The number that changes everything: 31 to 43 TWh in one year
Google’s latest environmental report contains a figure that reframes the entire conversation about Big Tech and climate accountability. The company’s electricity consumption climbed from 31 terawatt hours in 2024 to 43 terawatt hours in 2025 — a 39% surge in a single reporting period. No previous year-on-year increase in Google’s history comes close.
To understand the scale, consider the comparison: 43 TWh is roughly equivalent to the total annual electricity consumption of a mid-sized European country. One corporation. One year. That is not a data center expansion story. That is a structural transformation in how much energy a single tech company demands from the global grid.
The historical pattern makes the headline number even harder to dismiss. Analysts tracking Google’s energy consumption back to 2011 describe the current trajectory as exponential, not incremental. Previous increases — even the significant ones that accompanied Google’s earlier cloud and data center buildouts — look flat against the slope of the most recent reporting period. The growth curve does not suggest a one-time infrastructure investment followed by stabilization. It suggests acceleration.
The timing is not coincidental. Google has spent the past two years embedding generative AI features across Search, Workspace, and its cloud services platform. Every AI-powered query, every Gemini interaction, every large language model inference request carries an energy cost that dwarfs a standard web search by orders of magnitude. The cumulative demand from those billions of daily interactions is now visible in the electricity data in a way that corporate sustainability language cannot paper over.
Google’s own carbon emissions reporting tells the same story from a different angle. The company abandoned its previous net-zero trajectory — emissions have risen sharply since it began scaling AI infrastructure. The gap between the climate commitments Google made and the energy reality its AI expansion has created is no longer a matter of interpretation. The numbers measure it directly: 31 TWh to 43 TWh in twelve months, with no credible signal that the curve bends down from here.
AI is the accelerant: why generative AI makes this worse than the data-centre booms of the past
Generative AI is not a more efficient version of what came before. It is a categorically different energy beast. A standard Google Search query processes a handful of ranking signals and returns results using well-optimised, decade-old infrastructure. A single Gemini interaction — generating a multi-paragraph summary, drafting an email in Workspace, or powering an AI Overview in Search — requires running billions of parameters through layers of matrix multiplication, demanding orders of magnitude more compute per interaction than anything in Google’s previous product history.
That distinction matters enormously when you scale it to Google’s user base. Gemini is no longer a standalone chatbot people visit occasionally. Google has embedded it directly into Search, Gmail, Docs, Sheets, and Android — products used by billions of people every day. What were once lightweight, low-energy interactions now carry a significant computational overhead, baked invisibly into the user experience. Every AI Overview loaded in Search, every Smart Reply suggested in Gmail, every prompt processed in Workspace adds to a running energy tab that compounds across billions of daily sessions.
The 12 TWh spike in Google’s electricity consumption — from 31 TWh in 2024 to 43 TWh in 2025 — reflects exactly this dynamic. Previous data centre expansion cycles, driven by video streaming growth or cloud storage demand, produced gradual consumption curves. Generative AI inference at Google’s scale produces a vertical one.
Tech journalism has largely missed this. Coverage of Google’s AI rollout fixates on capability comparisons between Gemini and GPT-4, benchmark scores, and competitive positioning against OpenAI and Microsoft. Almost no mainstream reporting tracks the energy cost of deploying large language models at planetary scale. The carbon footprint of AI infrastructure, the electricity demand of machine learning inference, and the real-world environmental impact of AI product expansion remain afterthoughts — buried in annual sustainability reports that most readers never open. Google is counting on that inattention. The numbers in those reports tell a story the press releases don’t.
The broken promise: how this demolishes Google’s 2030 net-zero target
Google’s 2030 climate commitments looked ambitious when the company made them. They look like fiction now.
Google had pledged to operate on 24/7 carbon-free energy by 2030 — meaning every kilowatt-hour consumed matched, in real time, by clean power generation — and to reach net-zero emissions across its entire value chain by the same date. Those targets were already under strain before the latest environmental report landed. A single-year jump from 31 TWh to 43 TWh of electricity consumption has effectively shredded whatever credibility remained.
The arithmetic is brutal. A 39% annual increase in electricity demand turns renewable energy procurement into a race against a target that keeps moving faster than anyone can chase it. Google has spent years signing power purchase agreements and accumulating renewable energy certificates, and the company has genuinely led the tech sector in clean-energy investment. None of that changes what the numbers show: the volume of electricity now requiring clean-power coverage is expanding at a pace that outstrips even aggressive procurement strategies.
Most coverage of Google’s carbon footprint glosses over a critical distinction. Matching annual electricity consumption with renewable energy certificates — buying credits that represent clean power generated somewhere on the grid — is not the same thing as actually running on carbon-free electricity hour by hour. Google’s 24/7 carbon-free energy goal was designed specifically to close that gap, requiring clean supply to align with demand at every hour of every day, in every region where data centers operate. That standard becomes exponentially harder to meet when consumption grows nearly 40% in twelve months. The overnight demand spikes from AI inference workloads don’t pause while solar panels wait for sunrise.
Google’s greenhouse gas emissions have climbed 48% since 2019. The company acknowledges AI is a primary driver. Reaching net-zero across its full value chain — including Scope 3 emissions from suppliers and infrastructure — by 2030 now requires a clean-energy buildout that would need to run parallel to one of the fastest demand surges any corporation has ever recorded. The 2030 deadline is five years away. The electricity consumption curve is pointing straight up.
