AI reasoning models used 30 times more power on average to respond to 1,000 written prompts than alternatives without this reasoning capability or which had it disabled, according to a study released Thursday. The work was carried out by the AI Energy Score project, led by Hugging Face research scientist Sasha Luccioni and Salesforce Inc. head of AI sustainability Boris Gamazaychikov.

The researchers evaluated 40 open, freely available AI models, including software from OpenAI, Alphabet Inc.’s Google and Microsoft Corp. Some models were found to have a much wider disparity in energy consumption, including one from Chinese upstart DeepSeek. A slimmed-down version of DeepSeek’s R1 model used just 50 watt hours to respond to the prompts when reasoning was turned off, or about as much power as is needed to run a 50 watt lightbulb for an hour. With the reasoning feature enabled, the same model required 7,626 watt hours to complete the tasks.

The soaring energy needs of AI have increasingly come under scrutiny. As tech companies race to build more and bigger data centers to support AI, industry watchers have raised concerns about straining power grids and raising energy costs for consumers. A Bloomberg investigation in September found that wholesale electricity prices rose as much as 267% over the past five years in areas near data centers. There are also environmental drawbacks, as Microsoft, Google and Amazon.com Inc. have previously acknowledged the data center buildout could complicate their long-term climate objectives. 

More than a year ago, OpenAI released its first reasoning model, called o1. Where its prior software replied almost instantly to queries, o1 spent more time computing an answer before responding. Many other AI companies have since released similar systems, with the goal of solving more complex multistep problems for fields like science, math and coding.

Though reasoning systems have quickly become the industry norm for carrying out more complicated tasks, there has been little research into their energy demands. Much of the increase in power consumption is due to reasoning models generating much more text when responding, the researchers said. 

The new report aims to better understand how AI energy needs are evolving, Luccioni said. She also hopes it helps people better understand that there are different types of AI models suited to different actions. Not every query requires tapping the most computationally intensive AI reasoning systems.

“We should be smarter about the way that we use AI,” Luccioni said. “Choosing the right model for the right task is important.”

To test the difference in power use, the researchers ran all the models on the same computer hardware. They used the same prompts for each, ranging from simple questions — such as asking which team won the Super Bowl in a particular year — to more complex math problems. They also used a software tool called CodeCarbon to track how much energy was being consumed in real time.

The results varied considerably. The researchers found one of Microsoft’s Phi 4 reasoning models used 9,462 watt hours with reasoning turned on, compared with about 18 watt hours with it off. OpenAI’s largest gpt-oss model, meanwhile, had a less stark difference. It used 8,504 watt hours with reasoning on the most computationally intensive “high” setting and 5,313 watt hours with the setting turned down to “low.” 

OpenAI, Microsoft, Google and DeepSeek did not immediately respond to a request for comment.

Google released internal research in August that estimated the median text prompt for its Gemini AI service used 0.24 watt-hours of energy, roughly equal to watching TV for less than nine seconds. Google said that figure was “substantially lower than many public estimates.” 

Much of the discussion about AI power consumption has focused on large-scale facilities set up to train artificial intelligence systems. Increasingly, however, tech firms are shifting more resources to inference, or the process of running AI systems after they’ve been trained. The push toward reasoning models is a big piece of that as these systems are more reliant on inference.

Recently, some tech leaders have acknowledged that AI’s power draw needs to be reckoned with. Microsoft CEO Satya Nadella said the industry must earn the “social permission to consume energy” for AI data centers in a November interview. To do that, he argued tech must use AI to do good and foster broad economic growth.

Cyber Monday alone generated $1.03 billion (a 4.2% increase YoY) in online BNPL sales with most transactions happening on mobile devices, per Adobe Analytics. Overall, consumers spent $14.25 billion online on Cyber Monday. To put that into perspective, BNPL made up for more than 7.2% of total online sales on that day.

As for Black Friday, eMarketer reported $747.5 million in online sales using BNPL services with platforms like PayPal finding a 23% uptick in BNPL transactions.

