Artificial intelligence (AI) has the potential to become one of the largest innovation waves in history in terms of capital investment (capex), related earnings growth, and for debt and equity capital markets. Indeed, the ongoing capex boom ranks among the largest and fastest in history. Yet, like past periods of technological breakthroughs, there are likely to be rolling phases to this innovation cycle where the ultimate outcomes are highly uncertain. Indeed, even the leaders of the companies at the forefront of AI admit they don't know how this technology will impact the world.

Research from academics that builds on work from Carlota Perez and Joseph Schumpeter coalesces around the idea that major technological innovation waves tend to move through recurring stages: breakthrough, speculative investment, crash or reset, productivity lag, deployment, and mass diffusion. From our perspective, we're currently experiencing features of both the breakthrough and speculative investment phases. AI capabilities are advancing rapidly, but the physical buildout requires chips, memory, data centers, power, cooling, networks, etc. All of this buildout requires financing—and lots of it.

From our perspective, the outcomes—for society at large, the business sector, and financial markets—remain quite uncertain. Major factors include not only how the AI models themselves advance but how quickly data center and power needs are met, how the broader business sector is able to leverage the technology, and how it all gets paid for. Because of this uncertainty, we are tracking the broad AI innovation wave through scenario analysis that covers four potential scenarios: (1) a productivity boom, (2) delayed payoff, (3) an inflationary boom, and (4) deflationary bust.

We see core tensions revolving around the timing of these dynamics: significant capital is being spent well ahead of the value that's expected to be generated. AI's capabilities could continue to advance but if companies aren't able to translate these advancements into revenue growth and increased productivity, the payoff may be delayed or unrealized altogether.

The AI innovation wave is still in its early innings, and we expect the playing field to change as adoption, costs, revenues, and the models themselves continue to evolve. Based on the trends we see today, we believe the delayed payoff and inflationary boom scenarios are most likely. Productivity boom is an upside scenario, and deflationary bust is a downside scenario. However, the U.S. midterm election results and growing opposition to data center construction could potentially slow the pace of investment and adoption. This could raise the probability of a delayed payoff and a deflationary bust relative to other scenarios.

We're tracking these scenarios across a range of metrics that center on the following categories: capex, financing, adoption, and economic payoff/monetization.

AI infrastructure spending continues to rise, although the pace of growth may slow in 2027. Permitting delays, community opposition, and shortages of power, equipment, labor, and financing could constrain the buildout.

Hyperscalers are large cloud service providers that provide computing, storage, and networking infrastructure, including data centers and servers needed to support AI. Hyperscalers have funded most capacity and have lower cost of capital than smaller players, while neoclouds provide incremental capacity to hyperscalers, with less durable funding.

Capex could slow if supply catches demand, models become more efficient, competitors consolidate, or investor risk appetite weakens. Capex could accelerate if adoption broadens and users demonstrate higher revenue or lower costs.

Hyperscalers initially funded the AI buildout from operating cash flow, but heavy capex has dramatically reduced free cash flow (FCF), or operating cash flow less capex. This has pushed some companies toward debt, equity, leases, and structured financing. Because investment precedes revenue by years, continued expansion depends on investors' willingness to fund the gap in a leap of faith.

Debt and equity issuance has risen sharply, while higher yields, delayed or canceled initial public offerings (IPOs), and credit-rating pressure suggest markets are becoming more selective. Financing for neoclouds, which provide incremental capacity to hyperscalers and are smaller in nature, is another vulnerability: delayed capacity could cause hyperscalers to miss capacity and revenue targets. Alternatively, increased conviction about profit expectations could ease financing conditions.

AI usage is increasing as measured by tokens, a unit of data or compute when using AI models. Greater usage is aided by lower token prices and lower-cost open-weight models that encourage increased volume. Yet business token spending has recently leveled off, and adoption remains concentrated among early users. Data from the financial platform Ramp shows a wide spending gap between the top 1% of corporate adopters and the median company.

If adoption fails to broaden, demand may not absorb planned capacity. Conversely, wider production use and clearer revenue or cost benefits could support additional investment. New use cases or applications will likely be important to justify sustained spending.

Economic payoff, or monetization, is the hardest signpost to measure because adoption is early and investment occurs well before returns. Broad evidence of higher productivity or revenue remains limited beyond early applications, such as AI-assisted coding. In prior innovation cycles, profits from new technology didn't always accrue to those who put up the most investment capital.

Rising hyperscaler and model-provider margins would indicate improving returns on invested capital. Falling margins could signal higher costs or weak revenue, although margins could temporarily decline as new use cases and applications for the technology develop with a lag. Monetization is needed for investors to be repaid and for companies to be able to reinvest in capacity, models, and applications.

The investment stage of the cycle has created opportunities to the providers of chips, data centers, and power in the U.S. and emerging markets (EM), as well as select companies in Japan and Europe. In emerging markets, Taiwan and South Korea together comprise almost half of the EM index according to MSCI. Over time, leadership could shift to businesses that successfully use AI to increase productivity, expand margins, or create new sources of revenue. Changes in trends within signposts and signs of durability or vulnerabilities to earnings and corporate health will be important to monitor. As scenario probabilities change over time, equity market leadership may vary.

Evidence of broad paid usage, accelerating AI revenue, rising adopter margins, and measurable productivity would move our outlook toward a productivity boom. Slower than expected adoption, increased ability to finance capex—either through internally generated cash flow or via capital markets—and a lag in monetization would suggest a delayed payoff. Persistent bottlenecks, inflation and higher real yields would favor an inflationary boom. Weak utilization, widening spreads, cancellations, and impairments would signal that the capital structure is breaking before the technology adoption thesis plays out in a deflationary bust scenario. Until one of these patterns becomes decisive, we believe either a delayed payoff or inflationary boom are the most likely scenarios, but the U.S. midterm election outcome has raised the probability of a delayed payoff and a deflationary bust.

The AI investment cycle remains in its early stages, and its ultimate path is far from certain. Demand is still outpacing supply in parts of the AI ecosystem, but history shows that infrastructure booms can eventually produce excess capacity because spending decisions are made well before projects become operational and generate returns. AI may deliver substantial long-term advancements to society even if some of the companies funding today's buildout fail to earn an adequate return on that investment.

The U.S. midterm election outcome has added a new source of uncertainty to the AI cycle, with regulation and growing opposition to data center development potentially slowing the pace of investment and adoption. In our view, these risks, alongside market weakness in semiconductor and AI supply chain equities, raise the probability of a delayed payoff and a deflationary bust relative to our other scenarios.

For investors, this uncertainty reinforces the importance of diversification and selectivity. The companies and markets profiting from the initial surge in capital spending may not be the same ones that ultimately capture the greatest economic value. Leadership could shift from the providers of chips, data centers, power, and financing toward businesses that successfully use AI to increase productivity, expand margins, or create new sources of revenue.

Taken together, the signposts we've identified can help investors assess how the AI cycle is evolving. Conditions can change quickly; what looks like a roadblock, capacity shortfall, or slowdown has, at various points in this cycle, turned into renewed acceleration within weeks or months. A single month of data is unlikely to represent a durable trend. Rather than focusing on any one indicator, investors may want to watch whether the evidence across the signposts of capital spending, financing, adoption, and monetization increasingly points in the same direction.

Heather O'Leary, Senior Manager, Equity Research and Strategy, contributed to this report.