Google vs. Microsoft: who is really ahead, and why the comparison misleads
Google’s jump to 43 TWh of annual electricity consumption — up from 31 TWh the previous year — flips a narrative that had largely cast Microsoft as the hyperscaler with the most reckless energy appetite. Microsoft’s deep integration with OpenAI and its aggressive Azure expansion had made it the default villain in most coverage of Big Tech’s carbon footprint. Google has now overtaken it in absolute consumption terms, and by a margin that is difficult to explain away with accounting differences or reporting methodology.
The head-to-head framing is seductive but misleading. Treating Google versus Microsoft as a corporate energy race suggests that one company’s relative restraint would constitute meaningful climate progress. It would not. Both companies are locked into datacenter expansion cycles driven by generative AI infrastructure buildout — training runs, inference at scale, and the power-hungry cooling systems that keep it all from melting. Their electricity demand trajectories are steep and parallel, not divergent.
What the comparison quietly buries is the additive nature of the problem. Every major hyperscaler — Google, Microsoft, Amazon Web Services, Meta — is posting record power consumption figures. These numbers do not cancel each other out. They stack. The strain on grid infrastructure is cumulative. The competition for renewable energy capacity, long-term power purchase agreements, and clean energy certificates tightens every time another company announces a new gigawatt-scale datacenter campus. When Google absorbs more clean energy supply to offset its emissions, less of that supply is available to decarbonize the rest of the economy.
Framing this as a two-company rivalry also lets the rest of the industry operate in the background without scrutiny. The real story is a sector-wide acceleration in electricity demand driven by AI workloads that were not factored into any of the net-zero pledges these companies made just a few years ago. Google’s 43 TWh figure is a headline. The systemic pressure every hyperscaler is now placing on power grids and clean-energy supply chains is the actual crisis.
The missing accountability layer: what regulators, investors, and users should be demanding
Google published the number that matters — 43 terawatt hours of electricity consumed in 2025, up from 31 TWh the year before — inside a sustainability report that most investors, regulators, and users will never read. That is not a disclosure system. That is a filing cabinet.
No mandatory framework currently requires Big Tech companies to break out AI-specific energy consumption in real time, or even annually in a standardised format. The SEC’s climate disclosure rules remain contested. The EU’s Corporate Sustainability Reporting Directive is still bedding in. In the gap, companies self-report on their own schedules, in their own formats, with their own chosen metrics. Google’s 38% single-year electricity surge qualified, under this system, as routine environmental reporting.
Investors pricing climate transition risk into Alphabet’s stock are working from that same lagging, self-selected data. The divergence between Google’s published net-zero commitments and its actual consumption trajectory is a material financial risk — the kind that belongs in analyst models and shareholder resolutions, not buried in PDF appendices. Asset managers who absorbed the ESG commitments without stress-testing the underlying energy math now hold exposure they may not have fully priced.
The question for everyday users is blunter. Google has bundled generative AI features into Search, Gmail, Maps, and Workspace by default. Users did not opt into AI Overviews. They did not choose to have Gemini embedded across their productivity tools. Each of those interactions draws on the data centre infrastructure driving that 43 TWh figure. The environmental cost of Google’s AI rollout is being distributed across billions of users who were never asked whether they wanted to participate. Default-on deployment is a design choice with grid consequences, and it is one the company made unilaterally.
What accountability looks like in practice: mandatory, disaggregated AI energy reporting tied to product lines; independent verification of renewable energy matching claims; and opt-in rather than opt-out defaults for AI features until the carbon cost per query is publicly disclosed. None of those things exist yet. The data is alarming. The infrastructure to act on it is not.
What comes next: why the curve is unlikely to flatten
Google has already announced billions in new data centre construction across the United States, Europe, and Asia. Gemini is being embedded deeper into Search, Workspace, and Android — products used by billions of people daily. Each expansion compounds the electricity problem. If the jump from 31 TWh to 43 TWh represents a single year’s growth, the 2026 climate report will almost certainly show consumption climbing into territory that makes the current numbers look modest.
The hardware optimisation argument gets raised here, reliably. Google’s custom TPU chips do deliver real improvements in performance-per-watt. Efficiency gains are genuine. The problem is Jevons Paradox: when a technology becomes more efficient, it becomes cheaper to run, which drives higher usage, which consumes more total energy than the efficiency savings recovered. Google is not building more efficient AI infrastructure instead of more infrastructure. It is building both, simultaneously, at scale. The net result is the 39 percent electricity surge already on the books.
The harder question — the one most coverage sidesteps — is structural. What market or regulatory mechanism would actually force Google to slow down? Carbon offset markets are voluntary and widely criticised as unreliable. Renewable energy procurement commitments address the source of electricity, not the volume of it. There is no carbon price in the United States that makes energy-intensive AI training or inference economically painful. There is no regulatory body requiring Big Tech to cap its data centre energy footprint. Google’s competitors — Microsoft, Amazon, Meta — are expanding at comparable or faster rates, which means any unilateral restraint would simply hand market share to a rival with identical emissions problems.
Without a structural penalty — a carbon tax with real teeth, mandatory emissions disclosure tied to AI workloads specifically, or enforceable caps on data centre power draw — Google faces no incentive to flatten the consumption curve. Voluntary climate pledges, however sincerely intended, cannot compete with the commercial pressure to deploy more AI faster. The trajectory is not a projection. It is already written into the infrastructure spending commitments Google has made publicly.
