Exclusive: Amazon pursues "emissions-first" AI energy strategy

Zoom in: According to a blog post released Tuesday morning and shared first with Axios, Amazon is turning to AI and machine learning techniques to better utilize new battery storage systems. The company is pairing those with solar installations.

• This may allow for improvements in how carbon-free electricity is dispatched onto the grid at times when the sun isn't shining or the wind isn't blowing.
• In addition, when combined with other data, Amazon and its partners are developing predictive models to anticipate when power would best be distributed to the grid or stored in batteries on-site, based on future weather conditions and other factors.

Reality check: Improving the efficiency and utilization of battery storage through AI and machine learning will help, but it is unlikely to be enough to meet the onslaught of power demand from new data centers built to handle the generative AI rush along with other aspects of a decarbonized economy.

• AI, heralded as a game-changer for how nearly every industry works, is quickly gaining a reputation as an energy hog.
• Amazon is seeking to portray positive use cases for AI, illustrating that it cuts both ways — it can both use more energy and help companies figure out new methods to save electrons.
• More broadly, tech execs and many policymakers see AI as a double-edged sword. It presents challenges from the standpoint of its electricity demands, while also potentially helping to devise novel emissions-cutting breakthroughs.

Threat level: Microsoft released its 2024 sustainability report last week, which revealed a 30% increase in carbon emissions tied to the company's rush into generative AI.

• "The rise in our Scope 3 emissions primarily comes from the construction of more data centers and the associated embodied carbon in building materials, as well as hardware components such as semiconductors, servers, and racks," wrote Microsoft president Brad Smith and Melanie Nakagawa, the chief sustainability officer.
• Amazon Web Services leads Microsoft in terms of market share, and the company has a net zero carbon emissions by 2040 target.

Yes, but: Amazon declined to respond directly to Microsoft's emissions increase. Instead, a spokesperson detailed some of Amazon's activities in the energy space, noting the company is working on new sources of carbon-free energy, "including nuclear, wind and solar, and emerging technologies."

• "We're also focused on making electrical grids cleaner and more reliable for everyone, which includes advocating for the modernization of grid infrastructure, and enabling renewable energy projects in locations that still rely heavily on fossil fuels," they said.

The intrigue: Amazon highlighted a few facilities and partners it is working with on battery storage to grid technologies, as well as predictive modeling to optimize renewable energy assets.

• The Baldy Mesa solar farm in the Mojave Desert, for example, contains a battery storage system that, when paired with machine learning models powered by Amazon Web Services, helps to anticipate when and how the batteries should store or sell energy into the grid.
• Amazon is also working on an AI model that would utilize data from the company's San Bernardino Air Hub, which has a 2.5 MW battery energy storage unit.
• This model, focused on offering predictive insights on site performance and energy generation, could benefit other parts of the company and beyond.

Zoom out: Given Amazon's global reach, it is also working with companies internationally to use AI in ways that would reduce energy consumption or optimize energy storage systems.

What they're saying: Chris Roe, Amazon's environment director, told Axios in an interview that the company is approaching predictive systems for renewable energy facilities, among other actions, with an "emissions-first approach."

• That means focusing on what steps are best to cut emissions from the grid, be it optimizing the use of battery storage predictive systems or purchasing even more renewable power.
• "The cool thing about it is processing billions of data points a year and, five years ago, we would not have been able to understand what's the right way to optimize the performance of the system" in relation to grid needs, emissions-cutting opportunities and other priorities, Roe said of the Baldy Mesa facility.

The bottom line: Battery storage predictive systems are one of many promising AI applications that could reduce a major company's emissions.

• But at the end of the day, the coming surge in data center energy demand could still overwhelm such gains.