Over the past several years, we have written often about the growing risks beneath the U.S. economy and financial markets. Those discussions have included the possibility of recession, the effects of higher interest rates, weakening household finances, narrow market leadership, growing government debt, and the unusual amount of capital being directed toward artificial intelligence. As each new set of data became available, our goal was not to predict the exact date of the next downturn. It was to understand whether the pieces that normally appear before one were beginning to fall into place.

That work eventually led us to create the Canary Dashboard. The Dashboard gave us a better way to organize many different economic and market signals instead of relying on one indicator or one forecast. It brought together valuations, market breadth, consumer credit, employment, corporate credit, liquidity, monetary policy, interest rates, global markets, and other measures of financial stress. We have discussed the Dashboard with investors before, and each quarter we have used it to ask the same basic question: Is the system becoming safer, more fragile, or beginning to break?

The answer this quarter is important because the story has changed again.

The warning signs we discussed earlier in the year have not disappeared. In several areas they have become stronger. But the risks are also moving. Earlier concerns centered heavily on expensive markets, weak consumers, narrow market leadership, and the possibility that higher rates would eventually slow the economy. Today those concerns remain, but a new connection is becoming clearer. The cost of capital itself is starting to place more pressure on the areas that have done the most to support economic growth and market returns, especially artificial intelligence investment, long-duration assets, and private-sector hiring.

At the same time, we still do not see the broad signs of financial failure that would normally tell us a recession or major bear market has fully arrived. Corporate credit markets remain open. The unemployment rate remains low. Consumers are still spending. The Federal Reserve continues to describe economic activity as expanding at a solid pace. Federal Reserve This is why the current environment remains difficult. The surface of the economy continues to look fairly strong while the pressure underneath it is becoming harder to ignore.

Our proprietary Canary Dashboard now stands near 7.8 out of 10, compared with a lower reading earlier in the year. We describe this as late fragility with partial confirmation. The exact score matters less than the change underneath it. The most important development this quarter is that the risks are beginning to connect in ways they had not before.

The clearest example is the bond market. The 10-year Treasury yield has moved above 5%, reaching roughly 5.15% as of September 24. The Federal Reserve also raised its target range for short-term rates by another quarter point on September 16, bringing the range to 3.75% to 4.00%. The Fed said inflation remained elevated even though economic activity continued to grow. Federal Reserve In earlier quarters, we viewed high rates mainly as a future risk. Today they are becoming a current operating cost for households, businesses, the federal government, and investors.

This is an important change. A 5% Treasury yield does more than make bonds more attractive. It changes the price of money throughout the economy. Companies must earn a higher return before a new project makes sense. Buyers of homes face higher financing costs. Commercial real estate projects become harder to justify. Private-equity deals require stronger cash flows. Investors have less reason to pay very high prices for profits that may not arrive for years.

We have discussed the experience of 2022 many times because it showed how quickly rising interest rates can change stock-market values. The comparison is not exact. In 2022, rates were rising from extremely low levels. Today we are already starting from a much higher base. That may actually make the next move in rates more important, not less. The stock market may not need another rise of several percentage points to feel pressure. From current levels, a move in the 10-year Treasury toward the mid-5% range could create a much harder test for growth stocks, housing, corporate borrowing, and capital spending.

The reason this matters so much today is that the U.S. economy has become increasingly tied to a very large investment cycle in artificial intelligence.
We have been writing about AI spending for some time. The question has never been whether AI is an important technology. We believe the evidence already shows that it is. The more important investment question has been whether the amount of money being spent to build AI infrastructure can earn a return high enough to support the valuations and borrowing behind it.

That question has become more urgent this quarter.

The largest technology companies continue to spend extraordinary amounts on chips, data centers, power, networking, and cloud infrastructure. Some of these firms have enormous cash flows and very strong balance sheets. That is one of the major differences between the current AI boom and many of the companies that defined the dot-com period. Yet even strong companies face limits. Reuters has reported that the combined capital spending of several of the largest technology companies could eventually exceed their combined free cash flow if current plans continue. That does not mean the companies are running out of money. It means investors are beginning to ask whether the next dollar of AI spending will earn enough to justify the cost.

That question is now appearing in the bond market as well. Oracle provides one of the clearest examples. Roughly $18 billion of debt tied to its Project Jupiter data center project in New Mexico has recently traded below face value, around 89 to 91 cents on the dollar. Investors have also become more concerned about Oracle’s growing debt load, while efforts to distribute some of the project financing have become more difficult. Reuters We do not view one troubled financing package as proof that the AI cycle is ending. We do view it as evidence that lenders are no longer treating every major AI project as if capital were unlimited and risk did not matter.

