Google's Android antitrust defeat and a 37% power surge: the cost of winning AI

Lead

On 2 July 2026, the European Union's General Court in Luxembourg delivered Google its most consequential antitrust defeat of the decade. The court upheld, in substance, a 2018 European Commission finding that Google had abused its dominant position in mobile software by tying its Play Store and search engine to the Android operating system — a strategy the Commission said was designed to lock rival search providers out of the most lucrative real estate on the smartphone. The figure attached to that finding, now a confirmed liability, is €4.1 billion (roughly $4.7 billion at then-prevailing exchange rates).

The ruling lands at an awkward moment for the company. The same week, Google's own environmental disclosures recorded a 37% year-on-year increase in electricity consumption across its operations in 2025, a jump the company attributes directly to the build-out of artificial-intelligence infrastructure. Two storylines that Google's communications team has long treated as separate — regulatory hostility in Brussels, and the punishing physics of frontier-model training and inference — are now colliding in the same news cycle.

Nut graf

The lesson is not that Google is uniquely villainous. It is that the platforms now defining the next decade of the internet face a three-front squeeze that no previous generation of tech incumbents had to manage simultaneously: a hard legal limit on how they bundle their products, a hard physical limit on how much power their AI ambitions consume, and a soft but accelerating limit on how credibly they can promise investors a carbon-neutral future. Each front constrains the others. Any one of them would be a story. Together, they describe the cost of winning the AI race.

The court, the fine, and what 'Android openness' actually means

The Commission's original case turned on a contract. Google required handset makers who wanted pre-installed access to the Play Store to also install Google Search as the default, and to refrain from selling devices running "forked" versions of Android — a stipulation that, in practice, meant that any device shipping with Google's app ecosystem had to run Google's Android, not a competitor's modified variant.

According to the Commission's 2018 decision, the result was that "more than 95%" of GoogleSearch queries on Android devices went through Google's own engine, foreclosing rival search providers at the operating-system layer rather than on the merits. The €4.1 billion fine was at the time the largest antitrust penalty the EU had ever imposed on a single company, later surpassed only by a separate €4.34 billion fine on Apple's Irish tax arrangement.

Google's response, repeated on 2 July after the ruling, is that the framing misses the point. A spokesperson said the judgment "fails to recognise" the firm's "significant investment to ensure Android remains free and open," a position the company has held since the original decision. The argument, in essence, is that Android's openness is itself the product — that Google built and gives away the operating system on which a global device ecosystem runs, and that the Commission is penalising the firm for succeeding at distribution without weighing what it gave away.

That defence is not insubstantial, and it has supporters. Android is, by Google's account and by independent measurement, the most widely deployed mobile operating system on Earth. The Commission's own logic, however, is that being widely deployed is precisely what made the bundling abusive: a dominant platform cannot use its dominance to lock out the next platform's competitors. Courts in Luxembourg and Brussels have now told Google, twice, that this distinction is going to hold.

The AI electricity problem, in numbers

If the antitrust ruling is the legal limit, the energy story is the physical one. Google's 2026 environmental report — the document Ars Technica began unpicking on 2 July — records a 37% increase in electricity use across 2025, attributed squarely to data-centre expansion tied to AI workloads. The number sits awkwardly next to Google's long-running commitment to operate its entire business on carbon-free energy on a 24/7 basis by 2030.

The company is not alone. Amazon, Microsoft's hyperscale peers, and the AI-specialised cloud operators are all reporting similar curves. As TechCrunch put it on 2 July, the build-out is "a warning sign about AI's real cost" — not the per-token inference cost that has been falling sharply for two years, but the upstream cost of cooling, siting, and powering the GPUs that produced those falling inference costs in the first place.

