Reading the Gartner $2.5 Trillion AI Forecast as an IS Researcher

Gartner's January 2026 forecast says worldwide AI spending will total $2.5 trillion in 2026, up from $1.5 trillion in 2025. The first time I read that number I had to sit with it for a moment. A trillion dollars is not a concept that resolves easily into something you can actually picture. Two and a half of them, in a single year, in a single technology category, is the kind of figure that either means something transformative is happening or that something very expensive is happening, and those are not the same thing.

The number that followed was the one that gave me more to think about. AI agent software is projected to grow from $86.4 billion in 2025 to $206.5 billion in 2026 to $376.3 billion in 2027. That is more than a quadrupling in three years. GenAI model spending is growing at 80.8% in 2026 alone. These are not incremental growth rates. They are the kind of numbers that show up when a market is in its speculative expansion phase, when capital allocation runs ahead of demonstrated value and organizations buy because the category is considered strategic, not because they have verified ROI.

IS research has a name for the gap between spending and value. Brynjolfsson named it in 1993 when he documented the productivity paradox: decades of IT investment through the 1970s and 1980s had not produced visible productivity gains in macroeconomic statistics. His four explanations still organize the question. Mismeasurement says the value is there but our metrics do not see it. Time lags say the value is real but takes years to materialize as organizations restructure around new tools. Redistribution says value is created but passed to customers as lower prices rather than captured as profit. Mismanagement says the investment is real but the complementary organizational changes that would convert it into performance never happen. The mismanagement explanation is the one I keep coming back to when I look at the $2.5 trillion forecast alongside McKinsey's finding that only 7% of organizations have fully scaled AI.

Melville, Kraemer, and Gurbaxani (2004) built the integrative model that refuses the shortcut between spending and value. IT resources do not directly produce organizational performance. They interact with complementary organizational resources, produce changes in business processes, and those process changes drive performance outcomes, all moderated by industry characteristics and the competitive environment. The implication for the $2.5 trillion is uncomfortable: the dollars show up in Gartner's forecast regardless of whether the complementary organizational investments follow. The vendors get paid whether or not the buying organizations change their processes, build the data governance, develop the internal expertise, or redesign the workflows that would allow the technology to produce value. Spending is the input. Value requires everything else.

The Brynjolfsson productivity paradox was supposed to be resolved in the early 2000s, after Brynjolfsson and Hitt (1996) showed that IT capital had a higher marginal product than non-IT capital at the firm level and that the paradox was partly an artifact of aggregating across firms rather than a real absence of value. The partial resolution was real. But the deeper question was never about whether IT could produce value. It was always about whether it does, in specific organizational contexts, with specific investments, and specific complementary conditions. That question does not get resolved by a resolution of the original paradox. It reasserts itself every time a new technology category attracts capital at the scale AI is attracting now.

What worries me about the $2.5 trillion is not that the number is wrong. Gartner's methodology for these forecasts is rigorous and their track record is reasonable. What worries me is the structure of what the number represents. A substantial portion of that spending is infrastructure and vendor contracts: cloud compute for training and inference, enterprise software licensing, consulting fees for implementation programs. Those are real expenditures with real beneficiaries, primarily the hyperscalers and the major AI vendors. Whether the organizations doing the buying extract commensurate value depends on factors the spending figure cannot see. Do they have the internal knowledge to evaluate which AI applications are worth pursuing? Do they have the process maturity to redesign workflows around AI outputs? Do they have the governance structures to catch and correct model errors before those errors propagate into business decisions?

Absorptive capacity is the theoretical construct that captures this gap. Cohen and Levinthal (1990) showed that the ability to extract value from external knowledge depends on prior related knowledge. Organizations that had already built data capabilities, analytical routines, and evidence-based decision-making before the AI wave hit can absorb what the $2.5 trillion is buying. Organizations that had not built those foundations are buying access to something they are not yet equipped to exploit. The spending is real in both cases. The value extracted is very different.

I do not read the $2.5 trillion as evidence that AI is producing $2.5 trillion in value. I read it as evidence that organizations believe, or have been convinced, that AI spending is strategically necessary in 2026, which is a different claim and a weaker one. The McKinsey 7% scaling rate says the same thing from a different angle. The spend is massive. The organizational ability to translate that spend into scaled performance is rare. The gap between those two facts is where the interesting IS research lives, and it is also where the next version of the productivity paradox is quietly assembling itself.

The Gartner number will keep growing. The 2027 forecast for AI agent software alone is $376.3 billion. By the time the value question gets sorted out empirically, the spending will have moved on to the next cycle. That is how the paradox persists. Not because organizations refuse to learn, but because the technology and the capital cycle faster than the organizational capability to absorb what the capital buys.