The broader credit market is showing a similar change in tone. Large technology firms have issued roughly $220 billion of bonds over the past year to help finance AI expansion, and the sheer amount of supply has begun to affect how the market prices even high quality borrowers (like Treasury securities). At the same time, new financing structures are being used to support enormous data center projects while keeping some obligations away from the main corporate balance sheet. The Financial Times recently reported that technology companies are using guarantees connected with as much as $300 billion of AI-related exposure held through outside financing vehicles. Financial Times These structures may work exactly as intended. Still, their growth tells us something important: the AI investment cycle has become large enough that companies are searching for increasingly creative ways to fund it.

This is where our earlier concerns about free cash flow become more important. A company can report strong revenue and still face pressure if spending rises even faster. It can also remain profitable while the return on new projects falls. That is why we have been careful not to confuse AI adoption with AI investment returns. The technology can be successful while investors still lose money if too much capital is spent too quickly or at too high a price.

This quarter has added another piece to that discussion: the technology itself may face limits on how quickly it can be pushed forward.

Leaders at Anthropic and OpenAI have publicly discussed the need for stronger controls around advanced models as concerns have grown about AI systems being used in cyber attacks or behaving in ways developers did not fully expect. These comments do not mean AI development is stopping. Competition between the largest firms remains intense, and companies still have strong reasons to keep investing. But the discussion has changed. Safety testing, regulation, security, and the pace of model development are becoming financial issues as well as technical ones.

That matters because much of the current AI investment case assumes rapid progress. If model development takes longer, if safety rules increase costs, or if regulation delays deployment, some expected revenue could arrive later than investors currently assume. When Treasury yields are above 5%, delays matter more because future cash flows are worth less today. This gives us a new connection in the Canary Dashboard: higher interest rates can pressure AI economics at the same time that safety and regulatory concerns may slow the speed at which companies can turn investment into revenue.
The semiconductor market may already be reacting to some of this uncertainty. The sector had an extraordinary run earlier this year as investors priced in continued AI spending. More recently, semiconductor stocks have become much more volatile. We would not call one correction proof that an AI top has formed. Markets often pull back after very strong gains. What matters is whether weaker semiconductor leadership begins to occur at the same time that financing spreads widen, projects are delayed, and companies become more careful with capital spending. If those events continue to line up, the market may be telling us that the AI capital cycle is entering a different stage.

This is where the comparison with the late 1990s remains useful, but only if we use history carefully. The lesson from the dot-com period is not that a major IPO or a fast-rising technology automatically causes a crash. The internet was real. Its use continued to grow even after the market fell. The problem was that expectations and capital spending moved much faster than the profits needed to support them.

That distinction remains central to our work today. We are not arguing that AI is a bubble simply because the technology is popular. We are asking whether investors have assumed a level of growth, spending, and future profitability that leaves too little room for setbacks. The recent pressure in AI financing, combined with very high Treasury yields and greater safety concerns, means the answer to that question has become more important over the last quarter.

Consumer Strain

The consumer side of the economy tells a similar story of strength mixed with fragility. Earlier this year, we wrote about the K-shaped economy and the growing difference between households that own assets and households that depend mostly on wages. That divide remains.

The latest New York Federal Reserve Household Debt and Credit Report shows total household debt at approximately $18.8 trillion. Credit card balances increased to about $1.26 trillion, auto-loan balances rose to $1.71 trillion, and home equity borrowing also increased. At the same time, the report showed that overall delinquency rates changed little, and serious credit card delinquency transitions remained broadly steady. Federal Reserve Bank of New York This is important because it does not support the idea that consumers are currently falling apart. Instead, it supports the view we have been developing for several quarters: many households remain under pressure, but employment is allowing them to keep the system functioning.

That is why the labor market has become such an important part of the Dashboard.

The August unemployment rate remained only 4.1%, while total payroll employment increased by 162,000. On the surface, those figures still look healthy. Yet the details continue to show that job creation is becoming less broad. Food services and local government education were major sources of August job growth, while the information sector lost 23,000 jobs. Health care continued to add workers, although at a slower pace than during the prior year. Bureau of Labor Statistics

The six month picture reinforces that concern. Health care and related services have accounted for a large share of new private sector jobs, while some higher paying fields such as information and financial activities have lost employment. Government hiring is not the main reason total employment remains positive, which is an important correction to one possible reading of the data. The larger issue is that private-sector hiring itself is becoming concentrated in fewer industries.

We have therefore begun describing the labor market as low fire, low hire. Employers are not laying off workers in large numbers, but many are also becoming more careful about adding staff. That can create a long period in which the unemployment rate looks stable even though it becomes harder for someone who loses a job to find the next one.

This matters directly to the K-shaped consumer. A family with strong savings and investment income may be able to handle several months without a paycheck. A household carrying revolving credit card debt may not. The lower and middle parts of the K can keep spending longer than many forecasts assume as long as employment remains stable. Once employment weakens, however, the lack of savings can turn a gradual slowdown into a faster one.