What makes Google's case particularly visible is the asymmetry between its two public commitments. On one hand, the company tells investors and customers that it is on a credible path to round-the-clock clean energy. On the other, it is signing long-dated power-purchase agreements and, in some jurisdictions, siting data centres next to gas turbines or behind-grid interconnection queues, in order to bring AI capacity online faster than new renewable generation can be permitted and built. Clean-energy matching on an hourly basis is hard. Matching it while capacity roughly doubles every eighteen months is harder.

Why the two stories are the same story

There is a temptation to read the antitrust fine and the energy disclosure as a coincidence of timing. They aren't. Both constrain the same core business — platform distribution — and both raise the same structural question: what does it cost a platform to be the platform?

In the antitrust case, the cost is paid in cash and in lost optionality. Google can absorb €4.1 billion without operational damage; it cannot absorb a court-imposed requirement to unbundle Android from the Play Store without rewriting the architecture of how its software reaches a billion users. The company has said it will study the ruling before deciding on further appeals, but the procedural runway is narrowing. The fine itself is the price of admission; the conduct remedy is the price of staying in business.

In the energy case, the cost is paid in grid capacity and in carbon. If Google is to keep its 24/7 clean-energy promise, the AI build-out has to land in regions with surplus renewable generation and the transmission to move it — not in the regions where land or tax breaks are cheapest. That constraint will, over time, push some data-centre investment into markets with slower renewable buildouts and fewer skilled-labour pools, raising unit costs and lengthening deployment schedules. Alternatively, the company can adjust the timeline for the 24/7 target, which is what the report conspicuously does not yet do.

The structural reading is that platform dominance and frontier-AI capacity are converging on the same bottleneck: power. Antitrust regulators in Europe and the United States have spent the past decade arguing that Google's leverage over distribution is the asset that needs policing. Climate advocates and grid operators are now arguing that AI is the load that needs policing. Both camps reach for the same lever — the right of a firm to dictate terms downstream — and both will find, over the next several years, that the other camp has logic.

Stakes, in plain terms

For Google specifically, the immediate stakes are existential for one product line and reputational for the broader company. Android's economics are built on the bundle; unbundling it on European terms would make the platform less valuable to handset makers at exactly the moment when regulators in the United Kingdom, India, and Japan are watching the European ruling closely and drafting their own parallel cases.

For the broader tech sector, the stakes are competitive. The hyperscalers that can build or contract renewable generation fastest will be the ones able to deliver AI services at the lowest environmental cost, which increasingly matters to enterprise procurement and to sovereign-cloud buyers in Brussels and Riyadh. The laggards will find that the AI discount they offer customers on inference pricing is eaten by energy surcharges and carbon offsets.

For the climate itself, the stakes are arithmetic. A 37% jump in a single company's electricity use, in a single year, is a large but bounded anomaly. If every hyperscaler reports similar jumps in their next disclosures, global data-centre electricity demand will cross thresholds that several major grid operators have already warned will require new firm capacity — and new firm capacity, in most jurisdictions today, means new gas turbines built underwritten as "bridge" generation. The bridge has a tendency to extend.

Where the evidence thins

There are three points the public reporting does not yet resolve. First, the General Court ruling was on Google's appeal of the Commission's liability finding; the Commission has separately indicated it may revisit the conduct remedies in light of the 2026 judgment, but no fresh procedural step has been confirmed. Second, Google's energy disclosures are company-published and unaudited in the financial sense; the 37% figure is internally sourced and may be revised in subsequent filings. Third, the link between the antitrust loss and the energy problem is structural inference on this publication's part — neither the court nor Google's own documents draw the connection explicitly. The two stories are converging, but the precise elasticity between them — how much each constraint tightens the other — remains a question for analysts with better access to Google's internal cost data than the public record provides.

Desk note: Monexus has tracked the EU's Google cases as a single arc since the original 2018 decision, and has flagged in earlier coverage that the European approach to platform regulation diverges from the United States on remedy design. This piece is the first to read those cases alongside the company's energy disclosures as two surfaces of the same platform-under-pressure story.