Forrester: AI bubble fears premature, but math is getting harder

AI would need to lift US GDP by 5% by 2035 for current infrastructure investment to break even, according to new Forrester research that argues fears of an imminent AI bubble are premature — even as the financial pressure behind the build-out rises.

The gap between that target and reality is the story in one number. Published estimates of AI-driven productivity gains average 2%, which leaves roughly USD $1.3 trillion in additional value that must come from benefits beyond productivity — things like new products, better decisions, or business models that productivity metrics don't capture.

Forrester's assessment lands in the middle of a heated debate over whether the heavy spending on data centres, chips and networking can be justified by near-term returns. Its conclusion: no sudden break yet, but the economics are becoming more demanding as funding needs spread beyond internal cash flow.

The verdict, in the analyst's own words

Brian Hopkins, Vice President, Emerging Technology and Principal Analyst at Forrester, put the firm's position bluntly.

"Fears that an AI bubble is about to burst are premature. About nine-tenths of the money building today's AI computing infrastructure comes from large, healthy technology companies paying out of their own cash flow," Hopkins said.

That nine-tenths figure is the foundation of Forrester's case. The risk, in its view, is not spread across the market. It sits in two places: "a small number of builders under real financial strain, and circular financing deals with the AI labs."

Hopkins also rejected the idea that too much capacity, by itself, will trigger a crash. "Oversupply on its own won't decide whether a correction comes. Demand will, and that's the biggest unknown, because it depends on enterprises finding lasting value in AI beyond simple productivity gains," he said.

His advice to buyers is not to retreat. "Leaders shouldn't pull back. Instead, they should keep investing where AI delivers value they can demonstrate, watch the financial health of the companies building AI computing closely, and be ready for corrections, large or small."

How Forrester built the model

The research rests on a model that tracks three things at once: infrastructure supply, enterprise demand, and the financial health of the companies funding the AI build-out. That last leg is where the picture is shifting.

The largest builders remain financially healthy — all of them still score within the model's healthy range. But the margin has narrowed. Forrester's financial health score for the nine companies in its model fell from 98 to 82 over the past year.

That 16-point drop doesn't signal distress. It signals strain, and the strain shows up in how the spending is being paid for.

Take the last quarter. Amazon, Alphabet and Meta each spent more than the operating cash they generated. Microsoft kept funding spending from operating cash but also raised USD $25 billion in bonds.

The takeaway: Big Tech is starting to lean on borrowed money to keep the build-out running.

Where the real risk sits

Forrester said financial risk is moving closer to the centre of the market, even though most AI infrastructure funding still comes from large technology groups with strong balance sheets.

The emerging concern is a growing reliance on outside capital at the edge of the market. Middle-tier suppliers — AI-native startups, software vendors, neoclouds and model labs — can absorb capacity and make demand appear stronger than it is, for a time.

That dynamic matters because it can extend the infrastructure build-out even if end demand from enterprises remains uncertain. Capacity keeps getting added; whether anyone ultimately pays for it is a separate question.

Forrester's model suggests oversupply alone will not determine whether a correction takes place. The decisive factor is whether enterprises convert rising AI use into lasting economic gains.

The takeaway: watch who's borrowing, not just who's building.

Usage is exploding. Value isn't keeping up.

The report points to a widening gap between how much AI is being consumed and how much value it's actually producing.

Google's monthly token processing rose from about 9.7 trillion in May 2024 to more than 3.2 quadrillion in May 2026 — a more than 300-fold increase in two years. Yet Forrester's model puts AI's productivity uplift to US GDP at less than half a percentage point so far.

Forrester said that divergence shows why rapid growth in AI usage should not be treated as proof of sustainable demand. The key unknown, in its view, is whether enterprises can find durable value in AI beyond straightforward productivity gains.

This is the mechanism behind the whole debate. Usage statistics are easy to grow, especially when tokens are cheap and applications are free. Economic gains are harder, because they require companies to change how they work, not just how much they process.

The takeaway: rising token counts measure activity, not value.

What should enterprise buyers actually do?

Forrester's findings carry a practical implication for anyone signing long-term AI contracts. Because financial risk is migrating from the core of the market to its edges, supplier financial health becomes a more important factor when weighing long-term AI strategies.

Hopkins's closing guidance maps onto that: invest where value is demonstrable, monitor the balance sheets of the companies building AI computing, and prepare for corrections of any size.

Forrester's own model gives buyers a way to do the monitoring. Its financial health scoring tracks the nine largest players, and the 98-to-82 decline over the past year shows the metric can move quickly.

Why this matters beyond AI

The stakes here reach past the technology sector. Forrester's break-even math — 5% of US GDP by 2035 — ties AI infrastructure spending to macroeconomic performance. If the gains don't materialise at that scale, the adjustment won't stay contained within tech company earnings reports.

For now, Forrester concludes the market does not point to a sudden break. But until enterprise customers turn rising AI use into lasting economic gains at sufficient scale, pressure on the companies financing the expansion will remain the central issue — and the number to watch is demand, not supply.