This is why our view of consumer stress has changed slightly from earlier quarters. We are no longer looking primarily for an immediate collapse in spending. The latest Fed data do not show that. We are looking at fragility. Household balance sheets can remain stable for a long time when paychecks continue to arrive. The danger is that many households may have very little ability to absorb a true employment shock.

That brings the different pieces of the Dashboard together.

Higher Treasury yields raise corporate financing costs. Higher financing costs make new projects harder to justify. Companies become more careful with spending and hiring. Weaker hiring makes the labor market less forgiving. Households with little savings become more dependent on credit. If job losses then rise, consumer spending weakens. Lower spending puts pressure on company profits. Weaker profits can lead to more cuts in spending and employment.

For several years we have been watching these risks as separate Canaries. This quarter, the reason for our increased concern is that we can begin to see how they might connect.

Importantly, that process has not completed.

Broad corporate credit is not yet showing crisis conditions. Money markets and financial system liquidity are not showing the type of stress that normally appears during a serious funding event. The New York Fed’s household data show broadly stable delinquency trends rather than rapid deterioration. The Federal Reserve still sees solid economic activity. Payrolls are still growing. 

Those are not small details. They are the reason we continue to resist making a simple recession call based only on a high Canary score.

Our work over the years has taught us that being early and being wrong can look very similar for a long time. Expensive markets can become more expensive. Consumers can remain under pressure for years while continuing to spend. Companies can fund investment booms much longer than expected. Higher rates can remain high without immediately causing a recession. The purpose of the Dashboard is to keep us from turning any one of those risks into a prediction before the evidence supports it. The change this quarter is that the cushion appears thinner.

Our current score near 7.8 reflects an environment in which valuations remain high, market leadership is narrow, Treasury yields are putting more pressure on capital decisions, household finances remain stretched, job growth is becoming less broad, and the AI investment cycle is facing a more serious test from financing costs and safety concerns. At the same time, employment, profits, credit markets, and financial liquidity continue to provide support.

That combination is why we describe the current period as late fragility rather than crisis.

As we move into the final quarter of 2026, several developments will tell us whether this thesis strengthens or weakens. The first is the Treasury market. If the 10-year yield moves further into the mid-5% range, we believe the pressure on stocks and business investment would increase meaningfully. The second is private sector hiring. If hiring continues to weaken while unemployment claims begin to rise, the consumer could become a much more important risk. The third is corporate credit. If the widening that is now visible in parts of AI financing spreads into the broader high yield and investment grade markets, we would view that as much stronger confirmation. The fourth is AI capital spending itself. Individual project delays are worth watching, but broad cuts or slower guidance from the major technology companies would carry much more weight.

We will also continue to look for evidence that could prove the more cautious view wrong. A sustained decline in inflation could allow Treasury yields to retreat. Strong profit growth could justify current stock valuations. AI projects could begin producing cash flow faster than expected. Hiring could broaden. Household savings could improve. Credit markets could remain calm. If these developments occur, the Canary Dashboard should respond by moving lower.

That is the purpose of the framework.

Our earlier recession oriented research was meant to identify growing risks. The Canary Dashboard was built to make that process more disciplined. The quarterly reports are now the record of how those risks are actually evolving.

This quarter, the message is not that the recession we have discussed in prior years has finally arrived. It has not.

The message is that the path by which one could arrive has become easier to see.

Higher interest rates are no longer simply an abstract valuation problem. They are beginning to affect real financing decisions. AI is no longer simply a story about demand. It is becoming a story about return on capital. Consumer stress is no longer simply about inflation. It is becoming increasingly tied to employment stability. And the labor market is no longer simply about the unemployment rate. It is increasingly about whether enough industries are still willing to hire.

These are changes in degree, not proof of a turning point.

But those changes are exactly why we built the Canary Dashboard in the first place.
Our task remains the same as it has throughout this series of research: participate in economic and market growth while paying close attention to the signals that tell us risk is changing. Markets rarely send one clear warning before conditions turn. More often, the pieces begin moving slowly, then suddenly start moving together.

As of September 24, we believe more of those pieces are moving in the same direction than they were one quarter ago.

That deserves our attention. It does not require our panic. And it is the distinction between those two ideas that will continue to guide our portfolio decisions as we move into the final quarter of the year.

The information contained in this article is for information purposes only. Consult your personal advisor before making any investment decisions.

The Canary Dashboard is an internal research framework used to organize economic and market data. Its score is not a statistical forecast and does not predict future market returns. Investing involves risk, including possible loss of principal. Past performance is no guarantee of future results. Data and views in this report reflect information available through September 24, 2026.