AI power consumption will rise in 2025 and remain a major problem, while few generative AI companies will be transparent about their usage.
Resolved true The record shows it happened or was so.
Prediction: Statements about events outside the person's control, checked once they fell due: did they come true?
What would count as true or false
The claim has three parts: (1) AI power consumption rose in 2025; (2) it remained a major problem, meaning it was widely reported as a significant concern for energy grids, climate or costs; (3) few generative-AI companies were transparent about their power use, meaning most leading generative-AI companies did not publish specific energy figures for their AI models or services. True: all three parts held during 2025. Mostly true: all three broadly held, with a minor exception, for example one or two major companies publishing partial per-query energy figures. Mixed: two parts held and one did not, or a part held only partly. Mostly false: only one part held. False: none held, for example AI power use was flat or falling, was not treated as a major concern, and most leading generative-AI companies disclosed their usage.
Set before the research began.
Evidence (4)
-
E1 Data centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions - News - IEA International Energy Agency (IEA) · 11 September 2026
“Electricity demand from data centres soared by 17% in 2025, and that of AI-focused data centres climbed even faster – well outpacing growth in global electricity demand of 3%.”
-
E2 Data centre electricity use surged in 2025, even with tightening bottlenecks driving a scramble for solutions - News - IEA International Energy Agency (IEA) · 11 September 2026
“Over the past year, social concerns around AI have also grown, with data centres serving as a highly visible flashpoint for concerns around energy prices and the environment.”
-
E3 Transparency in AI is on the decline Stanford University (Stanford HAI story, 2025 Foundation Model Transparency Index) · 30 October 2026
“10 companies disclose none of the key information related to environmental impact: AI21 Labs, Alibaba, Amazon, Anthropic, DeepSeek, Google, Midjourney, Mistral, OpenAI, and xAI. This includes no information about energy usage, carbon emissions, or water use.”
-
E4 AI giants stay quiet on climate as California prepares to force disclosure Los Angeles Times (Bloomberg story) · 12 August 2026
“The companies haven't disclosed their greenhouse gas emissions, made net-zero pledges or published sustainability reports.”
Reasoning
The IEA reports that data-centre electricity demand rose 17% in 2025, and that demand from AI-focused data centres grew even faster (E1). The IEA also describes data centres over the past year, in a September 2026 release, as a highly visible flashpoint for concerns about energy prices and the environment (E2). Stanford's 2025 Foundation Model Transparency Index finds that ten leading AI companies, including OpenAI, Google, Anthropic, Amazon and xAI, disclose no energy-usage information (E3). Bloomberg separately reports that major AI firms had not disclosed their emissions (E4). All three parts of the claim are supported by independent sources.
Cite this claim
Gary Marcus, “AI power consumption will rise in 2025 and remain a major problem, while few generative AI companies will be transparent about their usage.” (said 1 January 2025). Verdict: resolved true. Reliability API, claim c_ca58f05fd92f, method 0.1.0+c876030c, dataset release v1.1.2. https://epistacklabs.com/claims/c_ca58f05fd92f
This claim is in dataset release v1.1.2, with the same verdict.
Reply to this claim
More prediction claims by Gary Marcus
-
Mostly true
“Some of its advice will be head-scratchingly bad.”
-
Mostly true
“GPT-5 will surely be better, a lot better than GPT-4.”
-
Resolved true
“GPT-5 really is about to drop, no foolin’.”
-
Resolved true
“It will still make a significant number of shake-your-head stupid errors, in ways that are hard to fully predict.”
GPT-5 will still make a significant number of obvious errors that are hard to predict.
-
Resolved true
“Guardrails (a la ChatGPT) may be in place, but the guardrails will teeter between being too weak (beaten by “jailbreaks”) and too strong (rejecting some perfectly reasonable requests).”
Verdicts are made automatically; how verdicts are made.