Big Tech Has a Massive Carbon Problem Thanks to AI

As the old joke goes, how do you know if a company will be carbon-negative by 2030? They tell you.

Microsoft famously shared its green promises with the world in 2020, promising to be carbon-negative by the end of the decade.

Yet before you could say “greenwashing“, we saw artificial intelligence take center stage — testing the strength of every organization’s Environmental, Social, and Governance (ESG) commitments.

Since then, many large enterprises have gotten in a jam by pushing back big climate promises so they can jump on the AI hype train instead.

In Microsoft’s case, since Satya Nadella made his original pledge in 2020, its emissions have increased by up to 40%.

In this article, Techopedia investigates how ESG commitments are being quietly downplayed by the global obsession with AI and why it matters.

Table of Contents Table of Contents

1. Key Takeaways
2. Why AI Poses a Problem for Big Tech Climate Goals
3. Is Nuclear Power the Solution to AI's Power Demands?
4. Small Reactors Bring Massive Responsibility
5. The Bottom Line
6. References

Show Full Guide

Table of Contents

1. Key Takeaways
2. Why AI Poses a Problem for Big Tech Climate Goals
3. Is Nuclear Power the Solution to AI's Power Demands?
Show Full Guide
4. Small Reactors Bring Massive Responsibility
5. The Bottom Line
6. References

Why AI Poses a Problem for Big Tech Climate Goals

Did you know that a short 100-word response from ChatGPT uses approximately half a liter of water? With chatbots like Microsoft CoPilot offering their wares to the public, Microsoft’s data centers in Arizona are projected to consume over 50 million gallons of drinking water every year.

It is an insatiable thirst, and this water usage is becoming a concern in hot U.S. states that are already facing water shortages.

ChatGPT currently has over 200 million active users per week, so it shouldn’t be surprising to hear that AI’s water usage could reach 6.6 billion m³ by 2027. To put this figure into perspective, it’s around half the amount the UK consumes yearly.

Every ChatGPT search uses 2.9 watt-hours of electricity, nearly 10 times more than a standard Google search. In some states, like Virginia, data centers consume over 25% of the electricity, adding to the already overstretched power grids.

Gartner predicts that power shortages will restrict 40% of AI data centers by 2027 if we carry on this path without making significant changes.

As big tech continues to underplay the scale of the problem, the Guardian reported that data center emissions could be 662% higher than some tech CEOs would have us believe.

Is Nuclear Power the Solution to AI’s Power Demands?

Amazon and Google recently hit Techopedia’s radar after announcing they would be turning to nuclear power to drive their AI ambitions.

For obvious reasons, this could be a hard sell for PR teams, especially regarding concerns for public safety. But there are pros and cons to this approach.

One of the biggest reasons big tech is attracted to nuclear power is that nuclear energy produces minimal greenhouse gases. It fixes two problems for the price of one by providing a reliable way to continuously power AI systems’ demanding needs while aligning with environmental sustainability goals.

Unlike solar or offshore wind solutions, nuclear power is not dependent on the weather. Small Modular Reactors (SMRs) also open up opportunities for nuclear power to be based near or within data centers, dramatically reducing transmission losses.

SMRs are significantly smaller and generate 50 to 300 megawatts of electricity, compared to 1,000 megawatts for traditional large reactors. They also have simpler designs with fewer moving parts, reducing the likelihood of failures.

Once built, nuclear plants could also offer stable operating costs and greater protection against volatile fossil fuel prices. So what’s the catch?

Small Reactors Bring Massive Responsibility

Big tech doesn’t have the best reputation when it comes to change management. We have all seen how a configuration change or software update can bring down the shutters of any retailer in the world.

But when a mistake is made with nuclear power, the stakes are much higher.

There are three major examples of nuclear meltdowns: Three Mile Island in 1979, Chornobyl in 1986, and Fukushima in 2011, which will always linger in the back of the mind when anybody mentions the word nuclear.

News that Microsoft has entered into a deal to revive a unit of the Three Mile Island plant will be enough to set off a few alarm bells, justified or not.

Despite ticking all of the safety and security boxes, the cost of any mistake has historically been at the expense of the environment and local inhabitants who are forced to flee the affected areas. Rebuilding trust in nuclear will understandably take time.

Despite the powerful promises, SMR designs are still unproven technologies, with questions about long-term performance and reliability remaining unanswered.

Five major hurdles block nuclear power’s path. Waste management, billion-dollar construction costs, regulatory red tape, public fears, and decade-long buildouts.

On paper, the smaller fuel inventory and power output of SMR minimize potential radiological release in case of an accident, and they are designed to withstand extreme events, including natural disasters, without releasing harmful radiation.

But these reassurances will not land with everyone.

The Bottom Line

As the demand for AI increases, the Electric Power Research Institute predicts that Data centers’ electricity demand will double by 2030, consuming up to 9% of U.S. Electricity.

If the pace of climate change continues, the problem could worsen with every AI search, and the rising demand for air conditioning that keeps our homes cool will ironically make the planet hotter.

Amazon, Microsoft an Google all share the belief that nuclear energy will pick up the baton and address the growing energy demands from AI data centers.

But the jury is still out on whether they can deliver on this promise or whether it is merely pushing the problem further down the road. How much road we have left is a debate for another day.