Google Is Paying Farmers to Change How They Grow Rice. Here’s the Real Bet Behind It.
How Google Rice Carbon Credits Work
Google rice carbon credits are built around an unusual idea: changing how farmers flood their fields could reduce methane while helping remove carbon from the atmosphere.
That unglamorous fact is the starting point for the largest carbon-removal deal Google has signed to date. On September 16, Reuters reported that the company would back a project covering more than 200,000 hectares of rice farmland in Rio Grande do Sul, Brazil’s rice belt, developed by a climate startup called Terradot. The project asks the same land to do two different jobs: change how farmers flood their fields to cut methane, and spread crushed volcanic rock across those same fields to pull carbon dioxide out of the air. Google is committing to buy one million tons of methane-abatement credits by 2030 and a further one million tons of carbon-removal credits by 2040.
The pairing sounds tidy, almost too tidy, as if one field could quietly solve two different problems at once. It also lands at an inconvenient moment for Google. The deal follows a year in which the company’s own greenhouse gas emissions rose faster than they ever have, driven largely by the AI hardware it is racing to build. Whether credits like these can get cheap and dependable enough to actually catch up to that growth is the real question this deal is testing, not just for Google, but for every technology company making similar promises.
Inside the deal
Terradot is not a household name, but Google has been betting on it for two years. The company took an equity stake in the Stanford-incubated startup in 2024, the same year it agreed to buy 200,000 tons of carbon removal for delivery in the early 2030s, at the time Google’s largest single purchase of carbon dioxide removal. The new Brazilian project dwarfs that earlier deal. Combined, the methane and carbon-removal credits add up to two million tons of climate impact, and the carbon-removal half alone is roughly ten times the size of any enhanced-rock-weathering project built before it, according to Reuters.
Randy Spock, who leads carbon credits and removals at Google, told Reuters the company sees the project as a template rather than a one-off: the first attempt, at this scale, to pair a fast-acting methane cut with a slower carbon-removal process in a single portfolio. The logic behind that pairing is fairly plain once it’s spelled out. Methane warms the atmosphere quickly and mostly clears out of it within a decade; carbon dioxide lingers for centuries. Spending everything on one timescale leaves the other problem to keep compounding in the background.
The scale is meant to be seen. Rio Grande do Sul is Brazil’s main rice-growing state, and the project’s 200,000-plus hectares work out to roughly a fifth of all the rice grown there. Google and Terradot have said that if the model holds up, it could eventually extend across Brazil’s 1.5 million hectares of rice fields and be tested in other major rice-producing countries, including India and Vietnam. Terradot plans pilot projects in several countries this year, with commercial-scale expansion targeted for 2028, a timeline that turns this one Brazilian project into something closer to a proof of concept for an entire industry.
How a flooded field becomes a climate tool
The methane side of the project relies on a technique called alternate wetting and drying, or AWD. Instead of keeping a paddy continuously flooded, the way rice has traditionally been grown, farmers let the water level drop on a schedule before reflooding it. The dry interval lets oxygen back into the soil, which slows the methane-producing microbes that thrive in waterlogged conditions, a mechanism straightforward enough to explain in a sentence and considerably harder to standardize across a quarter-million hectares of farmland with different soils and different rice varieties.
The carbon-removal side works through enhanced rock weathering, or ERW. Crews crush silicate rock, typically basalt, into a fine powder and spread it across the same fields. When rain hits the powder, it sets off a chemical reaction that pulls carbon dioxide out of the air and converts it into bicarbonate, some of which eventually washes into waterways and is stored, in effect, for good. It is the same weathering that happens to any exposed rock on the planet, just accelerated: ERW grinds the rock small enough, and spreads it wide enough, to compress a geological process into a human timescale.
The two credit types don’t run on the same clock, and that gap looks deliberate rather than accidental. Terradot plans to wait years before issuing removal credits from the weathering side, which is why the carbon-removal date is 2040 rather than something sooner, specifically to measure the effect less frequently and with more confidence, lowering both the uncertainty and the cost of verification. The methane credits, tracked through a more established agricultural monitoring process, can move on the faster 2030 timeline. Google describes the combination as a portfolio built to manage warming across two different timeframes. That framing also happens to describe, quite conveniently, a deal in which the two halves are, in practice, running years apart.
The price still isn’t where it needs to be
Neither company disclosed what Google is paying for the new credits. But the last publicly known price point is instructive: a 2024 purchase of 90,000 tons of Terradot’s carbon removal, arranged through the buyer coalition Frontier, worked out to roughly $300 a ton, some way above the $100-a-ton threshold that much of the industry treats as the point where demand really opens up. James Kanoff, Terradot’s chief executive, told Reuters the new project would be cheaper than the company’s earlier deals and, while still short of that mark, represented “a big step toward that.”
