Summary of the Core Content in Plain Language
Recently, there’s been a particularly bizarre but real situation in the industry: many AI data centers that cost hundreds of millions, even billions, are left locked up with their equipment gathering dust and unoperated. People might assume the projects failed because they couldn’t find customers, but that’s not the case. The companies that own these data centers have already secured all the computing power they need; there’s a long queue of companies eager to use AI services, just waiting for the power grid to connect them to the electricity supply. The main issue in the AI industry is no longer a lack of technology or customers, but rather a frantic competition to invest in graphics cards and build data centers. The expansion of computing power far outpaces the speed at which the power grid can allocate and deliver electricity. “Waiting for electricity” has become the biggest obstacle hindering the growth of the AI industry.
---
Detailed Explanation
1. Why Do Data Centers Have to Wait for Electricity?
Imagine an AI data center as a super kitchen dedicated to powering large AI models. All the owners have invested heavily in top-tier equipment (H100 graphics cards, cooling systems, and operational platforms) and hired maintenance staff, but when they go to the gas company to request electricity, they’re told that the gas quota for their area was allocated three years ago and the new pipeline is still under construction, with the earliest availability in 2027. In the past, the electricity demand of ordinary cloud data centers was not so high; a data center’s energy consumption was roughly equivalent to that of a small residential area, and the power grid’s infrastructure was well-planned to meet this demand. However, a medium-sized AI data center now consumes as much electricity as a county with tens of thousands of residents, creating an unexpected additional demand. Since the expansion of the power grid, new substations, and wiring require 3-5 years of planning, no one expected the AI industry to grow so rapidly, and no sufficient electricity capacity was reserved for it. As a result, it’s common for data centers to wait for years just to get power.
2. It’s Not a National Power Shortage; AI’s Demand Outpaces Existing Users
Some might think the problem is a lack of power generation. However, in 2023, there’s an excess of generating capacity nationwide, with much of the energy from wind and solar power going unused. The issue lies with the allocation of power quotas: the grid prioritizes residential and industrial uses. AI’s demand is new, and granting a large quota to an AI data center could deprive nearby manufacturing companies of necessary power, affecting employment and the real economy. It’s like trying to get a table at a popular restaurant when all tables are reserved months in advance; even if you offer triple the price, the owner can’t accommodate you.
3. Preferring to Keep Data Centers Idle to Secure Quotas
Companies are willing to wait for years just to secure a power quota. The competition in the AI industry is so intense that being late by three months could mean losing all major clients. Since power quotas are now in short supply, companies rush to apply, build data centers, and stock up on graphics cards. Even if the data centers remain idle for two years, it’s better than waiting for years to get the necessary power. The empty data centers you see are essentially “power slots” reserved by companies, with expensive equipment sitting idle, incuring daily depreciation costs. Without access to electricity, losing this opportunity would be a significant loss.
4. The Impact on Ordinary People
This situation affects us all. AI data centers’ electricity costs will eventually be passed on to consumers through higher prices for AI services. Additionally, more power infrastructure will be built in areas to support the AI industry, creating new jobs. However, this may also lead to stricter power restrictions for local manufacturers, potentially increasing the cost of goods we buy. In short, the competition for power in the AI industry has far-reaching consequences for everyone.