AI data centers become carbon bombs: annual greenhouse gas emissions exceed entire countries, OpenAI, Meta, Microsoft... decarbonization pledges collapse

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1745 words

New natural gas projects for 11 data center campuses in the U.S. have the potential to emit over 129 million tons of greenhouse gases annually, exceeding Morocco's total emissions for 2024; even if actual emissions are halved, they would still surpass Norway's national emissions. The decarbonization commitments of tech companies are being rapidly eroded by the AI compute arms race. (Context: Trump forces Big Tech to sign "cost-sharing for electricity price hikes"! AI data centers consume 12% of U.S. power, voter anger becomes a catalyst) (Background: Can the Earth's power supply feed AI? Analyzing SpaceX's acquisition of xAI to build a "space compute empire") 1 data center campus, 129 million tons of greenhouse gases. This is the draft climate bill for the AI compute expansion expected by 2026, and this only covers some of the natural gas projects currently under construction or permitted in the U.S. According to permit filings obtained by WIRED, these new natural gas projects powering AI companies such as OpenAI, Meta, Microsoft, and xAI have an annual emission potential that already exceeds Morocco's total greenhouse gas emissions for 2024. Even by the industry's most optimistic estimates, if actual emissions are cut in half, the greenhouse gases released would still exceed Norway's total national emissions for 2024. Note: Although burning natural gas produces about 40% to 50% less CO2 than burning coal, the massive scale of combustion still results in a significant impact. Behind this bill, a specific infrastructure model is spreading rapidly: BTM (behind-the-meter), where data centers bypass the public power grid to build their own natural gas power plants. At the beginning of 2024, the total BTM natural gas capacity developed specifically for data centers in the U.S. was only 4GW. By early 2026, this figure is approaching 100GW, showing exponential growth. The non-profit organization Global Energy Monitor documented this shift in a research report this January. The driving force is straightforward: AI companies cannot wait for the connection schedules of traditional power grids (which often take years), nor do they want their own electricity demand to trigger public backlash over rising electricity prices. By building their own power plants, they bypass regulations and the grid, leaving the emission problem confined to the numerical fields of air permit application files. Michael Thomas, founder of the clean energy research firm Cleanview, has been tracking data center natural gas permit applications for a long time. He describes the rapid spread of BTM as a "frenzied acceleration of emissions." What worries him even more is that high-efficiency gas turbines are in short supply globally (partly due to the data center race itself), forcing some developers to turn to less efficient models, which generate more emissions for the same amount of power. Energy researcher Jon Koomey also pointed out that data center loads are almost constant throughout the day, unlike grid power plants that have room for peak-valley regulation, making actual emissions more likely to approach the upper limits listed in permit files. The Stargate project led by OpenAI, along with its three related natural gas projects in Abilene, Texas, and Project Jupiter in New Mexico, has a combined annual emission potential of over 24 million tons. The turbines for xAI's Colossus 1 (Memphis) and Colossus 2 (Southaven, Mississippi) each have an annual emission potential of over 6.4 million tons per campus, equivalent to 30 medium-sized natural gas power plants operating simultaneously, enough to power 1.5 million households. Even larger is Fermi in Amarillo, Texas, whose two natural gas projects have an annual emission potential of over 40.3 million tons, higher than the annual emissions from all electricity sources in Connecticut. Pacifico Energy GW Ranch near Fort Stockton has an annual emission permit for a single project exceeding 33 million tons, claiming to be the largest single energy project in the U.S. As for Microsoft, its Chevron-supported natural gas project in West Texas has an annual emission potential of over 11.5 million tons, exceeding the annual emissions of Jamaica. Meta's three BTM projects in Ohio have a combined annual emission potential of 5.5 million tons, a figure equivalent to more than 10% of the 23.8 million tons of emission reductions Meta has publicly claimed over the past four years. There is a structural contradiction between the figures in these permit files and the sustainability commitments publicly made by tech companies. These companies have several standard responses: Williams Companies: The permit figures are "conservative theoretical upper limits, not actual forecasts." Microsoft: The company is "adopting a diversified energy portfolio approach." Fermi: Natural gas is a "bridge

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching6 items

2026-04-24

Jensen Huang sends internal memo: 10,000 NVIDIA employees mandated to switch to OpenAI Codex, GPT-5.5 to run on GB200 chips

Similarity 120%關鍵字 openai同分類 zh

2026-04-24

Jensen Huang sends all-hands email embracing OpenAI Codex: over 10,000 NVIDIA employees are already using it, with GPT-5.5 running on GB200

Similarity 120%關鍵字 openai同分類 zh

2026-04-23

Naval, author of The Almanack of Naval Ravikant, launches AI interval fund USVC, allowing participation in early-stage investments in xAI and OpenAI with as little as $500

Similarity 120%關鍵字 openai同分類 zh

2026-04-23