{"id":228374,"date":"2026-06-02T08:18:47","date_gmt":"2026-06-02T08:18:47","guid":{"rendered":"https:\/\/teknomers.com\/en\/we-understand-the-cost-of-ai-but-measuring-its-output-remains-a-major-challenge\/"},"modified":"2026-06-02T08:18:49","modified_gmt":"2026-06-02T08:18:49","slug":"we-understand-the-cost-of-ai-but-measuring-its-output-remains-a-major-challenge","status":"publish","type":"post","link":"https:\/\/teknomers.com\/en\/we-understand-the-cost-of-ai-but-measuring-its-output-remains-a-major-challenge\/","title":{"rendered":"We Understand the Cost of AI, but Measuring Its Output Remains a Major Challenge"},"content":{"rendered":"\n

Understanding the True Cost and Value of AI Investment<\/strong><\/p>\n

We know very well the cost of developing AI: the mammoth data centers, the skyrocketing electricity consumption, and the capex of technology companies soaring. However, a significant problem arises: it appears that these investments are not yielding sufficient returns to justify their scale. While concerns about a potential bubble are valid, another underlying issue may be at play: our measurement systems are inadequate.<\/p>\n

The Hidden Production of AI<\/h3>\n

In an insightful analysis presented in the newsletter Semianalysis<\/strong>, the concept of “dark output” is introduced, referring to the economic benefits that AI creates but are largely overlooked by existing measurement tools, which subsequently do not reflect on GDP. This hidden production can be examined through two primary lenses:<\/p>\n

Hidden Production by Substitution<\/h4>\n

This aspect reflects jobs once performed by humans that AI now handles at a fraction of the cost. For instance, writing a will used to cost around $400, decreased to $150, and astonishingly, has plummeted to just $0.50 due to AI capabilities. While the task is completed, the associated economic transaction vanishes from standard financial data collection.<\/p>\n

New Production That Remains Hidden<\/h4>\n

Conversely, there are now jobs that could not be carried out due to high costs but are feasible due to AI’s affordability. An example includes bibliographic reviews, which historically cost up to $2,000, making them exclusive. With AI, these reviews can now be produced much more cheaply, leaving minimal economic traces other than token usage or subscription payments.<\/p>\n

The Importance of Accurate Measurement<\/h3>\n

This analysis suggests we are not confronting a bubble but are facing measurement inaccuracies regarding the returns on AI investment. This is a critical problem that extends beyond mere statistical discrepancies. Macroeconomic data serves as a crucial tool through which investors gauge real growth, central banks make interest rate adjustments, and companies decide whether to automate or hire. Decisions based on flawed data can lead to dire consequences.<\/p>\n

The Challenge of Measurement<\/h3>\n

Assessing services and intellectual labor is vastly more complex than quantifying physical goods. For instance, a furniture factory can easily measure increases in chair production due to new machinery. In contrast, AI facilitates programming, writing, and summarizing tasks, where measurement often relies on token consumption. Increased token use might drive significant benefits or yield poor-quality outputs, highlighting that value lies in the actual results rather than expenditure alone.<\/p>\n

Historical Precedents<\/h3>\n

A historical parallel can be made with the computer boom of the 1980s and 1990s when macroeconomic data failed to capture the revolution’s true impact. It wasn\u2019t until 2013 that R&D and intellectual property investments were integrated into GDP accounting, retroactively adding $3.6 trillion\u2014showing that by 2000, this sector constituted 30% of GDP.<\/p>\n

Moreover, consider the “care economy,” encompassing unpaid domestic and caregiving work, predominantly by women. The International Labor Organization<\/strong> estimated that in 2018, 16.4 billion hours of unpaid care work equated to $11 trillion, or 9% of global GDP.<\/p>\n

Conclusion: A Call for Revised Metrics<\/h3>\n

While the need to update our measurement frameworks is evident, it doesn\u2019t negate the reality of substantial investments in AI infrastructure. Predictions indicate that by 2025, tech companies will invest around $410 billion<\/strong> in AI, potentially exceeding $650 billion<\/strong> in 2026. The chief economist of Goldman Sachs even noted that the current contribution of this investment to the U.S. GDP is \u201cbasically zero.\u201d It is equally risky to label this scenario as a bubble primed to burst due to overexpenditure, as it is to blindly assume enormous invisible wealth justifies every dollar spent.<\/p>\n

Ultimately, revising how we measure AI\u2019s economic impact is crucial for accurately understanding its contributions and guiding future investments. <\/p>\n


\n
General News – 2<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Understanding the True Cost and Value of AI Investment We know very well the cost of developing AI: the mammoth data centers, the skyrocketing electricity consumption, and the capex of technology companies soaring. However, a significant problem arises: it appears that these investments are not yielding sufficient returns to justify their scale. While concerns about […]<\/p>\n","protected":false},"author":1,"featured_media":228375,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[36399],"tags":[7955,2180,187,17582,5968,1994,3430],"class_list":["post-228374","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technology","tag-challenge","tag-cost","tag-major","tag-measuring","tag-output","tag-remains","tag-understand"],"_links":{"self":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/posts\/228374","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/comments?post=228374"}],"version-history":[{"count":1,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/posts\/228374\/revisions"}],"predecessor-version":[{"id":228376,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/posts\/228374\/revisions\/228376"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/media\/228375"}],"wp:attachment":[{"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/media?parent=228374"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/categories?post=228374"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/teknomers.com\/en\/wp-json\/wp\/v2\/tags?post=228374"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}