Big Tech Bought 8 Million Tons of Carbon Removal in 2024. The Industry Delivered 319,000.

Four point four percent.

That is the fraction of carbon removal credits purchased in 2024 that were physically delivered as actual CO2 pulled from the atmosphere. According to CDR.fyi's 2024 year-in-review, the carbon dioxide removal industry sold forward commitments totaling nearly 8 million tonnes. It delivered 318,600 tonnes. Everything else is a promise backed by facilities that do not yet exist, technologies still scaling, and timelines measured in years.

Meanwhile, the three companies responsible for more than 75% of those purchases are adding emissions faster than the entire removal industry can subtract them.

The Numbers Nobody Puts in the Same Paragraph

Microsoft's total greenhouse gas emissions rose 23.4% from its 2020 baseline through fiscal year 2024, driven overwhelmingly by data center construction and AI infrastructure. Its own sustainability report shows Scope 3 emissions, which account for 97% of the company's total footprint, climbed 26% over the same period. In absolute terms, that is roughly 17.4 million tonnes of CO2 equivalent per year.

Google reported 14.3 million metric tons of total emissions in 2023, a 48% increase over its 2019 baseline. Scope 2 emissions from energy purchases jumped 37% in a single year. Its data centers consumed more than 24 terawatt-hours in 2023, roughly 7 to 10% of global data center electricity, and that figure rose another 27% in 2024.

Meta disclosed 8.2 million tonnes of CO2 equivalent in its 2024 sustainability report, with 99% sitting in Scope 3. It purchased approximately 50,000 tonnes of carbon removals to cover its direct operational emissions.

Combined, those three companies emitted roughly 40 million tonnes in their most recently reported fiscal years. Across all buyers, the global CDR industry delivered 318,600 tonnes.

Ratio: for every tonne the carbon removal industry physically extracted from the atmosphere in 2024, Microsoft, Google, and Meta added approximately 125 tonnes.

Purchased Is Not Removed

The vocabulary of carbon removal markets obscures this gap by design. When Microsoft announced it had signed agreements covering 45 million tonnes of carbon dioxide removal in 2025, up from 20 million in 2024 and 5 million in 2023, headline writers reported it as climate progress. Language like "commitments" and "agreements" reads like the carbon has already been captured.

It has not. Microsoft's 2025 purchase volume alone is roughly 141 times the total CDR delivered globally in 2024. Those 45 million tonnes are forward contracts for removal that will happen over years, from facilities that are under construction, in permitting, or on drawing boards. Phil Goodman, director of Microsoft's carbon removal portfolio, described the purpose plainly: "By securing that forward demand commitment, suppliers can actually go raise financing, hire staff, and build out their projects."

A defensible industrial policy argument. Also an admission that the tonnes do not exist yet.

Climeworks operates the world's largest direct air capture facility, Mammoth, in Iceland. Its nameplate capacity is 36,000 tonnes per year. Net removal after accounting for energy inputs and operational emissions is lower. One Mammoth, running at full capacity for a year, removes roughly what Microsoft adds to the atmosphere every 18 hours.

The Growth Math

CDR.fyi reports that deliveries grew 120% year-over-year in 2024, from approximately 145,000 tonnes in 2023 to 318,600 tonnes. That is impressive growth off a tiny base. Run the exponential forward and you get something like this:

Year	Projected CDR Delivery (tonnes)	AI Data Center Emissions (Goldman Sachs trajectory, million tonnes)
2024	318,600	~80
2025	~700,000	~108
2026	~1.5 million	~136
2027	~3.4 million	~164
2028	~7.4 million	~192
2029	~16.3 million	~210
2030	~36 million	~220+

The delivery column assumes CDR deliveries sustain 120% annual growth, which is optimistic. Physical infrastructure does not compound like software downloads. Each new DAC plant requires site selection, permitting, grid interconnection, water access, and geological storage characterization. Climeworks spent years getting Mammoth from announcement to operation. Enhanced rock weathering requires mining, crushing, transporting, and spreading millions of tonnes of basalt. Biochar requires feedstock logistics and pyrolysis capacity. None of these are digital products.

The emissions column uses Goldman Sachs Research's August 2025 analysis, which projects that 60% of incremental data center electricity demand will be met by fossil fuels, adding approximately 220 million tonnes of CO2 by 2030. The IEA's April 2025 report projects global data center electricity demand will more than double to roughly 945 terawatt-hours by 2030, which is slightly more than Japan's total electricity consumption.

