Modern data center servers room with neon lights AI iot learning 3d rendering

AI is Here: Watch Your Electric Bill Start to Climb

July 8, 2024

Over the last twenty years, modest increases in America’s energy demand have been offset by more efficient energy production. However, the era of consistent electricity use is over.

For the remainder of the decade, US power demand is expected to climb between 13% and 15% yearly, according to forecasts from McKinsey, Boston Consulting Group, and S&P Global. That kind of growth is meteoric in comparison to former decades and greatly exceeds the capability of US electrical generators.

Among the primary forces behind this explosive rise in the need for power? synthetic intelligence.

Apart from the energy needed to charge your laptop, most people don’t consider computers to be very energy-intensive devices. However, your request is processed in a distant data center when you input a prompt into ChatGPT (or any large language model), and producing that response requires power.

Each model’s training requires a lot of energy as well: These AI models are fed an enormous amount of data that has been scraped from websites, Wikipedia, Reddit, and YouTube videos that have been transcribed. Thousands of hours are spent processing this data continuously by hundreds of graphics processing units, which are electronic circuits that can do quick mathematical calculations. And for all of that, gigawatts upon gigawatts of electricity are needed on a scale that makes the use of data centers in the past seem archaic.

This is made worse by a nationwide decarbonization initiative centered on switching to an electric grid fueled by clean energy sources from a variety of energy-consuming sources, including manufacturing, transportation, and heating (boosted by the CHIPS and Science Act and the Inflation Reduction Act).

As a result, as the demand for AI grows and as industry and autos move to the grid, a large percentage of the generators in the US grid are kept “on” nonstop, driving up costs.

The demand for energy fluctuates throughout the day and year, so the grid must have a flexible supply that ramps up during evenings when everyone is home as well as during hot spells and periods when air conditioning use soars.

However, utilities switch on peaker plants—which are meant to ramp up quickly but result in very expensive power output, costing $1,000 per megawatt—when demand exceeds the availability of electricity from these base power plants. The cost of electricity will climb exponentially for households, schools, and hospitals as a result of increasingly higher demand for the resource due to electric vehicles, AI models, and expanding manufacturing footprints.

In summary, energy prices with current generation are nonlinear. Because utilities are increasingly depending on expensive peaker facilities to meet demand, a 15% increase in power demand does not translate into a 15% rise in price. A doubling of household energy expenses is certainly conceivable given exponential cost increases. Beyond cost, we cannot justify turning off dirtier energy sources of generation, which would impede the advancement of climate targets, if the grid is straining to keep up.

In less than ten years, the grid infrastructure will not be able to meet the quickly increasing demand, even when all sources are operating continuously. This drives us closer to a future that most Americans can’t fathom: an unstable energy system characterized by frequent blackouts and brownouts. Remember the 2003 East Coast blackout that affected 50 million people, and the blackouts that occurred throughout Texas in the winter of 2021? This time, though, it won’t be a result of unfavorable weather or a safety mishap. Rationing electricity will become the norm in America as the system scrambles to expand its capacity.

By introducing the Federal-State Modern Grid Deployment Initiative, the Biden White House aims to draw attention to the nation’s increasing susceptibility and inadequate state of the national grid. The initiative’s nearly entire focus on boosting current infrastructure through grid-enhancing technology and lack of significant plans to expedite and fund new energy infrastructure match the enormity of the challenge, even though the diagnosis of the problem is accurate.

Furthermore, due in part to inadequate transmission of the current grid, the approval process for additional power generation to be connected to transmission networks has recently reached a five-year wait period. Currently, a power transmission line needs governmental approval to run through them. This is a very time-consuming procedure that frequently involves talking to nearby property owners and taking into account issues related to engineering, economics, safety, the environment, and historical preservation.

A single central monitoring authority is necessary for a truly national grid. The Streamlining Interstate Transmission of Electricity (SITE) Act suggests combining these powers. To meet the energy demand of an AI-powered, green economy, this must be done in tandem with a national push to develop and finance new electrical generation and transmission on the size of New Deal period building.

Abundant power is necessary for America to fulfill its goals in AI leadership, decarbonization, chip manufacturing, and electric vehicle production. However, even an AI chatbot is unable to circumvent onerous bureaucratic procedures that jeopardize grid upgrades. To do that, we need policies designed to quickly build and put new power generation online.

Keep Reading

Modern data center servers room with neon lights AI iot learning 3d rendering