November 16, 2025
Hot takes, hotter chemistry
Open Catalyst Project
Meta + CMU bet AI can find cheap climate catalysts — comments erupt
TLDR: Meta and CMU opened huge AI datasets to hunt cheap catalysts that could make clean energy storage practical. The crowd is split between hope for open science and skepticism about Meta’s motives, with hydrogen hype, leaderboard jokes, and a big question: who benefits if a breakthrough lands.
Meta’s Fundamental AI Research (FAIR) and Carnegie Mellon just dropped the Open Catalyst Project, aiming to use AI to discover cheap chemical helpers (catalysts) that turn surplus wind/solar into fuels like hydrogen. They released massive open datasets (OC20/OC22) with results from 260 million quantum physics calculations (DFT = a super-precise physics calculator), plus code and a leaderboard to let anyone compete. Cue the comments section: chaos, hope, and memes.
The loudest split? “AI can speed climate solutions” vs “this is Meta greenwashing.” Optimists cheer the openness (“finally, climate science you can fork on GitHub”), while cynics side-eye Big Tech motives. Chem nerds are thrilled (“my laptop could never do 260M DFTs”), and gamers love the leaderboard (“climate Fortnite, let’s go”). Hydrogen sparks its own war: fans say it’s the scalable way to store clean energy; skeptics warn about leaks, pipelines, and hype. Memes fly: “AI vs atoms”, “DFT stands for Don’t Fix Twitter,” and “please don’t A/B test the atmosphere.” The spiciest practical worry? Who profits if a breakthrough catalyst emerges from this “open” effort. Still, the vibe is electric: if AI can cut the time and cost to find the right materials, this could be a rare W for the internet.
Key Points
- •FAIR at Meta and CMU’s Department of Chemical Engineering collaborate on the Open Catalyst Project to use AI for catalyst discovery.
- •Renewable energy storage needs motivate finding cost-effective catalysts, with hydrogen conversion cited as a scalable solution.
- •DFT-based quantum simulations evaluate catalysts but are computationally expensive, limiting scale.
- •AI/ML approaches aim to approximate DFT efficiently to accelerate identification of effective catalysts.
- •OC20 and OC22 datasets (1.3M relaxations, 260M DFT calculations) plus open-source baselines and a leaderboard enable community participation.