Context is useful here. Direct air capture, which uses industrial plants to chemically scrub carbon dioxide from ambient air, still runs in the neighborhood of $500 a ton. Older, land-based offsets, paying to plant or protect trees, can cost as little as $5 to $20 a ton, but that low price has come with a credibility problem: several of the largest nature-based offset programs have been found, on closer analysis, to overstate how much carbon they actually kept out of the atmosphere. Terradot is positioning itself in the gap between those two extremes, pricier than a cheap and contested offset, cheaper than an industrial plant, betting that a rigorous enough measurement process is worth the premium.
One reason the price hasn’t fallen further is the sheer cost of proving the work happened. Terradot has said that third-party verification alone can add more than $100 a ton to enhanced-rock-weathering costs. Stretching out the timeline before credits are issued is, in part, a way to measure less often and still trust the number, which lowers the verification bill even as it lengthens how long a buyer has to wait to find out whether the credit it paid for actually held up.
What makes a flooded field so hard to measure
Direct air capture has one underrated advantage: the carbon dioxide moves through a machine, on a schedule, where it can be measured as it flows through a pipe. A rock spread across 200,000 hectares of working farmland does not cooperate the same way. Some of the carbon dioxide it absorbs does turn into stable bicarbonate. Some of it leaves the field dissolved in runoff water, some sits in the soil, and some can be released back into the atmosphere by soil microbes. Separating the rock’s actual effect from ordinary geochemical noise, the background variation in soil chemistry that exists with or without any treatment, is a genuine and unsolved measurement problem, not a paperwork one.
The methane side carries its own asterisk. A global meta-analysis of alternate wetting and drying found that, on average, the technique does cut methane emissions and water use, but it also tends to slightly reduce yield and increase emissions of nitrous oxide, a greenhouse gas with an outsized warming effect of its own. The results vary with how aggressively fields are dried, the local soil type, and the rice variety planted, which means an emissions factor calculated in one region can’t simply be applied to Rio Grande do Sul and trusted at face value. As of the announcement, neither Google nor Terradot had named the specific registry or crediting protocol that will certify the methane side of the project, a detail that, for a deal being pitched as a template for the rest of the world’s rice belt, will matter as much as the tonnage does.
Why Google needs the math to work
The deal did not happen in a vacuum. Google’s 2026 environmental report, published in June, disclosed the steepest one-year jump in greenhouse gas emissions the company has ever reported: up 18%, driven largely by the manufacturing of AI chips and servers. Electricity demand climbed 37% that same year, itself an acceleration from a 27% rise the year before and roughly three-and-a-half times higher than in 2019. Water use, mostly for cooling data centers, rose 34% to 10.9 billion gallons. Google has said publicly that it does not want AI’s growth to become a reason to loosen its environmental standards. The trend lines suggest that promise gets harder to keep with every new model the company trains.
That backdrop is why a rice paddy in southern Brazil matters to a company that makes its money in search advertising and cloud computing. Big technology companies are under growing pressure to show that their clean-energy and carbon-removal spending can keep pace with the emissions their data centers are adding. The Intergovernmental Panel on Climate Change has said the world will need carbon removal, alongside deep cuts to emissions, to hit its climate targets. Critics of the practice make a fair counterpoint: buying removal credits, especially ones that won’t be verified for years, can look a great deal like paying now for permission to keep polluting later.
Both things can be true at once.
The technology can be a genuinely necessary tool, and it can also be convenient cover for a company whose core business keeps adding to the very problem the tool is meant to solve.
Nothing about this deal will be settled soon. The methane credits are due by 2030, the carbon-removal credits by 2040, and the actual price Google is paying likely won’t become public before either date arrives. What’s being tested in the meantime is a bet about direction: that a technology priced at roughly $300 a ton in 2024 keeps sliding toward $100, the way costs for other climate technologies have occasionally fallen with scale, rather than stalling out the way plenty of promising climate technologies have instead.
Back in Rio Grande do Sul, the actual work is unglamorous and slow: draining a field a little earlier than usual, spreading a layer of crushed rock, waiting for rain to do its chemistry. None of it resembles the thing it’s meant to offset, a data center somewhere pulling gigawatts to train the next model. That gap is, in the end, the wager Google is making: that some of the receipts for one of AI’s largest and least visible costs can come from a rice farmer’s field.