Every human genome carries millions of small differences, single-letter changes in the code of DNA, and for most of them no one knows what they do. On September 9, Google DeepMind released a tool meant to change that: AlphaGenome Atlas, a catalog of predicted effects for roughly nine billion single-base changes in the human genome.
The company called it the most comprehensive map yet of the molecular consequences of genetic variation. Instead of reading a person’s DNA and flagging only the handful of variants already known to cause disease, the tool aims to score the likely effect of nearly every possible single-letter change, giving researchers a far wider view of what a given genome might mean.
The announcement sits inside a broader scientific push to turn AI from a text and image tool into a laboratory instrument. DeepMind has already produced AlphaFold, which predicts protein structures and won its creators scientific acclaim. AlphaGenome Atlas extends the same idea to the genome itself, where the raw data is vast and the interpretations are scarce.
The atlas builds on work the lab began with protein structure. Where AlphaFold predicts the shape a protein folds into, the new tool addresses the step before: whether a change in the DNA that encodes that protein will matter at all. Together they form a pipeline from variant to structure to function.
The practical promise is in medicine. Doctors and researchers hunting for the cause of a disease often face a long list of candidate variants with no clear culprit. A map that predicts the effect of each one could shrink that list, pointing faster to the mutation that matters and the one that is harmless noise.
But the release also arrived in a week when Google was managing a different kind of consequence: its own. On the same day, the company announced a four-year agreement with the Indian climate-tech firm Mitti Labs to buy one million tons of carbon credits for reducing methane from rice farming by 2030.
The credits will cover roughly 100,000 hectares of rice paddies across the states of Karnataka, Andhra Pradesh and Telangana, and Google described the deal as the largest publicly announced rice-methane carbon-credit purchase to date. Methane from flooded rice fields is a significant source of greenhouse gas, and the credits pay farmers to manage their fields in ways that cut those emissions.
The motive is written in Google’s own numbers. The company’s greenhouse-gas emissions rose 18 percent in 2025 from the prior year, driven largely by the data centers that power its AI ambitions. Carbon credits have become one of the ways the company tries to offset a footprint that keeps growing even as it promises to reach net zero.
Skeptics have long questioned whether carbon credits represent real reductions or convenient accounting. The rice-methane program is an attempt to answer that by paying for measurable changes on the ground, though the market’s critics say credits let companies keep emitting while claiming progress.
The third piece of Google’s week was quieter and pointed in a different direction. The company’s online store quietly removed the Pixel Tablet, the device it launched in 2023, and reports said the planned successor, a Pixel Tablet 2 built on the Tensor G4 chip, had been canceled.
The tablet’s retreat is a small story next to the others, but it says something about where Google is placing its bets. The company has pulled back from hardware categories that do not support its AI narrative, and a tablet line that never found a large audience was a natural place to cut.
Taken together, the three items sketch a company straining to balance its ambitions. On one side, DeepMind publishes tools that promise to advance biology and medicine. On the other, the company’s own operations produce the emissions it must then pay to offset, and its consumer hardware strategy narrows to the products that fit.
For scientists, AlphaGenome Atlas is a genuine addition, whatever Google’s own footprint. The data, freely released, gives the research community a new reference point, and the questions it will answer, about which variants matter and why, are among the most important in biology.
The scale is what makes it notable. Nine billion predictions is more than the field has ever had in one place, and the ability to query any single-base change against it could change how quickly geneticists move from raw sequence to testable hypothesis. The tool will be judged not on its size but on whether those predictions hold up in the lab.
Google’s week was, in the end, a study in the gap between aspiration and operations. The company can map nine billion variants and sign the largest rice-methane credit deal on record in the same breath that its emissions rise and its tablet quietly disappears. Both things are true, and both are now part of the record.
