Are We Adopting AI Fast Enough to Avoid a Bubble?

The $2 Trillion Question

The AI industry is spending at a historic rate. Hyperscalers are on track to deploy $700 billion in AI infrastructure in 2026 alone. To put that in context: it exceeds the combined peak spending on the Manhattan Project, Apollo, and the Interstate Highway System as a share of GDP. Bain and Goldman Sachs estimate that the AI economy needs to generate approximately $2 trillion in annual revenue by 2030 to justify this level of investment.

The question is simple: will it get there?

Exponential Thinking

The optimistic case is straightforward. The AI economy has grown from $190 billion in 2023 to $390 billion in 2025, a compound annual growth rate of 43%. If that rate continues unchanged, revenue reaches $2.1 trillion by 2030. Threshold cleared, no bubble.

This is the implicit model behind most bullish forecasts. It assumes growth continues at the same rate indefinitely, with no ceiling. When you are early on an adoption curve, that assumption feels self-evident. It is also, historically, always wrong at some point.

Why Exponentials Break

Nothing grows exponentially forever. Every technology adoption in history has followed an S-curve: a slow start, steep acceleration, then saturation as the market approaches its ceiling (the total addressable market). Electricity, radio, television, personal computers, the internet, smartphones. All followed this pattern without exception.

The S-curve matters because it decelerates. The same adoption speed that produces explosive early growth produces diminishing growth as the market fills up. A 43% CAGR today does not guarantee 43% in 2028. The rate that matters is not the headline growth rate but how much of the addressable market has already been captured. Early-stage exponential growth and late-stage S-curve deceleration look identical in the first three years of data.

Fitting the Data

Logistic S-curves fitted to the three years of known revenue data (2023 to 2025), assuming a total addressable market of $4 trillion (a generous estimate supported by UNCTAD's projection of a $4.8 trillion AI market by 2033), yield a clear finding.

Every S-curve that fits the known data within 10% error has an adoption speed parameter (k) between 0.29 and 0.48. This range is wide. It produces outcomes by 2030 that span from a clear bubble collapse to clear sustainability.

The best fit to the data is k = 0.38. In historical terms, that is smartphone-speed adoption: the fastest mass-technology diffusion ever observed prior to AI. At k = 0.38 with a $4 trillion TAM, the AI economy reaches approximately $1.7 trillion by 2030. That is $300 billion short of the threshold.

The breakeven adoption speed is k = 0.43. Slightly faster than smartphones, and faster than any enterprise technology adoption in history. At this speed, the AI economy just clears $2 trillion by 2030.

The Fan of Uncertainty

Three years of revenue data are consistent with outcomes ranging from $1.2 trillion to $2.3 trillion by 2030. That range spans from unambiguous bubble collapse to clear sustainability. The honest answer is that we do not yet know which path we are on.

The difference between the bottom and top of this range is not about whether AI works or whether people want it. It is entirely about how fast enterprises move from testing AI to paying for it at production scale. The OECD reports that 88% of firms have tried AI but only 20% have deployed it in production. Closing that gap faster pushes outcomes toward the top of the fan. A slow close keeps them near the bottom.

What This Means

The AI economy is underwater today. Revenue is below the ecosystem sustainability threshold and has been since the current infrastructure buildout began. This is not a prediction. It is the current state. OpenAI lost $5 billion on $3.7 billion in revenue in 2025. Anthropic spent $10 billion to earn $5 billion. The entire industry is cash-flow negative.

The question is not whether this constitutes a bubble. By any standard definition, it does. The more important question is what kind: one that resolves into long-term sustainability (as the dot-com bubble ultimately produced the modern internet) or one that collapses before revenue catches up with spending.

This is not a forecast. It is a measure of how much of the current uncertainty space sits in the danger zone. To be in the safe 26%, enterprise adoption needs to slightly exceed smartphone-speed diffusion. That has never happened before for an enterprise technology category.

Where We Could Be Wrong

This analysis rests on two assumptions that could both break in the same direction, potentially overstating the risk significantly.

The TAM Could Be Larger Than $4 Trillion

The $4 trillion ceiling is a generous but bounded estimate, supported by UNCTAD's $4.8 trillion projection for 2033. If AI genuinely restructures the entire knowledge economy, as PwC's $15.7 trillion GDP impact estimate suggests, then the ceiling is far higher and the S-curve would not begin to decelerate before 2030. With a $15 trillion TAM, the best-fit k = 0.38 curve comfortably clears $2 trillion and the exponential projection becomes essentially correct.

The Adoption Speed Could Accelerate

The model fits the 2023 to 2025 trajectory and projects it forward unchanged. But AI capabilities are improving rapidly. Each step-change in capability opens new use cases and converts trial users into paying customers. A Will Smith eating pasta video was a joke in early 2024; Higgsfield was producing commercial-grade video by 2026. We are fitting a curve to three data points in a technology that reinvents itself every twelve months. Step-changes in capability can steepen the S-curve in ways that historical precedent cannot capture.

Both errors point in the same direction: the 74% probability of missing the threshold may overstate the risk. The model captures what the data says today. The data is young and the technology is not standing still.

The Next Data Point Will Be Decisive

2026 full-year AI revenue will narrow the uncertainty fan substantially. If revenue comes in above $600 billion, the slow scenarios are ruled out and the industry is likely on track. If it comes in below $500 billion, the fast scenarios are eliminated and the current spending trajectory becomes very difficult to justify.

We will know considerably more by early 2027.