How AI Is Changing Beer

There is a bar in Beijing’s Zhongguancun tech district, China’s answer to Silicon Valley, where you order a 9.9 yuan glass of house lager called AGI, connect your laptop to the bar’s Wi-Fi, and get free, unlimited access to an AI coding model running on two Nvidia DGX Spark computers sitting on the counter. The owner, an AI developer named Song De, says he opened AGI Bar because AI and alcohol share three things in common: hallucinations, bubbles and distillation. He is reportedly losing money on the place. He is opening a second location anyway.

It is a good punchline, and also a fair snapshot of where AI and beer actually stand right now, somewhere between a genuinely useful industrial tool, a marketing gimmick, and a running joke that people can’t quite stop telling. Here is what’s real, what’s hype, and what it might mean for the beer in your hand.

This Isn’t New, It Just Got a Louder PR Team

The instinct is to treat AI in beer as a post-ChatGPT phenomenon. It isn’t. According to a 2024 peer-reviewed overview of the field by Thomas Schlechter of the University of Applied Sciences Upper Austria, published in the academic journal BrewingScience, breweries have been using electronic “noses” and “tongues”, sensor arrays trained with AI to detect off-flavors and classify beer styles, since research on the technology began in 1993. By 2023, that approach could catch common off-flavors like diacetyl with around 95% accuracy using cheap, low-cost hardware.

What changed with ChatGPT wasn’t the technology’s arrival in brewing, it was that suddenly anyone, not just a lab with a spectrometer, could ask an AI model to do something beer-related and get an answer back in seconds. That’s the split worth understanding: quiet, unglamorous AI has been running quality control and fermentation monitoring in breweries for years. Loud, headline-friendly AI showed up in 2023 and started writing recipes.

The Recipe Gimmick Every Brewery Has Now Tried

When Beck’s turned 150 in 2023, the AB InBev-owned German brewery handed the recipe pen to ChatGPT, gave it hops, yeast, water and malt, and got back “Beck’s Autonomous,” a lager one Daily Mail reviewer rated above the standard lineup. It was far from the last. Atwater Brewery in Detroit released an AI-designed citrusy IPA. St. Austell in Britain made a tropical IPA called “Hand Brewed by Robots.” Night Shift Brewing’s Michael Oxton fed ChatGPT a brief and got back a full hazy IPA recipe, instructions and water profile included, in minutes, a job he says would normally take his team weeks of research and testing. Coedo Brewery in Japan asked an AI model to profile drinkers by age bracket and brewed four separate beers to match.

The most serious attempt so far belongs to Beau Warren of Species X Beer Project in the US, who has a background in both beer microbiology and data analytics. Starting in 2022, he trained seven different AI models on his own proprietary recipes, yeast strains, water chemistry and hop data, then used the results to guide real brewing decisions from 2024 onward. In one case, the model suggested pairing Maris Otter malt, normally a stout ingredient, into a lager. “I would never have thought of doing that in a lager, ever,” Warren told The Economist. “We brewed it anyway, and I thought it was one of the best beers we’d ever made.”

Where the Real Money Actually Shows Up: The Fermentation Tank

Recipe stunts get the headlines, but the money is in process optimization, and it isn’t small. Deschutes Brewing in Bend, Oregon, built an AI system called BreweryPi on cheap Raspberry Pi hardware to monitor and optimize fermentation. It delayed the need for an extra chiller and saved the company close to $1 million in the short term, with reported savings reaching roughly $8 million longer-term. Sugar Creek Brewing used AI-based monitoring to catch bottling and filling mistakes that were causing beer to over-foam and get wasted, a fix worth about $30,000 a month. Carlsberg ran a project with Microsoft, nicknamed Carlsberg Fingerprinting, using automated analysis inside its lab to speed up how fast new recipes get developed and tested.

A separate research group built a foam-pouring robot called RoboBEER in Australia, using computer vision to standardize how beer is poured so foam stability and protein content can be measured consistently, the kind of unglamorous quality-control work that never makes a press release but keeps a brewery’s product consistent can to can.

Can AI Actually Predict What Tastes Good?

