What’s going on with Microsoft’s energy intensity?

One major takeaway from the release of Microsoft’s sustainability report earlier this week: its 29.1% uptick in emissions since 2020. The tech giant attributes that increase to its strategic double down on data centers amid the artificial intelligence boom.

But buried elsewhere in the report was another consequence of the tech giant’s race to AI supremacy: an even steeper rise in energy intensity.

• The top line: The amount of electricity Microsoft consumes per dollar of revenue increased by more than 50% from 2020 to 2023, and 20% since 2022 alone. The company told Latitude Media that the steep rise can be attributed to “growth in our data centers' electricity usage,” raising questions about how the company will get the electrons it needs — much less the clean electrons — as its computing needs and revenue surge.
• The market grounding: The electricity appetite of U.S.-based companies developing AI are predicted to increase by terawatts in less than five years. At the same time, Microsoft and its peers have committed to fast-approaching decarbonization targets that seem increasingly at odds with ambitions of massive computing power.
• The current take: There’s a “real risk” that tech company climate commitments will take a back seat to their AI aspirations, Brian Janous, former VP of energy at Microsoft and CEO of Cloverleaf Infrastructure told attendees at a recent Latitude Media event. “There’s a huge economic pie sitting in front of all these tech companies, which is AI,” he said. ”And whoever can get there first and get the biggest piece of that pie is going to reap rewards for decades to come.”

While 2023 represented a more dramatic increase than in past years, energy intensity has been growing since before the AI boom took off with the release of ChatGPT in November 2022. Meanwhile, Microsoft’s total greenhouse gas emissions per dollar of revenue decreased by around 3% between 2022 and 2023.

In an interview with Latitude Media, Microsoft’s chief sustainability officer Melanie Nakagawa outlined the company’s view that compute power will get more efficient, and that AI’s energy use “may not be longitudinal.” Microsoft is “at the starting point” of seeing just how efficient its model training can be, she added. 

She did not indicate how efficient the company’s activities may become or how quickly that evolution may happen, and Microsoft representatives declined to speculate on whether increased efficiency (or increased AI revenue) would eventually bring down the company’s overall energy intensity. 

Janous, though, is skeptical that increased efficiency would lead to anything other than increased buildout. If Nvidia for example, invented a chip that is twice as efficient as what’s currently on the market, “then Meta or Microsoft or Google will plug in twice as many chips,” he said. “They’re going to build as big of a model as they can.”

That’s what differentiates the AI boom from prior technological eras, he added: “There’s very much an incentive to use more electricity.”

In 2020, Microsoft said it would become carbon-negative in the next decade, reducing emissions and abating via carbon removal more than it emits. And by 2050, the company has also pledged to remove emissions equivalent to its historic releases. 

Skeptics have called these and similar targets greenwashing, because some companies — including Microsoft — have relied in part on carbon offsets to meet them.

The grid problem

That’s not to say that companies aren’t incentivized to become more efficient in their AI activities, though.

“I believe those incentives for efficiency are a function of the fact that these companies don't believe utilities are going to be able to make enough power to meet their demands,” Janous said. 

How much of a grid’s energy comes from renewable sources is a major consideration as Microsoft attempts to meet the immense energy needs of its recent ramp-up on AI, Nakagawa said. 

“We are trying to go to countries and regions and places that offer carbon-free assets and clean power assets for us to reach out and get those clean energy credits or the PPAs,” Nakagawa said. “What we’re finding though, as a global company, there’s still many regions and grids around the world that don’t have a large source of carbon-free assets.” 

Ironically, meeting the steep energy demands of AI giants has pushed several utilities, particularly in the Southeast, to move to build new gas power and delay coal plant retirements. Georgia Power, for instance, decided to add 1.4 gigawatts of gas generation to meet the soaring demand. That move prompted Microsoft to push back in a filing to the state’s Public Service Commission, arguing that the utility was overestimating load growth in the region.

Part of Microsoft’s strategy involves doubling down on PPAs. Just two weeks before the release of its sustainability report, the company announced a 10.5-GW clean energy megadeal with Brookfield, which Nakagawa pointed to as an example of Microsoft’s role as an “energy innovator.”

But renewables projects take a lot longer to get off the ground than a data center. The company’s total energy consumption jumped by nearly 29% between 2022 and 2023 and — though Microsoft contracted nearly 20 GW of renewable electricity last year — its share of direct renewable electricity decreased. Nakagawa attributed that phenomenon to the fact that most economies (and local grids) aren’t on track to meet global climate targets.

For the world’s largest tech companies, who have always faced the challenge of scaling quickly while balancing sustainability, said Janous, there’s a major concern that utilities won’t be able to meet their energy needs, at least not in a clean way. 

“They want to go really fast,” he said. “They want to grow big, and now, what’s defined as ‘big’ is very different than it was five years ago.”