AI’s Next Test: Paying for the Promise
Corrado Tiralongo - Oct 23, 2026
AI remains a powerful growth story, but investors are asking a tougher question: can revenues and profits justify the spending behind the AI boom?
The recent sell-off in AI-related stocks reflects a more demanding question from investors: can the revenues justify the spending? That question has become harder to answer as borrowing costs rise, inflation limits central banks’ flexibility, and market leadership narrows. These pressures reinforce one another, leaving less room for companies to disappoint.
We believe this interaction is central to understanding the volatility. Confidence in AI’s long-term potential can remain intact while investors reassess the profits, financing and valuations attached to its development. A technology can change the economy without rewarding every company building it, or every investor financing the expansion.
The latest decline does not establish that the investment boom has ended. It reveals how dependent markets have become on its continuation.
A narrower foundation, a more expensive investment cycle
Headline indices were concealing considerable weakness before the latest sell-off. On October 6th, the S&P 500 reached a record high, yet the average company in its information technology sector remained more than 20% below its own peak. By October 15th, the Philadelphia Semiconductor Index was more than 20% below its mid-June high, while the S&P 500 remained within 2% of its recent record. Strength in a handful of large companies was masking a substantial adjustment elsewhere.
Narrow leadership does not make a reversal inevitable. Exceptional businesses can sustain exceptional results. But when market value and expected earnings growth depend heavily on the same companies, disappointment has consequences well beyond the business reporting it.
Valuations offer a mixed message. Technology’s forward price-to-earnings premium over the rest of the market has narrowed, but those earnings forecasts still need to be delivered. In “The Crash That Wasn’t” , Owen Lamont explores how equities can withstand higher real bond yields if expectations for future profits improve sufficiently. Stronger investment demand could itself contribute to higher rates. Market resilience is therefore consistent with a favourable AI outlook, but it also places more weight on that outlook proving correct.
Even then, economic success does not ensure investment success. AI could deliver substantial productivity gains while competition passes much of the benefit to customers through lower prices. The companies financing the expansion must generate enough demand and retain enough profit to reward the capital committed. Rapid adoption alone does not settle either question.
Recent developments have brought that question closer to AI’s financial foundations. By October 15, some of CoreWeave’s unsecured bonds were trading at distressed levels[1]. Separately, Reuters reported that Anthropic’s revenue continued to grow rapidly, but at a slower quarterly pace that disappointed some investors[2]. These developments raised questions about financing conditions and the demanding growth expectations underpinning the AI investment cycle.
The potential feedback is straightforward. Weaker revenues can make lenders and investors less willing to fund expansion. More expensive financing can make new projects less attractive. Slower spending then affects suppliers whose earnings forecasts depend on continued expansion.
A recent paper by Columbia Business School’s Stijn Van Nieuwerburgh, Financing the AI Buildout[3], shows why this exposure extends beyond debt issued directly by technology companies. Data centres and computing equipment are increasingly funded through leases, joint ventures, private credit and separately financed projects. These arrangements expand funding capacity, but substantial project borrowing can coexist with a strong technology-company balance sheet. Lease commitments and guarantees can keep that company economically exposed, making the distribution of risk harder to trace.
The paper also highlights a mismatch between asset lives and financing obligations. Buildings and power connections can remain useful for decades, while computing equipment may need replacement much sooner. Revenues must cover operating costs, financing and repeated hardware investment. Higher borrowing costs make that hurdle more demanding; faster technological change can raise it further.
Inflation and growth are pulling markets in different directions
The inflation backdrop compounds the problem. The unresolved U.S.-Iran war and disruption to energy supplies have kept pressure on fuel prices. We expect the global economy to withstand that pressure, but our outlook for easing inflation depends partly on energy prices declining in 2027. A prolonged disruption would delay that relief.
