By Ellen Hazen, CFA®, Chief Market Strategist
What We’re Watching in November
- Will the government re-open? The government shutdown is now the longest on record, the previous record being the 2018-2019 shutdown, which lasted for 35 days. In addition to many government agencies already closed, on November 1 additional funding stopped for programs including the Supplemental Nutrition Assistance Program (SNAP) and the Smithsonian museums
- Tariffs and Trade In other political news, the Supreme Court will hear arguments on the legality of the reciprocal tariffs. We believe that if these are struck down, President Trump will use other, well-supported methods to enact tariffs on the same order of magnitude. If they are ruled invalid, there may be a period of time in which importers are reimbursed for tariffs previously paid. President Trump continues to negotiate new trade deals with various countries; recently the US agreed to hold off on 100% tariffs while China agreed to hold off on rare earth export controls
- Economic Data Although the government was shut down in October, some economic data did come out, albeit late.
Inflation – The Consumer Price Index (CPI) and the Producer Price Index (PPI) were released on October 24 at 3.0% and 2.9% respectively, continuing the trend of above-target inflation. We expect inflation to remain higher for longer, supporting our allocation to Treasury Inflation-Protected Securities (TIPS)
Labor market – The Bureau of Labor Statistics did not release nonfarm payroll data, but we did get private nonfarm payroll data from private company ADP. In September, ADP reported 32,000 fewer jobs, providing additional evidence that the labor market is weakening. Interestingly, a recent paper by a Federal Reserve Bank of Dallas economist suggests that the breakeven rate of monthly job growth – the rate at which the unemployment rate neither rises nor falls – has fallen from over 200,000 in 2023 to only 30,000 now as a result of slower population growth due to both demographics and immigration
- Federal Reserve meeting in December The next Fed meeting is in December; after October’s 0.25% cut, the market is giving a 65% chance of an additional 0.25% cut at December’s meeting. The Fed also announced that starting December 1, it will stop letting Treasury securities roll off the balance sheet (a process known as Quantitative Tightening, or “QT”). Both of these could further ease financial conditions
- Credit quality A number of high-profile bankruptcies have raised the specter of possible broader credit deterioration. Two of the more notable failed companies were subprime automobile finance company Tricolor and automobile parts supplier First Brands, resulting in losses of hundreds of millions of dollars. Banks from JP Morgan to Fifth Third reported losses; more losses have been reported to be in the private credit realm
- Earnings Growth Continues Strong Approximately 75% of S&P 500 companies have reported earnings, with average earnings growth of 12% above expectations. The highest growth has been in Technology, Financials, and Industrials. Currently, analysts expect 7% growth for Q4 2025 and 12% for the full year 2026
|
Investing IS uncertainty. Anyone who says differently is selling something. |
Many clients are asking whether artificial intelligence spending and/or stock prices are in a bubble. Investors remember well the Internet bubble of the late 1990s, which burst in early 2000 and resulted in an 80%+ decline of the Nasdaq-100 Index. At that time, the new technology was the Internet and the overinvestment took place in fiber optic cable by companies like Global Crossing, Qwest, and Level 3 Communications. Most of these were funded with debt (debt-to-equity was typically 2x – 4x) and they built for demand that did not immediately materialize. Unfortunately for them, the Fed raised interest rates six times between June 1999 and May 2000, rendering them unable to meet their obligations. At that point, the companies – and their stock prices – collapsed.
Let’s be clear: The quantity of money being spent today on AI is very large, and is quickly rising. Just the four largest datacenter providers – called “hyperscalers” – are spending $349 billion in 2025, up over 50% from $217 billion in 2024, and this is expected to grow to $464 billion in 2026. Although this is large, in the context of other large infrastructure rollouts, this size does not yet raise alarms as a percentage of GDP, as I will discuss below.
The media is full of articles claiming that AI spending is indeed a bubble, however, it is also full of articles claiming that it is not. There are a lot of moving pieces and the industry is rapidly evolving, making certainty elusive. It is easy to cite evidence that suggests that AI spending is in a bubble. It is also fairly easy to do the opposite. We aim to provide a balanced perspective, recognizing that we will likely not know whether current spending is too high for some time. Our best judgment is that while spending on AI datacenters may ultimately result in a bubble, spending is not yet in a bubble.