Likewise, digital financial services company Zip reported 1.6 million transactions throughout 280,000 of its locations over the Black Friday and Cyber Monday weekend. Millennials (51%) accounted for a chunk of the sizable BNPL purchases, followed by Gen Z, Gen X, and baby boomers, per Zip.

The Adobe data showed that people using BNPL were most likely to spend on categories such as electronics, apparel, toys, and furniture, which is consistent with previous years. This trend also tracks with Zip’s findings that shoppers were primarily investing in tech, electronics, and fashion when using its services.

And while some may be surprised that shoppers are taking on more debt via BNPL (in this economy?!), analysts had already projected a strong shopping weekend. A Deloitte survey forecast that consumers would spend about $650 million over the Black Friday–Cyber Monday stretch—a 15% jump from 2023.

“US retailers leaned heavily on discounts this holiday season to drive online demand,” Vivek Pandya, lead analyst at Adobe Digital Insights, said in a statement. “Competitive and persistent deals throughout Cyber Week pushed consumers to shop earlier, creating an environment where Black Friday now challenges the dominance of Cyber Monday.”

This report was originally published by Retail Brew.

The Future of Life Institute’s latest AI safety index found that major AI labs fell short on most measures of AI responsibility, with few letter grades rising above a C. The org graded eight companies across categories like safety frameworks, risk assessment, and current harms.

Perhaps most glaring was the “existential safety” line, where companies scored Ds and Fs across the board. While many of these companies are explicitly chasing superintelligence, they lack a plan for safely managing it, according to Max Tegmark, MIT professor and president of the Future of Life Institute.

“Reviewers found this kind of jarring,” Tegmark told us.

The reviewers in question were a panel of AI academics and governance experts who examined publicly available material as well as survey responses submitted by five of the eight companies.

Anthropic, OpenAI, and GoogleDeepMind took the top three spots with an overall grade of C+ or C. Then came, in order, Elon Musk’s Xai, Z.ai, Meta, DeepSeek, and Alibaba, all of which got Ds or a D-.

Tegmark blames a lack of regulation that has meant the cutthroat competition of the AI race trumps safety precautions. California recently passed the first law that requires frontier AI companies to disclose safety information around catastrophic risks, and New York is currently within spitting distance as well. Hopes for federal legislation are dim, however.

“Companies have an incentive, even if they have the best intentions, to always rush out new products before the competitor does, as opposed to necessarily putting in a lot of time to make it safe,” Tegmark said.

In lieu of government-mandated standards, Tegmark said the industry has begun to take the group’s regularly released safety indexes more seriously; four of the five American companies now respond to its survey (Meta is the only holdout.) And companies have made some improvements over time, Tegmark said, mentioning Google’s transparency around its whistleblower policy as an example.

But real-life harms reported around issues like teen suicides that chatbots allegedly encouraged, inappropriate interactions with minors, and major cyberattacks have also raised the stakes of the discussion, he said.

“[They] have really made a lot of people realize that this isn’t the future we’re talking about—it’s now,” Tegmark said.

The Future of Life Institute recently enlisted public figures as diverse as Prince Harry and Meghan Markle, former Trump aide Steve Bannon, Apple co-founder Steve Wozniak, and rapper Will.i.am to sign a statement opposing work that could lead to superintelligence.

Tegmark said he would like to see something like “an FDA for AI where companies first have to convince experts that their models are safe before they can sell them.

“The AI industry is quite unique in that it’s the only industry in the US making powerful technology that’s less regulated than sandwiches—basically not regulated at all,” Tegmark said. “If someone says, ‘I want to open a new sandwich shop near Times Square,’ before you can sell the first sandwich, you need a health inspector to check your kitchen and make sure it’s not full of rats…If you instead say, ‘Oh no, I’m not going to sell any sandwiches. I’m just going to release superintelligence.’ OK! No need for any inspectors, no need to get any approvals for anything.”

“So the solution to this is very obvious,” Tegmark added. “You just stop this corporate welfare of giving AI companies exemptions that no other companies get.”

This report was originally published by Tech Brew.