AI's Dark Output: The Hidden Economic Value We Can't Measure 2026

AI is producing an incalculable amount of economic value, and that is precisely the problem: we don't know how to calculate it. The cost of developing AI is well known—massive data centers, skyrocketing electricity consumption, and soaring capex from tech companies. But the return on that investment seems insufficient, fueling bubble fears. However, the real issue may be that our measuring tape is broken.

The Hidden Production: What Is 'Dark Output'?

In a detailed analysis in the Semianalysis newsletter, the term 'dark output' is used to describe the economic value generated by AI that current measurement systems fail to capture, and therefore does not impact GDP. This hidden production has two main facets: first, the value created by AI tools that are free or subsidized, and second, the productivity gains that are not reflected in traditional metrics.

For example, when an AI assistant helps a worker complete a task faster, the time saved might be used for leisure or other unpaid activities, which doesn't show up in GDP. Similarly, AI-powered services that are offered for free (like improved search or translation) create consumer surplus that is invisible in economic statistics.

Why It Matters: The High Stakes of Mismeasurement

The thesis of the analysis is that we are not facing a bubble, but rather that we are not measuring the return AI is producing correctly. This is a problem that goes far beyond a simple statistical debate. Macroeconomic data are the metric investors use to detect real growth, central banks adjust interest rates based on them, and companies decide whether to hire or automate.

Making decisions of that magnitude based on inaccurate data can have serious consequences. If AI's true contribution is underestimated, we might underinvest in the technology or misinterpret inflationary pressures. Conversely, if we overestimate it, we risk misallocating capital.

The Difficulty of Measuring Intangible Value

Services and intellectual work are much more complicated to measure than goods. GDP was designed for an industrial economy where output was tangible—cars, steel, crops. Today, value is increasingly created through software, data, and algorithms, which are intangible and often free.

AI accelerates this trend by automating cognitive tasks and generating new forms of value that don't fit neatly into existing categories. For instance, an AI that writes code or designs a logo produces something that might have been paid for before, but now is often done in-house or by the user themselves, reducing market transactions that GDP would capture.

Key Takeaways

• Dark output refers to AI-generated economic value that current measurement systems miss.
• Bubble fears may be misplaced; the real issue could be inadequate measurement.
• GDP was not designed for an economy of intangible, often free, digital services.
• Mismeasurement affects investment decisions, monetary policy, and business strategy.
• Improving metrics is crucial to understanding AI's true economic impact.

How AI's Value Escapes Traditional Metrics

Traditional metrics like GDP focus on market transactions. But AI often creates value outside markets: free tools, improved efficiency without additional spending, and new products that are given away to drive adoption. This value is real but invisible in national accounts.

Moreover, AI can lead to consumer surplus—the difference between what consumers are willing to pay and what they actually pay. When AI makes a service better or cheaper, consumer surplus increases, but GDP may not reflect it. For example, if an AI-powered translation tool replaces a paid human translator, GDP could actually fall even though the user is better off.

Comparing Measurement Approaches

Different methods attempt to capture AI's economic impact, but each has limitations. The table below summarizes key approaches and their ability to detect dark output. The longer history behind the story is traced in AI's Dark Output: The Hidden Economic Value We Can't Measure.

As the table shows, no single metric fully captures AI's value. A combination of approaches is needed to get a clearer picture. The background to this story is set out in Nadella Warns AI Bubble: Without Broad Economic Growth, It Won't End Well. A related thread runs through AI Boom Economic Impact: Are Datacentres Worth. Earlier coverage in Business: latest articles shows how the argument developed.

What to Watch Next

Policymakers and statisticians are beginning to recognize the need for better measurement. Initiatives to incorporate digital services and free goods into national accounts are underway, but progress is slow. Meanwhile, investors should look beyond GDP to indicators like corporate productivity, patent activity, and AI adoption rates. A closer look at the same subject appears in SpaceX IPO Reveals Gulf States as Key AI Tech Investors. Anyone following the wider picture can start with SoftBank CEO: Calling AI a Bubble Is Blasphemy Against AI. The same theme surfaces in Intel's $5.4B Ireland Chip Investment: What It Means for Europe, which takes a different angle.

For businesses, the lesson is to develop internal metrics that capture AI's impact on efficiency and innovation, rather than relying solely on macroeconomic data. The dark output is there—we just need to shine a light on it. Context on how this played out before sits in UK Heatwaves Cost Economy £4.4bn in Lost Output. A companion piece, Heatwaves Cost Europe €180bn in Lost GDP and Productivity, approaches the same subject from another angle.