1. Evidence that artificial intelligence spending is in a bubble
Spending is growing rapidly. As noted above, capital spending just for the four large hyperscalers (Amazon, Alphabet, Microsoft, and Meta) is in the hundreds of billions of dollars and has grown over 50% for each of the last two years. The estimates keep increasing: Two years ago, analysts expected Meta to spend $31 billion in 2027; today, they expect it to spend $119 billion. Capital spending estimates (2027) for the other companies have likewise doubled or tripled over the past 2 years.
Companies are not yet seeing return on investment. MIT Media Lab recently published a survey in which 95% of companies reported that generative AI pilots generated no return on investment.
Funding is changing. Although initially funded out of existing cash flow, some new AI spending is being funded by debt or by vendor financing. The large hyperscaler companies have easily funded their very large datacenter spending out of cash generated from their operations; these are some of the largest, most profitable companies in the world. In 2027, for example, Alphabet is expected to generate cash from operations of $221 billion and spend only $130 billion on capital expenditures, leaving ample free cash flow for other purposes. (Note that this $130 billion was expected to be only $68 billion at the beginning of 2025). However, other players are taking on debt or vendor financing to afford the enormous spending. OpenAI, which developed ChatGPT, has signed a dizzying array of financing agreements with NVIDIA, AMD, Oracle, Microsoft, and CoreWeave. In the case of AMD, AMD gave OpenAI the rights to own 10% of AMD in exchange for OpenAI agreeing to use AMD’s chips for 6 gigawatts’ worth of power. OpenAI clearly cannot afford the multiple hundreds of billions of dollars of purchase commitments it has made on its existing revenue, which is about $15 billion a year. Oracle recently issued $18 billion in debt to help fund its spending; Meta issued $25 billion last week.
So, spending is large, it is growing, there is no positive return on that investment (yet), and it’s increasingly being funded by debt or “circular” vendor financing arrangements. It’s clearly a bubble, right?
2. Evidence that artificial intelligence spending is not in a bubble
Most spending is still out of existing cash flow. Although some of the incremental spending is from debt or circular arrangements, the majority is out of existing cash flow. Moreover, an additional source of demand is sovereign nations. Goldman Sachs estimates that countries will account for close to $50 billion of spending just on NVIDIA chips within two years.
Investors are not buying the AI stocks simply because they’re going up. In the famous Dutch tulip craze of the 1630s, tulip bulb prices vastly outstripped intrinsic value, and investors bought them purely with the belief that they could be sold to somebody else at a higher price. Today, stocks are primarily increasing because companies’ earnings are increasing, although there has also been some expansion of companies’ price-to-earnings multiples. A National Bureau of Economics paper studied stock price behavior in periods of technological change and found that the companies spending money are acting rationally, not irrationally, even though the spending may look irrational and bubble-like.
Corporate operating margin expansion may ultimately justify AI spending. Research provider Empirical Research estimates that between replacing some knowledge workers, using AI to algorithmically price goods and services, and getting some companies and users to pay subscriptions for AI services, operating margins could expand by 0.50% in aggregate even after paying for the capital spending on AI datacenters.
The absolute size of spending is modest compared to prior infrastructure buildouts. The amount of money spent on AI in 2025 is estimated at $406 billion, which is approximately 1.3% of GDP. The railroad buildout in the late 1800s peaked at 5.4% of GDP, the housing bubble in 2008 peaked at 3.5% of GDP, and the internet bubble of the late 1990s peaked at 4.3% of GDP.
So is AI spending in a bubble? We don’t know. Given that it has only been 3 years since ChatGPT rocked the world, that companies are still very much in the experimentation stage, and that spending continues to rapidly increase, we believe it is early in the rollout and penetration of artificial intelligence. We are watching a variety of indicators that we believe may provide early warning signals for when demand starts to slow. These include cash flow at the large spenders, prevalence of vendor financing deals, orders and backlogs for chips and for electricity, datacenter capacity utilization, debt and equity raises to fund infrastructure construction, and power prices, among others.
Even when demand slows, we believe AI will be transformative, increasing productivity in areas from medical diagnoses to self-driving cars to software development. The most successful internet companies of today, like Meta and Alphabet, were enabled by all the optic fiber that was laid during the internet bubble. Similarly, we expect that entirely new business models may arise out of today’s AI infrastructure buildout.
Investing in AI today can feel uncomfortable, because there is so much uncertainty about both the pace of spending and the ultimate return on that spending. At the same time, many companies can easily afford to keep spending, and given how the market has rewarded those companies, not investing in those companies can also be uncomfortable. At the end of the day, investing IS uncertainty, and as the Dread Pirate Roberts would say, anyone who says differently is selling something.