GenAI & Productivity - Executive Summary

In October, Bloomberg published an encapsulation of the staggering cost of the genAI data center buildout and the Everest-sized revenue-growth mountain the tech industry must climb to justify that investment. In 2025, Microsoft, Google, Amazon and Meta are projected to approach $400 billion combined on AI infrastructure spending (chart below). We quote Bloomberg:

Source: The Wall Street Journal

At Sage Road, we look for narrative polarization when identifying topics for reports—opportunities where our independence and research rigor can cut through maximalist noise to offer clients unbiased and differentiated investment guidance. We saw this with “Deregulation”, “US Reshoring”, and “The Retailization of Private Markets” (learn more here). Yet, nowhere have we found this opportunity more glaring than with genAI.

For months, the genAI debate has been fractured by bubble fears and wild speculation about the technology’s future potential. Researching this report, two headline-grabbing academic claims struck us as particularly illustrative of this extreme analytical polarization: An MIT report concluding that 95% of all AI initiatives across industries have “failed”. And the recent claim by University of Louisville computer science professor Roman Yampolskiy that AI will lead to 99% unemployment by 2030. How do investors and allocators diagnose risk and opportunity when the range of potential outcomes is so staggeringly vast?

Seismic productivity gains are required throughout the global economy to justify the extraordinary sums being invested in genAI today. The revenue growth necessary to pay for today’s investment can only come from three places: tech giant market-share expansion, labor replacement, and economic growth. Given their already monopolistic or duopolistic dominance in their respective markets, market-share expansion doesn’t appear an adequate long-term path to ROI for tech giants. So, it comes down to labor replacement and an economic growth acceleration and both will require significant per-worker productivity gains.

For most of the past two decades, US labor productivity has grown at a below average rate, costing the US nonfarm business sector roughly $11 trillion cumulatively, or $95,000 in output per worker, according to BLS calculations. Myriad factors have been credited for the slowdown, including regulation, globalization, industry consolidation, the GFC, and mismeasurement. Yet, we believe one factor matters more than all others: the innovation slowdown. As Stanford researchers concluded in a 2020 paper: “Economic growth arises from people creating ideas…In many contexts and at various levels of disaggregation, research effort is rising substantially, while research productivity is declining sharply. Steady growth, when it occurs, results from the offsetting of these two trends.” Will genAI prove a game-changing offset, unlocking new ideas and triggering productivity across the economy?

There are compelling, non-speculative reasons to believe the answer is yes. For years, AI integration in different forms has contributed to skyrocketing tech giant productivity and therefore, profit growth, optimizing everything from product and content curation to ad targeting, cloud services, and operational efficiency. Netflix has estimated that its AI-driven recommendation engine saves the company $1 billion annually in customer retention. Meta has claimed that its AI recommendation systems led to a 5% increase in time spent on Facebook just in 3Q25.

GenAI promises to democratize AI benefits, enabling scalable, adaptable, and intuitive integration of genAI capabilities across industries. Adoption rates suggest those benefits could manifest rapidly. ChatGPT reached 100 million users in just its first two months after launch, the fastest adoption rate ever recorded for an application. As of October, it had over 800 million active weekly users, reaching more than 10% of the global population, a “speed of global diffusion [that] has no precedent," as Duke and Harvard researchers put it. Enterprise adoption has been swift too. Based on McKinsey survey results, roughly 80% of enterprises now use genAI to some degree in their operations, a number that has nearly tripled in just two years.

Yet, to our mind, it’s the right-hand side of the chart above that begs the preeminent question of concern. Roughly 60% of enterprise AI initiatives remain in their pilot or experimentation phase. Will they actually prove productivity enhancing when scaled? There is clear reason for doubt. The MIT study has been the most cited—and maligned—academic claim about genAI inadequacy. Yet, MIT is not alone in voicing concern. In September, Harvard Business Review published survey results suggesting 40% of workers had to deal with “AI workslop” in the month prior—subpar work generated by colleagues that dragged on rather than enhanced productivity. Then, in October, software company Atlassian released survey results echoing MIT’s finding—96% of businesses "have not seen dramatic improvements in organizational efficiency, innovation, or work quality.”

ChatGPT’s own data begs questions. Between June 2024 and June 2025, ChatGPT use for work tasks plummeted as a share of total ChatGPT usage (first chart below). This may simply reflect exceptional engagement for non-work tasks. Or it may be because users are finding more value in ChatGPT as an assistant in completing lower-consequence household tasks than higher-consequence work-related tasks. More recent data paints an even more problematic picture. In late November, the Census Bureau released survey data suggesting genAI usage at work has actually fallen, down a percentage point since the beginning of this year (second chart below).

Source: NBER

Source: The Economist

While researching this report, much of the anecdotal evidence we accumulated talking to clients and contacts across industries mirrored these survey results. Hallucinations are compromising trust and therefore, reliance on genAI. Workslop is transferring drudgery up the organizational value chain. And obsolescence-fearing C-suites are making vague and misguided demands about AI integration, leading to a lot of talk with little productive action. Our own experience has told a similar story. Since forming Sage Road, we have relied more than ever on genAI in our research process. However, its productivity impact has been limited. It’s more efficient than traditional search for research discovery. It’s ineffective as an analytical assistant—too banal and prone to hallucination to rely on for interpretation of the complex topics we tackle.

As JP Morgan noted in early October: “AI-related stocks have accounted for 75% of S&P 500 returns, 80% of earnings growth and 90% of capital spending growth since ChatGPT launched in November 2022.” And as Apollo’s Torsten Sløk has noted, the contribution of data center investment to real US GDP growth equaled consumer spending’s contribution in the first half of 2025. Undoubtedly, there are already targeted, productivity-enhancing use cases for genAI across industries, which we will detail in this report. However, we also see compelling reason for concern that near-term, step-change improvements will be necessary for the technology to drive significant productivity gains across the economy. And those gains are required to sustain market optimism about the timeline to ROI.

From tech luminaries to pundits, analysts, and C-suite executives, many seem to assume that improvement is inevitable. We do not approach our analysis with assumptions. Will the technology’s rapid-rate of improvement sustain or will intelligence gains slow or worse, hit a comprehension wall? Will more secondary factors like unit economics, circular dealmaking, energy undersupply, or a political backlash hold back progress?

These are some of the questions we dissect in this report. We dug into academic, consultant, and analyst studies, worker testimonials, reported applications across industries, and speculation about future applications. Obviously, we remain in the nascent phase of genAI. There is much we cannot anticipate regardless of how deep we research and contemplate. To be clear, we believe in AI’s transformative potential—it is the definitional technology of the next half century and beyond. But will its true disruptive force manifest in 2026, 2036, or even later? With this report, we are focused on a one- to three-year investment horizon. We do not fear a dotcom-like market bubble burst as much as many others—not when the epicenter of the so-called “bubble” is the most profitable companies in the world. We do, however, believe the better investors and allocators understand the technology’s current strengths and weaknesses, and trajectory of improvement, the better they can anticipate how, when, and to what extent productivity gains will be realized and shape investment returns across asset classes, regions, and sectors. The report encapsulates how we see that equation today.