We don’t think much about our carbon footprint when we punch some words into Google. Yet did you know your average monthly Google searches are roughly equal to running a 60-watt light bulb for three hours?
Thirteen years ago, Google revealed a closely held industry secret: The company’s data centers — massive facilities across the globe that house millions of computer servers running 24/7 — continuously draw almost 260 million watts of energy. That’s enough to power roughly 200,000 homes, and that consumption has likely grown as Google has built even more data centers.
Now multiply a Google search by 10. That’s roughly how much electricity it takes for OpenAI’s ChatGPT model to process a single request. The promise of artificial intelligence, we’re told by our billionaire tech overlords, is that it will help solve intractable real world problems through increased productivity and innovation. In Texas, though, we fear the artificial intelligence revolution could crash our wobbly grid by adding way more load to the electric grid than it can currently handle. On a global scale, it could undermine efforts to slow climate change. Mike Sommers, the CEO of the American Petroleum Institute, told us recently the vast amount of energy required by AI development would increase the need for “dispatchable” energy sources such as natural gas and coal.
The International Energy Agency projects that the world will need an additional 10 terawatt-hours of electricity generation every year if Google were to one day fully implement AI requests into its search engine. That’s the equivalent of lighting 10 million homes for a year. Here in the U.S., data centers already account for 4% of power consumption, but that number will surely grow.
Grid operators are adjusting their demand forecasts accordingly. Pablo Vega, the CEO of ERCOT, Texas’ grid manager, acknowledged during a recent podcast interview that the speed at which data centers are being built in Texas has “fundamentally changed” how the state will have to plan for transmission.
Texas has roughly 276 data centers across the state. The vast majority of these facilities are in the Dallas area, where companies are taking advantage of affordable real estate and cheaper power costs than most other regions. Last year alone, 386 megawatts of data center space came online in the Dallas-Fort Worth area, the equivalent power demand of about 77,000 homes. Meta is in the process of building an $800 million data center in Temple to support its AI development.
This boom in power demand will surely test a shaky grid which is already warning of power shortfalls months before the summer temperatures keep our air conditioners running all day. ERCOT reported last month that peak power loads will rise 6% by 2030, though that projection seems cautious. It doesn’t fully account for the additional 62 gigawatts of load waiting to connect to the grid, including major energy guzzlers such as data centers, semiconductor plants and cryptocurrency mining facilities. And unlike the computers generating new cryptocurrency that can shut down at a moment’s notice when grid conditions get tight — and get paid handsomely for it — data centers must run continuously to meet the demands of cloud-based computing, streaming, gaming and analytics.
“Historically, we’ve always been able to have years to contemplate a massive manufacturing facility coming online and the potential supply built to support that and transmission to support it,” Vega said. “Today, it’s a whole different paradigm. We need to be able to plan and invest in transmission based on reasonable and prudent forecasts of where load is going to go.”
The good news — or bad if you believe the advancement of artificial intelligence is an apocalyptic nightmare in the making — is that AI technology is rapidly becoming more efficient. Nvidia, one of the leading AI developers, announced in March that its latest AI “superchip” is 25 times more energy-efficient. In the same way that Y2K-era fears that the Internet would hopelessly drain our grid never quite came true, these developments give us reason to believe AI will eventually be a seamless fit.
Yet Vega is correct to underscore the transmission challenges. The fundamental problem with Texas’ grid is our transmission lines can’t funnel energy where it’s most needed. Even when energy-rich areas such as West Texas have wind turbines producing massive amounts of electricity, it often ends up congested because there is not enough transmission capacity to get it to urban centers such as Houston. That should soon change thanks to a bill passed by the Legislature last session which allows the Public Utility Commission to expedite transmission development by requiring the state to consider future load growth in permitting construction.
What would help inform this construction planning is if we could track the energy impacts of artificial intelligence data centers, in Texas and across the nation. Sen. Ed Markey, D-Mass., has introduced a bill that would do just that. This bill would order the National Institute of Standards and Technology to develop needs and standards for AI’s energy usage. It would also require the EPA to study the environmental impacts of AI within two years. We urge Congress to pass this bill.
AI development doesn’t have to be all doom and gloom. We believe in its positive applications, even as we’ve warned that it requires substantial regulation. We are eager to see Texas play a role in realizing AI’s promise, as long as we can keep the lights on to do so.
Houston Chronicle
