AI Is Killing Big Tech's Climate Pledges — Google +25%, Amazon +16%, Meta +64%

Google’s carbon emissions climbed 25% year-over-year. Amazon’s shot up 16%. Meta’s jumped 64% in a single year. All three companies pledged net-zero carbon emissions this decade — and all three are watching AI datacenter demand tear those pledges apart.

The annual sustainability reports released last week by Google and Amazon confirm what the AI infrastructure boom has been signalling for two years: the energy cost of frontier AI is now materially reshaping the carbon trajectory of the world’s largest technology companies. Microsoft, which has not yet released its 2026 report, posted a 23% emissions increase over its 2020 baseline last year and is expected to reveal a similar or larger spike in coming weeks.

🔍 THE BOTTOM LINE

AI is not just a compute problem — it is a carbon problem. Four companies that spent the last decade branding themselves as climate leaders are now building gas-fired power plants as fast as they can build datacenters. The net-zero pledges have not been formally abandoned. They have been quietly redefined from concrete targets into “climate moonshots” — speculative projects that may or may not materialise. The emissions numbers tell the real story.

The Numbers That Broke the Pledges

Google’s 2026 environmental report is striking for what it admits and what it doesn’t promise. Carbon emissions have risen every year since 2023. By 2025, the company stopped speaking in terms of concrete 2030 goals and instead framed its ambitions as “climate moonshots” — language Google uses internally for speculative, uncertain projects.

Amazon’s sustainability report struck a similar note: “We recognize that the path to being a more sustainable company is not a straight line. Though our emissions increased in 2025, we remain steadfast in our commitment to sustainability.” That’s corporate language for: the line went up, not down, and we don’t know when it reverses.

Microsoft’s 2025 report documented a 23% increase against its 2020 baseline. The company’s AI infrastructure spending has only expanded since, making 2026’s disclosure likely to show further growth. Meta’s 2025 report showed emissions up 64% year-over-year — the steepest climb of the four — despite a net-zero-by-2030 pledge that now looks aspirational rather than achievable.

74 Gas Plants and Australia’s Worth of Pollution

The shift from clean energy to fossil fuels is not theoretical. According to The Guardian’s reporting, all four companies have turned to fossil fuels to power AI datacenters, signing contracts for large volumes of gas-generated electricity in Texas, Indiana, and Louisiana.

The Environmental Integrity Project released a report last week documenting 74 gas-fired power plants planned for datacenters across the US. Combined, these facilities could emit upwards of 660 million tons of greenhouse gas pollution per year — roughly equivalent to the entire national emissions of Australia.

This is the structural reality behind the US datacenter recall movement we covered earlier this year: the $130 billion in stalled AI infrastructure projects are not just about noise, water, and property values. They are about the fuel source. When residents see gas turbines going up next to their schools — as they did with xAI’s unpermitted turbines in Mississippi — the climate pledge gap becomes a local air quality issue.

The ESG Era Is Over for AI Infrastructure

A decade ago, these companies fashioned themselves as climate leaders because investors prioritised ESG metrics. Wind and solar investments were cheap signalling — good for share prices, good for press releases, and achievable because datacenter demand was manageable.

That calculus has inverted. As The Guardian’s TechScape column notes, stock prices now depend on AI integration capacity, and AI needs energy that clean sources cannot yet deliver at scale. The companies are making a straightforward choice: build the compute, buy the gas, and reframe the climate promises later.

Google’s own report concedes the dilemma: “The environmental footprint of the data centers that power AI is growing, creating a dual challenge: managing that environmental footprint while simultaneously building infrastructure to meet growing demand and realize AI’s full potential.” That is the most honest sentence in all four reports — and it amounts to an admission that the two goals are in direct tension.

Meta’s Panic Pivot

The climate story sits inside a broader narrative of AI-driven strategic chaos at Meta. The company’s stock is down roughly 9% year-to-date despite the broader AI boom. It laid off about 10% of its global workforce in May, including over 2,000 workers at its Menlo Park headquarters. CEO Mark Zuckerberg conceded in a town hall that its AI agent development was going slower than expected.

Meanwhile, Bloomberg reported Meta is entering the cloud computing business to sell excess AI compute. The New York Times reported it is building a prediction market app called “Arena” to compete with Polymarket. Forrester’s VP research director Mike Proulx called it “Meta’s familiar copycat playbook” — and noted the irony of a company under litigation for addictive product design launching another habit-forming product.

The 64% emissions increase sits inside this picture: a company spending billions on AI infrastructure it hasn’t yet figured out how to monetise, with rising carbon output as a side effect of the scramble.

NZ Angle — The Renewable Advantage That Isn’t Enough

New Zealand has been pitching itself as an AI datacenter destination on the back of clean electricity and a cool climate — genuine advantages for reducing operational carbon. But the US experience shows the problem is not just where the power comes from. It is scale. The 74 planned gas plants exist because renewable capacity cannot expand fast enough to meet AI demand, even for companies with the resources of Google and Amazon.

NZ’s grid is small. A single large AI datacenter could consume a meaningful fraction of national generation capacity. The “clean energy advantage” holds only if the country builds new renewable generation in lockstep with datacenter demand — and if the companies signing the leases actually commit to using it rather than falling back on gas when the wind stops blowing.

❓ FAQ

Which company’s emissions grew the most? Meta, at 64% year-over-year, despite a net-zero-by-2030 pledge. Google’s rose 25%, Amazon’s 16%, and Microsoft’s were up 23% against a 2020 baseline.

Why can’t these companies just buy more renewable energy? They are, but AI datacenter demand is growing faster than renewable capacity can be built. All four have signed gas contracts in Texas, Indiana, and Louisiana to bridge the gap. The Environmental Integrity Project identified 74 gas plants planned specifically for US datacenters.

What does “climate moonshots” mean? Google’s term for speculative, uncertain projects. The company shifted from concrete 2030 emissions targets to “moonshots” in its 2025 reporting — a signal that firm commitments have been replaced with aspirational R&D.

Does this affect New Zealand? NZ’s clean grid is an advantage, but its small size means a single large AI datacenter could strain national generation. The renewable advantage only holds if new generation is built in lockstep with datacenter demand.

Has any company abandoned its net-zero pledge? Not formally. All four maintain their pledges but have softened the language and timelines. The emissions data — rising every year since 2023 — tells the real story.

🔍 THE BOTTOM LINE

The same companies that built their brands on climate leadership are now the biggest drivers of new fossil fuel demand in the US. The net-zero pledges haven’t been cancelled — they’ve been redefined into meaninglessness. When 74 gas plants are being built for datacenters and emissions rise 25-64% year-over-year, the pledge is the marketing and the gas plant is the reality.