This is the more scientifically interesting question, and Belgian researchers have made real progress on it. A team at KU Leuven, led by Professor Kevin Verstrepen, built a machine learning model trained on 226 chemical compounds measured across 250 Belgian beers, cross-referenced against tasting scores from a 16-person expert panel and more than 180,000 public ratings pulled from an online beer review site. The model can predict a beer’s flavor profile from its chemistry, and suggest specific tweaks, like adjusting lactic acid or ethyl acetate levels, to improve it. In unpublished follow-up work, Verstrepen says non-alcoholic beers adjusted this way became indistinguishable from their alcoholic counterparts in blind tastings.

A similar project in Australia, run by associate professor Niusha Shafiabady and detailed by Orange’s Hello Future, trained neural networks on 284 production records covering 36 separate brewing parameters, malt type, filtration, boil method, alcohol level, CO2, and so on, to predict beer quality before a batch is even brewed, filling in missing data points the brewery hadn’t recorded.

Both projects land on the same conclusion: AI is good at finding patterns in chemistry and production data that a human brewer would take months to notice, but it still needs a real brewer to decide what to do with that information. Berry Gerrits, a postdoctoral researcher at the University of Twente who trained a model on BrewDog’s open-source recipes and Untappd ratings, found this out directly when his algorithm suggested a recipe that mixed flavor characteristics that don’t actually work together. “An IPA is hoppy, a stout is dark and heavy. You can’t just mix those flavours,” he said. “A brewer had to adjust the recipe. That immediately shows: domain expertise remains essential.”

Big Beer Is Betting on the Boring Version

While independent breweries experiment with recipe gimmicks, the world’s largest brewers are quietly building AI into everyday operations. Heineken rolled out a generative AI tool called CoBrain to more than 150 breweries and 80% of its factory workforce, letting floor staff query decades of internal standards and troubleshoot problems by comparing historical data across its global network. It’s also using AI to cut water and energy use in bottle washers, pasteurizers and wort coolers, part of a wider “Digital Backbone” strategy meant to unify data across more than 70 markets.

It’s worth being precise here: Heineken itself credits that cash flow jump mainly to working-capital timing, not AI directly. The honest read is that AI is one piece of a much larger operational overhaul, not a standalone line item on the balance sheet, which is a more believable story than the version some coverage has been telling.

Not Everyone Is Cheering

The pushback has gotten just as creative as the marketing. In August, challenger brands Liquid Death and Garage Beer, the latter co-owned by NFL center Jason Kelce, released a 90-second music video asking fans to donate their urine to help cool AI data centers, which can use up to 5 million gallons of water a day. The campaign leans into real public sentiment: a Gallup poll found more than seven in ten Americans oppose AI data centers being built in their area.

It is a joke campaign with a real point buried in it. AI’s water and energy footprint is a legitimate industry-wide concern, and beer, an industry that already uses enormous volumes of water to brew, is one of the more pointed places to make that argument.

The Cautionary Tale: IntelligentX

Before any of this was trendy, a London startup called IntelligentX tried to build an entire brand around AI-driven brewing, adjusting recipes based on direct customer feedback collected through a messenger bot. It launched in 2016 to real buzz. It is now bankrupt. The idea wasn’t wrong, using real drinker feedback to steer a beer’s development is exactly the kind of thing AI is good at, but a good technical concept isn’t automatically a good business. It’s the clearest reminder in this whole story that AI can make a brewery smarter without making it succeed.

So, Does Any of This Actually Make Beer Better?

Sometimes, yes, in specific and fairly narrow ways. Non-alcoholic beer is the clearest case, KU Leuven’s flavor-prediction work is directly aimed at closing the taste gap with alcoholic beer, which is currently the single biggest complaint holding the category back. Quality control is another clear win, catching a batch-ruining flaw before it ships is a genuinely good use of the technology, and it’s been quietly working for three decades. Recipe generation is the most overhyped piece of it: it’s a real research tool for an experienced brewer like Beau Warren, and a fun marketing stunt for everyone else, but nobody serious is claiming AI replaces a brewer’s palate.

The most honest summary of all of this comes from the very academic paper that first documented AI’s role in brewing back in 1993: AI is a tool, not a brewmaster. It’s very good at finding patterns in huge amounts of chemical and production data that a human would take months to notice. It is not good at deciding what those patterns mean for a beer people actually want to drink. That part is still, for now, entirely human.