How AI Rewrote the Climate Dictionary

In late July, Meta did something that would have been unthinkable for a Silicon Valley giant a decade ago: it withdrew from RE100, the global corporate renewable energy initiative it had belonged to for ten years. The Climate Group, which runs the initiative, said Meta could no longer meet the technical criteria "due to investments made in new gas power" ^[1] — ten gas plants, among them the ones feeding its Louisiana Hyperion campus and a 200-megawatt facility in Ohio. Then Meta issued its statement.

After several in-depth conversations between Meta and Climate Group, Meta has withdrawn from the RE100 initiative, as it is no longer able to meet the technical criteria due to investments made in new gas power. — Climate Group

The goal, the company insisted, had not changed. Only the framework that measured it had. The same pattern is playing out across the three companies most exposed to the AI buildout's energy demands. In February, Microsoft paused its purchases of unbundled renewable energy certificates and carbon removal credits — the instruments that had underpinned its carbon-negative pledge — to redirect the money elsewhere.

While AI infrastructure is driving demand for energy, water, land and materials, sustainability solutions are not scaling fast enough to meet demand. — Microsoft

Microsoft's emissions had risen 25 percent, to roughly 20 million metric tons, driven by data center expansion ^[2]. The old tool — buying offsets to cancel out emissions — was being swapped for a new one: building generation. The pledge stayed. The method changed. Google's version of the maneuver is the most vivid. In April, the company partnered with Crusoe Energy on a Texas data center powered by a 933-megawatt natural gas plant projected to emit 4.5 million tons of CO₂ annually — more than the city of San Francisco's total emissions ^[3]. Google's own emissions have risen 48 percent since 2019. Its response:

We don’t have a contract in place for the plant in Texas. — Chrissy Moy

Three companies, three preserved phrases, three changed practices. The vocabulary of decarbonization — "100% clean energy," "carbon-free," "net-zero," "climate moonshots" — remains intact. What those words describe does not. The mechanism that connects all three is a two-word phrase that has quietly become the organizing principle of AI energy procurement: "firm power." It means electricity that is reliable, dispatchable, and available 24 hours a day — something intermittent renewables like solar and wind, absent storage at scale, cannot yet provide. A Meta executive made the logic explicit.

You can’t sacrifice uptime for sustainability. — Daniel Sottosanti

That sentence is the hinge. Once reliability is the non-negotiable and sustainability is the variable, the energy mix writes itself. In El Paso, Meta is funding a 366-megawatt natural gas plant. El Paso Electric explained the choice plainly.

In this case, the customer requested and is funding a natural gas resource as the fastest available option to meet a large, time-sensitive power need. — El Paso Electric

El Paso Councilman Chris Canales saw it differently.

The company has made public commitments to power their project with renewable energy sources including on-site solar generation, and I think they need to be held to those commitments. — Chris Canales

But the plant is moving forward ^[4]. The same logic has pulled nuclear power into the "firm power" frame — and with it, a semantic shift. Vistra Corp signed a 20-year agreement with Meta for 2,609 megawatts of what it calls "carbon-free" power from nuclear plants, plus a 1,200-megawatt deal at Comanche Peak ^[5]. The label "carbon-free" — once the language of wind and solar commitments — now applies to nuclear. Constellation Energy signed a 20-year deal with Microsoft to restart Three Mile Island and another with Meta for the Clinton nuclear plant in Illinois, while simultaneously providing natural gas through its Calpine unit in Texas ^[6][7]. "Carbon-free" nuclear and fossil gas are being procured side by side, under the same logic, with the same vocabulary. The nuclear renaissance underway is not being sold as a climate strategy. Trump ordered a quadrupling of U.S. nuclear capacity with an $80 billion Westinghouse partnership; the UK is investing £18 billion; the EU approved €14.2 billion for Poland's first nuclear plant; roughly 40 countries pledged to triple nuclear capacity by 2050 ^[8][9]. The frame is AI energy security, not decarbonization. Brookfield Renewable — a company built on hydroelectric, solar, and wind — is now leveraging its 50 percent stake in Westinghouse to provide nuclear power to Microsoft and Google ^[10]. Even the renewable companies are speaking the new language. The federal government has removed the one mechanism that would have made the gap between the old words and the new practices visible. In May, the SEC formally proposed rescinding its 2024 climate disclosure rules. Chairman Paul Atkins offered a rationale.

Let the Environmental Protection Agency do their job and we stick to our job. — Paul S. Atkins

The rules would have required companies to report their emissions and climate risks in standardized, auditable form — a baseline against which a pledge could be measured ^[11]. Without them, a company can say its goal "does not change" and there is no federal filing to check it against. Counter-movements are forming, but they are fighting the economic force of a buildout McKinsey estimates at $7 trillion globally by 2030 ^[12]. The UN Secretary-General launched an AI Environmental Transparency Initiative in June, calling on firms to disclose emissions and commit to 100 percent renewable energy by 2030 ^[13]. A coalition of nine organizations including the U.S. Green Building Council and the Climate Bonds Initiative formed the Greening AI Data Centers Coalition. The Climate Bonds Initiative warned of the stakes.

Trillions are going into building AI data centers, but without clear standards, it risks becoming a climate disaster. — Climate Bonds Initiative

U.S. states — Michigan, Oregon, Minnesota — are enacting emissions-free energy mandates by 2040, and New York and California are proposing legislation requiring large data centers to derive 90 percent or more of their energy from renewables ^[14]. Australia is moving to require data center operators to fund new renewable projects, with climate groups pushing for additionality rules that would bar operators from using carbon offsets ^[15]. But the companies most exposed to AI's energy demands are the ones leading the redefinition, and the economic forces behind them are immense. NextEra Energy, one of the largest U.S. utilities, has abandoned its 2045 zero-emissions target, citing data center-driven generation demands ^[16]. Energy Transfer has locked in 20-year natural gas supply deals directly tied to AI data centers, contracting over 6 billion cubic feet per day of new pipeline capacity expected to generate $25 billion in future revenue ^[17]. The most telling sign that the semantic swap is complete comes not from a tech company but from a renewable energy advocate. Aligned Climate Capital, a firm that invests in clean energy, issued a statement that adopted the new vocabulary wholesale.

Solar, wind, and batteries are the fastest, cheapest, most scalable way to add firm power—and anyone who thinks otherwise is clinging to an outdated narrative. — Aligned Climate Capital LLC

The word "firm" has won the argument. No one noticed it was contested. The debate is no longer about whether carbon reduction is the goal — it is about what counts as "firm power." Solar and batteries, the renewable camp now argues, are the best kind. Gas and nuclear, the tech giants have already decided, are the fastest. The old question — is this energy clean? — has been replaced by a new one: is it reliable? The climate pledges still use the old words. But the new question is the one being answered.