Artificial intelligence has a complicated relationship with climate change — it’s both helping to combat emissions and adding to them as it grows in popularity.
On the heels of the boom of OpenAI’s ChatGPT, artificial intelligence is being considered as a solution to climate change, with industry experts looking to harness its power to optimize electrical grids and increase the efficiency of renewable resources. But the technology at the heart of AI relies on thousands of specialized computer chips that consume an immense amount of electricity – and could emit a significant amount of carbon emissions.
JOHN WHITMIRE DEFEATS SHELIA JACKSON LEE IN HOUSTON MAYORAL RUNOFF ELECTION
By 2027, AI servers could use between 85 to 134 terawatt hours annually, according to an analysis published by the scientific journal Joule. That’s roughly similar to the yearly energy consumption of some countries, such as the Netherlands or Argentina.
“If we start introducing new energy-hungry technologies like AI that could drive up our total power demand, and we are already utilizing our renewable capacity to the maximum extent, it means that something is going to be powered with fossil fuels – and that's, of course, not great from a climate change perspective,” said Alex de Vries, a data scientist who wrote the analysis.
To come up with that projection, Vries used an estimate that Nvidia — the tech company dominating the market for AI hardware — could ship out 1.5 million of its A100 servers by 2025 and multiplied that number by its servers’ electricity use. However, the study acknowledges that these scenarios include the assumption that utilization rates will be 100%, while realistically, rates are likely to fall below that threshold. But another factor to be considered is “overhead electricity consumption” — to cool down its servers, for example, which could push the totals higher.
The question that remains to be answered, however, is how much carbon emissions could be emitted as new technologies emerge in high demand. And the reality is tech companies don’t have a uniform formula for calculating emissions, so it’s nearly impossible to get an exact estimate of their carbon footprint.
“One of the reasons that a lot of companies haven’t done it today is that there’s not really an easy way of doing it,” said Daniel Castro, the Vice President of the Information Technology and Innovation Foundation. “There’s a lot of debate about how they should actually measure [emissions], what parts should be measured — there’s a lot of complicated questions that go into that.”
This underlines the complex dynamic between AI and climate change as tech companies navigate their own pledges to reduce their carbon emissions to zero on net — the technology that constitutes the core of their business model could be the thing that threatens the legitimacy of their climate promises.
Although Nvidia and OpenAI haven't established their own net-zero targets, other tech companies have set individual climate pledges over the last couple of years. Three of the largest tech companies have vowed to be net-zero or carbon-negative by 2030: Apple, Google, and Microsoft.
But with the boom of AI comes a focus on energy efficiency – a crucial component to limiting energy consumption and carbon emissions. In Nvidia’s annual fiscal year 2023 Corporate Responsibility Report, the company emphasized its efforts to bolster energy efficiency and performance, optimizing applications and developing cooling improvements.
“Global data center electricity usage is more than 200 trillion watt-hours per year, accounting for 2% of global energy usage, and that consumption is expected to continue to grow,” the report reads. “This is an unsustainable trend that demands attention, which is why we are building products that aim to advance efficiency and sustainability of all data centers globally.”
Efficiency improvements that have been implemented helped to moderate the growth in energy demand from data centers globally. Since 2010, data center energy use — with the exception of crypto — has grown only modestly, even as global internet traffic has expanded 25-fold, according to the International Energy Agency. But in the most recent years, rapid growth in workloads handled by large data centers has resulted in a substantial increase in energy use, of about 20-40% annually. There is further uncertainty over long-term trends of energy consumption.
Per the IEA, data centers and transmission networks are each responsible for 1% of energy-related greenhouse gas emissions, and emissions from data have grown only gradually since 2010, thanks to efficiency improvements.
According to Castro, the conversation around energy efficiency and costs will grow to be even more relevant as the AI boom progresses and it becomes increasingly more expensive to support the technology. If companies don’t figure out how to be cost-effective, it could prove to be unsustainable — even for a $1.7 trillion company like Google.
“That’s where they have to figure out, ‘Okay, if we want to make this product widely available, we’re either going to reduce the cost substantially … or this is going to be a premium service,’” Castro said.
POTENTIAL STRESSES ON THE GRID
It goes without saying that data centers are big energy consumers — a large cloud service's data center can use as much power as 80,000 households do, according to management consulting firm McKinsey. By 2030, demand is expected to reach 35 gigawatts of power consumption annually, up from 17 gigawatts in 2022.
Much of this is bringing concern that the new wave of artificial intelligence will bring its own stresses to the electric grid. But as much of the developed world looks to transition away from fossil fuels toward renewable energy; experts are grappling with the reality that energy sources such as wind and solar are unable to match the capacity needed to power these huge data centers, as they cannot produce energy that’s available all the time.
“If you want clean energy and you want it 24/7, you have to have something called clean, firm energy — energy that you can count on,” said Judi Greenwald, the Executive Director of the Nuclear Innovation Alliance.
Some forms of this “clean, firm energy” include nuclear, geothermal, and carbon capture with existing fossil fuel sources such as natural gas, according to Greenwald. And some companies are embracing nuclear power as federal investment grows in the sector (notably incentives embedded in the Democrats’ Inflation Reduction Act — passed into law last year). Microsoft, for example, is looking to use small nuclear reactors to power its data centers.
Although SMRs are cheaper, safer, and take less time to build than traditional large sources of nuclear, they are still extremely complex systems that will need to comply with stringent safety requirements. Furthermore, approval from the Nuclear Regulatory Commission — an agency responsible for overseeing public health and safety related to nuclear energy — will need to be granted before anything is built. The NRC approved its first SMR design at the beginning of 2023.
However, critics contend the commission is ill-prepared for the approval of new projects, as a litany of problems has slowed the licensing process, ranging from the lack of licensing timelines for certain reactors to the agency having limited personnel to approve the projects. Greenwald’s organization, Nuclear Innovation Alliance, has been calling for reform of the agency for quite some time.
“In the past few decades, we have gotten slower at building things for a whole bunch of reasons, but one of them is that we’ve kind of implicitly decided that it's okay if it takes years or even decades to build stuff,” said Greenwald. “But it’s particularly not okay now that we know about climate change. Modernizing infrastructure is really important, and it’s really important that we make decisions about infrastructure efficiently.”
THE TIDE THAT’S CHANGING
Tech companies have been historically opaque about their carbon emissions, but that has shifted over the last couple of years, with the three largest tech majors reporting their carbon emissions in annual environmental progress reports.
CLICK HERE TO READ MORE FROM THE WASHINGTON EXAMINER
But even without the individual efforts of companies, some states will require mandatory reporting of their emissions. California recently enacted legislation that requires companies making more than $1 billion in annual revenues to report both their direct and indirect emissions — applying to more than 5,300 businesses in the state. Many of the companies residing in Silicon Valley are likely to be affected by the new law, which will likely be enforced by 2026.
On a national level, the Securities and Exchange have proposed rules that would require companies to disclose climate-related information in their annual reports, ranging from companies’ greenhouse gas emissions to environmental risks their operations could impose. However, a final rule has yet to be issued by the agency.