In Amazon’s second quarter earnings call, president Andy Jassy said the company’s cloud computing arm AWS has more demand than capacity — and its “single biggest constraint” is power.
“I don’t believe that we will have fully resolved the amount of capacity we need for the demand that we have in a couple of quarters,” Jassy said in response to an analyst question. “I think it will take several quarters.”
AWS is just one example of a hyperscaler grappling with more demand than it can handle. The boom in artificial intelligence has strained the already congested grid as Amazon and its peers race for power. New data centers to power AI are flooding utilities with load requests; certain facilities are having to wait nearly a decade for power; and phantom requests are complicating things further.
Against this backdrop, power access has become the most important input for tech companies looking to build — more important even than fiber infrastructure, which long governed where data center clusters sprang up. And a partnership announced last year between Google, Intersect Power, and TPG Rise Climate is an example of the “energy-first” development models that are shifting how data centers get built.
Meanwhile, tech giants are increasingly prioritizing speed to energy over clean energy, despite their ambitious net zero commitments. Clean firm power, however, remains the holy grail, prompting a boom in investment in nascent advanced nuclear technologies, enhanced geothermal, and long-duration energy storage.
Amazon, for instance, invested $500 million in the advanced reactor company X-energy and partnered with Energy Northwest to deploy a nuclear project in Washington. (That said, restarting shuttered nuclear plants is arguably the faster route. Microsoft announced a year ago that it will restart the remaining reactor at Three Mile Island, and just last week it was reported that the power giant NextEra is aiming to restart Iowa’s closed Duane Arnold facility as well, and has asked FERC to reclaim interconnection rights for the plant over a proposed solar farm. Amazon so far has not made any moves to tap existing large-scale nuclear.)
AWS has also pushed efficiency for its data centers; in an interview with Latitude Media in late 2024, AWS VP of global data centers Kevin Miller said that efficiency in the age of AI is necessitating that the company rethink how facilities are designed and operated. These include simplifying electrical and mechanical systems, repositioning racks to reduce stranded power, and rolling out a new control system.
AWS grew 17.5% year-over-year in the second quarter of 2025; Amazon as a whole beat expectations.
AWS is spending a combined $31.4 billion in quarterly CapEx on chips, data centers, and power, and has plans to continue to invest in all three “to pursue this unusually large opportunity that we have in generative AI,” Jassy said.
But AI’s growth — and especially the potential of the collaboration of robotics and generative AI — is still in early days, Jassy said; the company expects AWS’ $123 billion annual revenue run rate to increase as the market continues to escalate.
Rising emissions
The earnings call made no mention of the emissions resulting from its data center growth, nor of the company’s commitment to achieving net zero across its global operations by 2040. However, just last month Amazon released its 2024 sustainability report, which like its 2023 report doesn’t report emissions or electricity use tied to AWS or data centers specifically.
The report does provide metrics for the company’s carbon footprint, though, and emissions are growing. In 2024, the company’s total carbon emissions were 68.25 million million metric tons of carbon dioxide equivalent gases, up from 65.1 MMT in 2022. And the company’s direct emissions have increased as well, from 13.02 MMT of carbon dioxide equivalent gases in 2019, to 15.13 MMT in 2024. (In 2019, long before the AI boom took off, direct emissions were 5.76 MMT.)
These numbers are virtually impossible to compare to those of Amazon’s peers. Last month, however, the company published its updated “carbon methodology,” which it said meets the “widely adopted international standard of the GHG Protocol.” That Protocol actually allows for a wide range of reporting approaches, such that there is no true international standard — a problem that the rise of AI has thrown into sharp relief.
In 2023, Amazon was dropped from the UN’s Science Based Targets Initiative, meaning that its reporting methods don’t align with SBTI requirements.