On October 14th, stronger-than-expected U.S. core inflation prompted investors to raise their expectations for Federal Reserve tightening. Treasury yields initially jumped and equities fell. Much of the yield increase subsequently reversed, but the initial reaction showed how sensitive markets had become to even a modest inflation surprise.
Persistent inflation would leave the U.S. Federal Reserve ("Fed") with less flexibility to respond to weaker growth. For businesses financing substantial expansion, that is an uncomfortable combination: revenue expectations face greater scrutiny while borrowing costs remain elevated.
What happened next was more revealing. Treasury yields fell sharply alongside equities despite the stronger inflation readings, suggesting that concerns about AI-led growth had begun to outweigh concerns about inflation. Earlier semiconductor sell-offs had often coincided with rising oil prices, leaving both stocks and bonds under pressure. This time, U.S. Treasuries provided a counterweight.
That distinction matters. When inflation dominates, stocks and bonds can decline together. If a substantial slowdown in investment becomes the greater concern, government bonds could benefit from weaker growth expectations and a reduced need for monetary tightening. Their behaviour depends on the source of the shock.
Corporate bonds face a different balance. Increased borrowing by AI-related companies has extended exposure to the investment cycle beyond equities. Falling government yields could support corporate bond prices, but wider credit spreads could offset some of that benefit. Borrower quality and financing structure therefore matter as much as the direction of interest rates.
A correction, with consequences worth watching
Our assessment is that a temporary correction remains more likely than an immediate, sustained unwind. We expect AI investment to remain strong through the first half of 2027, although an earlier slowdown is a meaningful risk. Continued revenue growth, available financing and easing energy prices could sustain investment and restore confidence.
Nor do the structures described in Van Nieuwerburgh’s paper establish that a financial crisis is approaching. They expand funding capacity and can distribute risk. The concern is whether leverage, concentrated demand and uncertain asset values could amplify losses if expectations disappoint.
A more adverse scenario would involve revenue disappointments leading to tighter financing and reduced spending. The consequences would extend beyond U.S. technology. Much of the hardware supporting the American AI buildout is imported, exposing Asian suppliers to a reversal. Falling equity wealth could also weaken U.S. consumption. A significant market adjustment need not cause a recession to affect earnings and growth.
Moving away from technology is not a complete answer. France’s recent experience shows how fiscal concerns can pressure sovereign debt, corporate financing and equities together. Different markets carry different vulnerabilities.
Canada faces a distinct balance. September employment fell by 68,000 and unemployment reached 6.5%, although the details were less weak than the headline suggested. Youth employment accounted for much of the decline,s, subdued wage growth and labour-market softness point to a different domestic backdrop from that facing the Fed.
Higher energy prices support parts of the Canadian economy, while trade uncertainty restrains others. Canadian and U.S. monetary policy need not follow the same path, a distinction that matters when assessing both growth prospects and fixed income.
For Canada Life Investment Management portfolios, we believe the priority is to identify exposures that depend on the same cash flows. Technology shares, data-centre property, infrastructure investments and private loans can all rely on a small group of tenants and continued growth in demand for computing capacity. Different investment labels do not necessarily provide different sources of resilience. The relevant questions are who ultimately pays, where the borrowing sits, and what happens to asset values if demand disappoints.
Higher real bond yields also change the portfolio comparison, a point Lamont emphasises. U.S. Treasuries offer a more credible alternative to equities than when real yields were near zero or negative. They may provide a counterweight if growth weakens, although further increases in yields would still put their prices under pressure. For equities, the implication is a higher hurdle: expected profits must justify both the price paid and the additional uncertainty relative to bonds.
In my opinion, the next stage of the AI cycle will place greater weight on revenues, cash generation and returns on investment. Spending announcements alone will provide less reassurance. We will be watching whether financial pressure remains confined to market prices or begins to change investment plans. That is where a correction in expectations could become a more consequential adjustment in economic activity.
Sincerely,
Corrado Tiralongo (he/him) Vice President, Asset Allocation & Chief Investment Officer Canada Life Investment Management Ltd.