Even under the most optimistic CDR growth scenario, the removal industry covers approximately 16% of the AI data center emissions increment by 2030. It does not touch the other 84%. And that 16% assumes perfect exponential scaling with no construction delays, no permitting bottlenecks, no energy price spikes, and no supply chain constraints on sorbents, basalt, or biochar feedstock.

The $/Ton Reality

To remove 40 million tonnes annually at current CDR costs would require $12 billion to $24 billion per year, depending on the mix. Direct air capture runs $300 to $600 per tonne at leading facilities. Enhanced rock weathering sits at $50 to $150 per tonne. Biochar costs $100 to $300 per tonne. BCC Research estimates the total global CCUS market at $3.4 billion in 2024, growing to $9.6 billion by 2029.

For context, Microsoft, Google, and Meta collectively spent more than $200 billion on capital expenditures in their most recent fiscal years, much of it on data centers and AI infrastructure. Removing the carbon those data centers generate would cost roughly 6 to 12% of the capital budget creating the emissions. That is not a prohibitive number. But it requires physical carbon removal infrastructure that operates at a scale roughly 125 times larger than what exists today.

The Buyer Concentration Problem

CDR.fyi's data reveals a structural fragility in the removal market that goes beyond the delivery gap. More than 75% of all carbon removal purchases in 2024 came from Microsoft, Google, and the Frontier consortium (which includes Stripe, Alphabet, Meta, Shopify, and McKinsey). Unique purchasers grew just 7%. First-time buyers declined 18%.

This means the CDR industry is economically dependent on three buyers who are also the largest emitters driving the need for removal. If any of them scaled back purchases due to a recession, an earnings miss, or a shift in corporate priorities, the supply side collapses. Equity investment in CDR companies already fell 30% in 2024, to $836 million. Only 36% of suppliers listed on CDR.fyi have made a single sale.

IPCC modeling says the world needs to remove 5 to 16 billion tonnes per year by 2050 to limit warming to 1.5 degrees Celsius. At 318,600 tonnes, the 2024 delivery figure is 0.002% of the low end of that target. Closing that gap requires not just scaling the technology but creating a buyer base that extends far beyond three tech companies and their subsidiaries.

The Strongest Counterargument

The best case for Big Tech's carbon removal strategy is that forward commitments are how you build an industry, and criticizing the delivery gap misunderstands what stage this market is in. Climeworks did not build Mammoth because atmospheric chemistry demanded it. Climeworks built Mammoth because Microsoft, Stripe, and other advance-market-commitment buyers guaranteed demand, which unlocked financing, which enabled construction. The 4.4% delivery rate is not failure; it is the normal pattern of an industry where capital must flow before capacity exists.

This argument has merit. Advance market commitments have worked before. The pneumococcal vaccine AMC, launched in 2009, guaranteed purchase prices for vaccines that did not yet exist, and the resulting supply ultimately protected more than 150 million children. Microsoft explicitly models its CDR procurement on this precedent.

But the vaccine AMC succeeded because the gap between commitment and delivery was measured in months to a few years, the unit economics improved with standard pharmaceutical learning curves, and the number of buyers expanded rapidly through GAVI and national health systems. Carbon removal faces a gap measured in decades, unit economics that improve slowly because they are constrained by thermodynamics and physical logistics, and a buyer base that is shrinking, not growing. Forward demand signals are necessary. They are not sufficient.

Limitations

This analysis uses the most recently published emissions data from each company, which covers different fiscal years (Microsoft FY24, Google calendar 2023, Meta calendar 2024). Year-to-year comparisons across companies are approximate. CDR.fyi tracks only the voluntary carbon removal market; government-funded projects and compliance markets are not included. The 120% growth projection for CDR delivery is an extrapolation of a single year's performance and should not be treated as a forecast. Goldman Sachs' 220 million tonne estimate for AI-related emissions depends on assumptions about the pace of data center construction and the carbon intensity of grid electricity, both of which vary by region. A Guardian analysis found that actual emissions from Google, Microsoft, Meta, and Apple facilities may be 7.62 times higher than officially reported, which would make these ratios conservative.

The Bottom Line

Carbon removal is real, necessary, and scaling. It is also three orders of magnitude short of what the atmosphere requires and 96% short of what the market has already sold. The companies buying the most removal credits are the same companies whose AI ambitions are accelerating the need for removal. That is not hypocrisy; it is the structural condition of an industry where the cure and the disease share a balance sheet. But spreadsheets full of forward commitments do not cool the planet. Operational tonnes do. And the planet is keeping score in molecules, not purchase orders.