It hurts to see bad people win, but Meta, the company that burned billions trying to make the metaverse a thing, remains eye-wateringly profitable, thanks almost entirely to its advertising business, which accounted for 98 percent of its total revenue in the second quarter of 2025. And soon, if Meta executives have their way, that advertising will be powered by so-called artificial intelligence—and shocking quantities of fossil fuels—all so they can convince you to buy more stuff.
In their earnings call October 29, Meta executives touted the growth of their tentpole advertising product, “Advantage+” which seems tuned toward letting companies fire marketing employees and replace them with AI. “The number of advertisers using at least one of our video generation features was up 20% versus the prior quarter,” Meta CFO Susan Li told investors. “We’ve also added more generative AI features to make it easier for advertisers to optimize their ad creatives and drive increased performance.”
This is one application for generative technology that could actually be profitable. The advantage of generative software, as Jason Koebler recently explained in a 404 Media report describing the call, is that it can automatically produce reams and reams of images and videos, each slightly different. These plug into Meta’s preexisting trove of user preference data, ushering in a completely unprecedented era of personalised marketing.
This is bad for a collection of reasons you’ll immediately guess. Automated generation of text, images, and videos from a massive database of mostly stolen human-created content frequently spews out results that are nonsensical, racist, or false, and sometimes violent or abusive. In Meta’s case, specifically, Business Insider reports one company that found Meta’s software automatically replaced a top-performing ad for millennial clothing with “an AI-generated photo of a cheerful yet unnatural granny sitting in an armchair.”
Meta’s own terms for its generative AI tools warn that the outputs of its marketing tools could be “inaccurate, incomplete, misleading, offensive and/or inappropriate,” and make no guarantee that “the Ad Creative AIs will be safe, secure or error-free, or will function without disruptions, delays or imperfections,” nor that the outputs will not “infringe third-party rights.”
In other words, the secret sauce here will not be the marketing genius of “intelligent” generative software systems. No, Meta’s advertising system will “work” because it will brute-force the mass production of a billion buckets of plagiarized shit to throw at a wall, to see what sticks.
Mass-producing sensory visual triggers to get people to buy more stuff will be bad both for people and the planet. Excessive purchase and discard of goods bought online is already a clear and major contributor to the accelerating climate crisis. But there is a more direct and likely more significant way that this trend will worsen the climate crisis.
As many people already know, generating a short sentence comes with a small energy cost. It’s becoming common for tech companies to disclose the energy cost of a single query sent to a chatbot, like Google’s Gemini or OpenAI’s ChatGPT.
But large amounts of text or image generation are much more energy intensive. Generating video is comfortably the worst, by orders of magnitude. One recent study found that “video diffusion [is] roughly 30× more costly than image generation, 2,000× than text generation, and 45,000× than text classification” in terms of energy, and expanding the dimensions of the video (height, width, and time) comes with a quadratic increase in the energy cost, implying “rapidly increasing hardware and environmental costs.”
Companies’ disclosures about the energy needed to use their tools are misleading for another reason too: The bulk of the energy draw isn’t from consumers intentionally triggering queries. Rather, these systems are being tightly woven into the fabric of our digital existence. As a recent study by French academics laid out, features like Google’s “AI Overview” ensure that you are activating generative output whether you like it or not.
There’s deep historical precedent for developments like this one benefiting fossil fuels. The plastics manufacturing industry fabricated its own necessity, an outgrowth of the fossil fuel industry’s strategy to extend the date of its death as far into the future as possible. More recently, the car industry’s bloating of vehicle models to enhance profit margins directly benefits an oil industry anxious about the growth of electric alternatives.
Now Meta’s decision to go all in on the single most energy-intensive form of generative software, in a way that can only work through brute-force mass production and cannot be boycotted or turned off by users, will almost certainly boost a fossil fuel industry still smarting from the unexpected growth of wind, solar, and climate regulations.
Fossil gas has been America’s fastest-growing climate pollution source for some time, but the latest “Global oil and gas exit list” published by German nongovernmental organization Urgewald found that the U.S. has just overtaken China as the number one global developer of new fossil gas power stations, and that “37% of the capacity proposed in the US in the last two years is linked to data centers and AI infrastructure.”
As the Trump administration enacts “permitting reform” to allow tech companies to bring their own fossil fuel plants, Meta has become an industry leader in building new fossil gas to power generative systems. Entergy, a utility that recently announced it’s expecting to fail its 2030 climate targets, is partnering with Meta in the construction of more than two gigawatts of new fossil gas.
Meta has also grown notorious for using ominously named LLCs (“Will-Power,” “Sidecat”) and projects (“Socrates,” “Plato South”) to mask its direct involvement in increasingly controversial developments.
In its latest sustainability report, Meta disclosed a 5x increase in its electricity consumption since 2018 and a corresponding increase in the best-estimate power grid emissions associated with that. But by purchasing questionable renewable energy certificates, the company estimates its “adjusted” emissions will still meet climate targets.
In 2021, every active Meta user cost around one kilogram of carbon dioxide equivalent, or CO2e—a way of measuring greenhouse gases—in terms of direct and unadjusted power grid emissions. By 2024, that had risen to 1.8 kilograms of CO2e. Similarly, every $1 million of revenue caused about 26 tonnes of CO2e in 2021. That’s now 36 tonnes. Meta did not respond to a request for comment for this story.
Meta’s push to force-feed advertising slop into every single corner of the massive digital space it controls could not have worse timing. The country it’s headquartered in is pouring cash into resuscitating coal and accelerating gas. The United States just saw the first two half-year periods with an increase in coal generation for at least half a decade (probably longer), as the Trump administration sets about pouring millions into keeping coal on life support.
Power demand is not neutral. The climate crisis and the need to update energy systems from dangerous to safer means every new megawatt-hour of demand has its own moral valence. There is no choice but to ask whether any of this is needed. If so, do we need it more than we need that precious electricity for hospitals, electrification, or heating and cooling during climate-induced disasters? The answer, for Meta’s great big video-slop tsunami, is an absolutely solid no.
