Engagement Over Efficiency: Why Big Tech's Power Numbers Don't Add Up

The following represents the analytical perspective of Ampergram's editorial team and is intended to provoke informed discussion among technically sophisticated readers.

There is a particular kind of corporate document that arrives each spring, glossy with renewable energy commitments and net-zero pledges, populated with carefully selected metrics designed to convey progress without inviting scrutiny. The annual sustainability report from a major technology company is, in many respects, a masterclass in selective measurement. And for an audience that understands the difference between apparent power and real power, the discrepancy is difficult to ignore.

This is not a piece about whether tech companies are lying. Most of what they publish is technically accurate. It is a piece about what they choose not to measure—and what independent analysis reveals when someone actually bothers to run the numbers.

The Engagement-Energy Paradox

Here is the central tension that no sustainability report adequately addresses: the business model of every major social media and cloud platform is built on maximizing engagement. More time on platform means more ad impressions. More ad impressions mean more revenue. More revenue funds more infrastructure. More infrastructure consumes more power. This is not a conspiracy; it is arithmetic.

When Meta reports that its global operations are powered by 100% renewable energy, that claim is technically structured around renewable energy certificates and power purchase agreements—financial instruments that allow a company to claim renewable attribution for grid power it did not physically receive through a dedicated clean energy connection. The electrons flowing into a data center in DeKalb County, Georgia, are drawn from whatever the regional grid is producing at that moment. In the Southeast, that mix still includes a meaningful share of natural gas and coal.

This is not unique to Meta. Google, Microsoft, and Amazon Web Services employ similar accounting frameworks. The renewable energy claim is real in a contractual sense. Whether it reflects the actual carbon intensity of the electrons powering a given server rack is a more complicated question—one that the companies themselves rarely encourage their audiences to ask.

What Independent Measurement Actually Reveals

Several research institutions and independent analysts have attempted to construct bottom-up estimates of Big Tech's actual power consumption, rather than relying on corporate disclosures. The results are instructive.

The International Energy Agency estimated in its 2023 data center and transmission networks report that global data centers consumed between 240 and 340 terawatt-hours annually—a range wide enough to obscure significant uncertainty. That uncertainty is not accidental. Major operators disclose aggregate consumption figures but resist facility-level granularity that would allow external verification.

In the United States, the Environmental Protection Agency's ENERGY STAR program for data centers provides a voluntary benchmarking framework, but participation is not mandatory and self-reported figures are not independently audited. A 2023 analysis by researchers at Lawrence Berkeley National Laboratory found that actual US data center energy consumption likely exceeded earlier projections by 15 to 30%, driven primarily by the accelerating deployment of AI inference workloads—workloads that the major platforms had not prominently featured in their prior efficiency narratives.

For context: a single large language model training run can consume upward of 500 megawatt-hours. The inference workload—serving responses to users at scale—is less intensive per query but runs continuously and at extraordinary volume. When a social media platform deploys an AI-driven content recommendation engine across a billion daily active users, the aggregate inference load is not a rounding error. It is a material portion of the facility's power budget.

PUE as a Distraction Metric

Power Usage Effectiveness has become the de facto headline metric for data center efficiency, and the major operators have become very good at optimizing it. Google has reported facility-level PUE values as low as 1.06. Microsoft has published similar figures for its newer hyperscale campuses. These are genuine engineering achievements.

But PUE is a ratio, not an absolute measure. A facility that draws 10 gigawatt-hours annually with a PUE of 1.1 consumes 1 gigawatt-hour on non-compute overhead. A facility that draws 100 gigawatt-hours with a PUE of 1.1 consumes 10 gigawatt-hours on overhead. The ratio looks identical. The absolute consumption is an order of magnitude larger. As platforms scale their infrastructure to support higher engagement volumes, a stable or improving PUE ratio can coexist with dramatically increasing total energy draw. This is precisely what the data suggests has been occurring.

The technically honest metric would be energy consumed per unit of useful compute output—what the industry sometimes calls compute efficiency or carbon per workload. This figure is rarely the centerpiece of a sustainability report, because it requires disclosing the denominator: how much compute is being consumed in total, and for what purpose.

The Audit Gap and What It Costs

The absence of mandatory, independently audited energy disclosure for major US technology operators is a structural problem with measurable consequences. Without standardized reporting requirements, there is no reliable mechanism for external analysts to verify corporate claims, compare operators on consistent terms, or identify whether stated efficiency improvements are translating into actual grid impact reductions.

The Securities and Exchange Commission's proposed climate disclosure rules, which have faced significant legal and political opposition, would have required large public companies to disclose Scope 1 and Scope 2 emissions—the latter covering purchased electricity. Even if implemented in their original form, those rules would not have required the facility-level granularity necessary for meaningful independent verification of data center efficiency claims.

For the technically sophisticated reader, this represents a familiar problem in a new domain: you cannot manage what you do not measure, and you cannot verify what is not disclosed. The power metering exists inside these facilities. The data is being captured. The decision not to share it externally is a policy choice, not a technical limitation.

What a Real Audit Would Look Like

An independent, credible audit of a major platform's energy footprint would require, at minimum: facility-level power consumption data broken down by workload category; real-time grid carbon intensity data for each facility's regional grid interconnection; a transparent accounting of renewable energy certificate methodology versus physical delivery; and compute efficiency metrics normalized by active user count and engagement volume.

None of this is technically infeasible. Several smaller cloud operators and co-location providers already publish more granular data than their hyperscale counterparts. The constraint is not instrumentation—it is disclosure policy.

Until that changes, the sustainability reports will continue to arrive each spring, full of carefully chosen ratios and renewable energy attributions. They will be accurate in the ways that are easiest to be accurate, and silent on the measurements that would tell the most complete story.

The Ampergram Standard

At Ampergram, we apply a straightforward editorial standard: metrics that matter should be measured, disclosed, and subjected to independent scrutiny. That standard applies to voltage regulation in a distribution network, to engagement rates on a content campaign, and to the power consumption of the infrastructure that makes digital life possible.

The technology sector has demonstrated extraordinary engineering capability in optimizing the metrics it chooses to prioritize. Engagement per server rack. Revenue per watt. Renewable attribution percentage. The question worth asking—the question this editorial team intends to keep asking—is which metrics are being left off the dashboard, and why.