One potential consequence of the boom in data centers: soaring demand for carbon removal.
That’s because the emissions resulting from the rapid, generative artificial intelligence-fueled build-out of data centers is projected to be enormous. According to an analysis that Morgan Stanley provided to its clients, the sector’s boom is expected to emit 2.5 billion metric tons of carbon dioxide equivalent gasses by 2030.
A scenario anticipating a 180% growth of data centers in the United States found it would amount to an emissions increase of 200 million metric tons per year by 2030, respectively. Over half of the anticipated build-out is expected to be in the U.S. On a global scale, a 200% growth is expected to yield a 400-million-metric-ton increase.
“The hyperscalers will own a portion of these emissions in their value chains,” the analysis found. “And while some of these emissions may be mitigated or unrealized if the hyperscalers deploy CCUS, green materials, or other carbon efficiency solutions, we think that their own emissions estimates under-estimate the magnitude of increase due to this expansion.”
The expected emissions are largely — though not exclusively — the result of the enormous energy demand of generative AI processes, which recently has become the primary bottleneck for AI’s growth; just a few years ago, chips were the main source of concern. The construction of the data centers, though, will also result in emissions from cement, steel, and other elements of the infrastructure.
In order to avoid that potentially grim climate toll, that carbon will need to go somewhere. And for that reason, Morgan Stanley is anticipating that the AI boom could indirectly boost the still-nascent carbon removal industry.
The report characterizes carbon removal as a “last resort with substantial growth potential” for addressing the emissions impact of data centers, especially given the enthusiasm of major hyperscalers like Microsoft, Google parent Alphabet, Amazon, Meta, and Apple for CDR technologies. The analysis found that the removal credits that these tech companies use each year “will increase dramatically,” or by at least 20 times, before the end of the decade.
The current announced capacity in the tech-based CDR approaches currently totals roughly 30 million metric tons per year, which mainly comes in the form of bioenergy with carbon capture and storage. Other pathways, including direct air capture, are significantly further from scale.
Morgan Stanley found that the hyperscalers have already begun to purchase CDR credits; between Amazon, Microsoft, and the advance market commitment Frontier (of which Alphabet and Apple are both members), they have contracted at least 18 million metric tons for delivery in the next decade. But it’s likely that substantially more purchases are to come.
“After 2030, we expect the hyperscalers to continue to strive to be carbon-neutral every year and to keep buying CDRs,” the report notes. “Their absolute emissions should continue to fall [year-over-year] as investments in decarbonization solutions materialize but we do not expect them to be able to achieve absolute zero.”
Offsetting the outstanding emissions, the analysis found, would likely involve a nearly six-fold increase in their annual use of removal credits as compared with their most recent fiscal years, given their current emissions goals.
But those goals were largely set before the build-out of generative AI infrastructure, which began in earnest in 2022; accordingly, Morgan Stanley expects that “emissions from these companies will grow more than was expected between now and 2030, and some of that shortfall will need to be neutralized with [carbon removal credits].”
This is just the latest emerging technology to be floated as a fix — or at least a Band-Aid — for the looming problem of AI energy demand. The U.S. colocation data center market has doubled in size over the last four years. And as capacity grows, it’s prompting broader questions about whether the boom in data centers can speed up the maturity curves for technologies that aren’t yet being deployed at scale.
For instance, Meta is also betting on geothermal; the Facebook parent company last week announced it would buy 150 megawatts of geothermal baseload power for its data center infrastructure in 2027 from the Houston-based advanced geothermal company Sage Geosystems. This follows another geothermal deal in June between Google and Fervo Energy, which involved a new rate structure dubbed the “clean transition tariff.”
Meanwhile, Microsoft and other players have an eye on green hydrogen, though there’s some skepticism from experts in the field about whether data centers are a viable use case of the fuel, especially while supply is relatively scarce.
However, it is not only clean baseload power that is seeing growth. The company Enchanted Rock is selling utilities on the use of onsite natural gas-powered microgrids to power data centers.
