Why AI bubble fears remain premature despite uncertain enterprise returns

An AI bust may be premature, but enterprise returns remain unproven

An AI supplier can keep expanding while its customers are still working out whether the technology earns its cost. That gap explains Forrester’s assessment that fears of an imminent AI bubble burst are premature, even as infrastructure investment exceeds measurable value. For enterprise leaders, the financial strength supporting expansion offers near-term reassurance, but the long-term investment case remains unresolved.

The investment case rests on two questions: who can keep paying for data centres, chips and networking, and whether enterprise demand will ultimately justify that spending. Forrester examines both through a model combining infrastructure supply, enterprise demand and funders’ financial health. The research and advisory firm has a commercial interest in demand for technology analysis and advice; its assessment allows for continued expansion alongside growing financial pressure, with corrections of different sizes still possible.

Strong funders can sustain expansion as financing pressure rises

Forrester’s near-term reassurance rests on the finances of the companies providing most of the money. Brian Hopkins, Vice President, Emerging Technology and Principal Analyst at Forrester, says about nine-tenths of today’s AI computing infrastructure funding comes from large, financially healthy technology companies using their own cash flow. The cash their operations generate gives them substantial capacity to keep investing.

But that capacity has weakened, according to Forrester’s financial-health model. The aggregate score for the nine companies it tracks fell from 98 to 82 over the past year, although the largest infrastructure players remained within the model’s healthy range. They can still fund expansion, but their margin of safety has narrowed. Their ability to finance construction also leaves open whether it will earn an adequate return.

Recent spending shows why that narrower margin matters. Forrester reports that Amazon, Alphabet and Meta each spent more in the last quarter than they generated in operating cash, the cash produced by running their businesses. Their operations therefore did not cover spending during that period, making other funding more important. Forrester identifies greater use of borrowing, which must be repaid, and share issuance, which raises money in exchange for ownership, as responses to this pressure.

Yet external financing needs to be considered alongside operating cash, because companies borrow from different financial positions. Forrester reports that Microsoft continued to cover its spending through operating cash while also raising USD $25 billion in bonds. Its borrowing adds to capital-market activity without indicating an operating-cash shortfall. Forrester interprets the broader shift toward external funding as mounting financial pressure rather than imminent distress.

The pressure is also uneven, so enterprise buyers need to examine individual suppliers’ finances. Hopkins identifies a small number of builders under real financial strain, while Forrester points to growing dependence on external capital among less financially secure participants. The strongest companies dominate aggregate funding, so their health can obscure weaker positions elsewhere. An enterprise remains exposed to the supplier supporting its systems, even when the market-wide assessment is healthy.

Forrester’s break-even case requires value beyond productivity

Financial strength gives expansion time to continue, but Forrester’s economic estimate sets a separate test for success. The firm estimates that AI must lift US GDP by 5% by 2035 for current infrastructure investment to break even. GDP measures the value of goods and services produced in an economy, so this requirement concerns economic output rather than companies’ ability to finance construction. AI use must ultimately generate enough value to cover the spending.

Published estimates of AI-driven productivity gains fall short of that requirement in Forrester’s analysis. Those estimates average 2%, leaving roughly USD $1.3 trillion in additional value to come from benefits beyond productivity. Doing existing work with fewer resources can produce economic gains, but Forrester’s framework requires enterprises to create other lasting benefits as well. Whether those benefits can reach the scale needed to justify the investment remains unresolved.

The scale required changes how leaders should interpret a successful deployment. Evidence that employees complete work faster can support an individual investment decision, while the infrastructure case depends on the value organizations generate across their use of AI. A useful application establishes a local benefit; market-wide expansion requires sufficient gains across the economy. Forrester’s break-even requirement looks ahead to 2035, and the average published productivity estimate describes expectations rather than benefits already realized.

Because those returns take time to emerge, Forrester expects financing pressure to remain central until enterprise economic gains reach sufficient scale. Financially strong funders can support further investment during that period, allowing expansion to continue while its eventual returns remain uncertain. So assessing progress means distinguishing growing activity from durable value for customers.

Usage growth and capacity purchases leave durable demand uncertain

Google’s processing growth illustrates how quickly activity can rise. Tokens are units of content processed by AI systems, and Google’s monthly total rose from about 9.7 trillion in May 2024 to more than 3.2 quadrillion in May 2026. Processing volume has grown sharply. The economic return customers receive is a separate measure.

Forrester places that growth alongside its model’s estimate that AI has added less than half a percentage point to US GDP through productivity so far. Google’s company-level usage figures and the economy-wide estimate cover different activities and scales, so they cannot establish how effectively Google’s processing produces economic gains. Rising consumption still leaves the return question open. Enterprises need lasting benefits from the work those systems perform.

The demand question also extends to companies buying computing capacity between infrastructure providers and enterprise customers. Forrester identifies a middle tier comprising AI-native startups, software vendors, neoclouds (specialized cloud-computing providers), and model labs. Their purchases create demand for infrastructure builders and operators, supporting further expansion. Whether that demand lasts depends on what happens further along the chain, including whether enterprise customers gain enough value to sustain their spending.

Intermediary purchases can therefore support infrastructure growth while durable enterprise demand remains uncertain. Forrester’s concern is the gap between capacity being taken up now and end customers establishing lasting economic benefits. Suppliers’ purchases can keep that gap open for longer. Forrester consequently sees excess capacity as one part of the correction risk, with its significance depending on how much demand develops.

Financing relationships add another risk within that chain. Hopkins identifies circular financing deals involving AI labs, arrangements in which money moves back through linked participants, among the areas Forrester is watching, alongside the financially strained builders. Ordinary capacity purchases do not by themselves establish such a relationship. Circular arrangements create additional exposure between participants while the economic strength of end demand remains uncertain.

Demonstrated value and supplier finances guide investment decisions

This uncertainty leads Forrester to recommend selective continued investment. “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,” Hopkins says. The recommendation makes demonstrated results the basis for continued spending while treating supplier durability as a separate condition for long-term reliance.

Supplier durability matters more when a company increasingly depends on outside capital to meet its commitments. The financial health of the largest technology groups cannot substitute for examining the company that provides an enterprise’s computing capacity or AI services. A supplier may face funding strain while the wider market keeps expanding, so its ability to finance obligations matters throughout the period an enterprise expects to rely on it.

Main highlights

• Strong funders support continued expansion: Financially healthy technology companies supply about 90% of AI infrastructure funding, giving the market room to grow. Procurement teams need to assess individual suppliers’ cash generation and reliance on external capital, because financial pressure varies across providers.
• Enterprise returns need broader sources of value: Forrester estimates that AI must lift US GDP by 5% by 2035 for infrastructure investment to break even, exceeding average published productivity estimates of 2%. Business sponsors need to measure lasting economic benefits beyond time saved.
• Usage growth leaves demand quality uncertain: Rising token volumes and intermediary capacity purchases show activity, while durable demand depends on enterprise customers gaining sufficient value. Technology buyers need evidence of sustained customer benefits when assessing suppliers’ growth claims.
• Demonstrated value supports selective investment: Forrester recommends continued spending where AI delivers measurable benefits, alongside monitoring supplier finances. CIOs and CTOs need investment plans that account for supplier funding strain and market corrections of varying sizes.