, and during peak times, the grid has to rely on expensive peak-load power plants that are not used under normal circumstances. These additional costs are then reflected in the electricity bills of all users.

3. Delayed decarbonization increasing expenses: Overloaded grids force the extension of the operation of old coal-fired power plants, which not only pollute the environment but also slow down the transition to cleaner energy sources, resulting in additional costs for users. For example, the electricity prices in PJM, the largest grid in the US, have increased by 75% in the past year due to the surge in demand from AI data centers.

Why Doesn’t the UK’s Reduction in Electricity VAT Help Your Wallet?

The UK’s reduction of the VAT on household electricity bills from 5% to 0 may seem like a way to save money for consumers, but it only provides temporary relief and does not address the underlying issues:

  • The tax cut is temporary: It only alleviates short-term financial pressure and does not solve the long-term problems of grid expansion and increased power generation costs.
  • Cost allocation is the key: The proper approach would be for data centers to bear the additional costs. For instance, the UK’s energy regulator has suggested that data centers should pay a 2.5%-7.5% “access fee” or fund the construction of their own grid infrastructure. Companies like Google and Meta have already begun to purchase green energy directly to reduce their reliance on the public grid, which is a more sustainable solution. The UK has chosen a politically appealing approach but has not resolved the fundamental issue.

Can “East Data West Computing” Solve the Problem?

China’s “East Data West Computing” initiative is a smart attempt to relocate data centers to the western regions, where electricity prices are lower and power supply is more abundant, thereby reducing the burden on eastern areas. However, even with these measures, electricity prices in eastern cities like Beijing have still risen, indicating that:

  • The relocation of computing power only provides partial relief: The demand for AI services in the east continues to grow, and some data centers, due to their proximity to users (e.g., those requiring low latency services), still need to remain in the east, meaning the demand-side pressure has not been completely alleviated.
  • Regional equity issues exist: While the western regions benefit from investment and job creation, they also have to bear the costs of grid upgrades and the consumption of land and water resources. If these costs are still shared by all users, residents in the west may be subsidizing digital services in the east.

The direction of “East Data West Computing” is correct, but additional measures are needed, such as requiring data centers to bear more of the local costs or providing compensation to residents in the western regions.

Global Response Is Chaotic, and a Fair Rule Is Needed

Currently, countries are acting independently without a unified and fair approach:

  • United States: State policies vary; for example, Virginia imposes taxes on data centers, while North Carolina has abolished tax exemptions.
  • UK: The UK uses tax cuts to mask the problem and avoids addressing the long-term cost allocation.
  • China: There is a national-level strategy, but details such as consumer electricity price protection and direct green energy purchasing mechanisms need to be improved.

The core principle is simple: Whoever benefits should pay, and whoever causes the costs should bear them. AI companies, which profit from their operations, should bear the costs of their electricity consumption and not burden ordinary households and the regions that host data centers. This is not just an energy issue but also a matter of fairness. After all, the development of AI should not lead to higher electricity prices for the general public.

The allocation of electricity costs associated with AI is as important as that of chips and computing power. If not addressed properly, it will not only increase living costs but also exacerbate regional disparities. The key to the future is to establish a cost-sharing mechanism that is acceptable to AI companies, users, and all regions.