High Expectations for Artificial Intelligence
At one time the phrase “A watched pot never boils” meant time seems to pass more slowly when one is waiting for something to happen. The time waiting also allows expectations to grow and I remember when the event I had been waiting for finally arrived, it was underwhelming since my expectations had grown so big. I grew up understanding that “A watched pot never boils” meant by paying attention an event that wasn’t desired could be avoided. One would remove the pot from the burner before it boiled over and created a mess. This interpretation of the phrase seems appropriate when evaluating the fervor surrounding Artificial Intelligence and everything related to it. This entails watching the soaring demand for AI chips, data centers, electricity, the impact of AI spending on economic growth, and the influence AI stocks are having on the S&P 500.
JP Morgan identified 41 AI related stocks that now account for 47% of the S&P 500’s total capitalization, even though they only represent 8% of the 500 stocks. In March 2023 those 41 stocks comprised 26% of the S&P 500, so their weighting has almost doubled. On October 28 the S&P 500 closed at a new all-time high after gaining +0.23%, but only 104 stocks were up and 396 closed lower because AI related stocks were up. Since 1990, the S&P 500 has never had weaker breadth on a day that it closed positive. The Equal Weight S&P 500 (each stock has a .2% weight) is underperforming the S&P 500 by the largest margin in more than 22 years.
ChatGPT was released on November 29, 2022. Since the introduction of ChatGPT, the value of the S&P 500 has increased by $32.5 trillion. Of that total, the 41 AI related stocks increased by $25.0 trillion and account for 70% of the total increase in capitalization, while the other 459 stocks added $7.5 trillion.
The “Magnificent Seven” (Apple, Amazon, Alphabet (formerly Google), Meta, Microsoft, Nvidia, and Tesla) companies now make up nearly a record 37% of the S&P 500’s total market capitalization and have consistently outperformed the index. All of these companies are great and have grown rapidly over the last decade, and they all currently have a market cap of more than $1 trillion. The S&P 500 performed well from the end of 2015 through 2024, with an overall return of 178%. However, it doesn't come close to the Magnificent Seven, which had a 698% combined return over that same time span.
Since 2010 the percent of assets that are invested in passive ETFs and mutual funds that track the S&P 500 has soared from 19% to 51.6%. In September 2025 there was $18.0 trillion in passive funds versus 16.87 trillion in active funds. Investors are drawn to the low management fees without knowing the inherent lack of diversification they’re getting with Mag 7 and AI concentration being so high. The strength that is a plus now is creating a vulnerability that will burn passive investors when a more sober reality takes hold.
There is another old saying that is applicable to the current impressive dominance of the AI stocks. “All good things must come to an end”. Imbedded in this saying is a second important truth “All bad things must come to an end”. Combined, these expressions summarize life on planet earth: All things must pass and are only temporary like a fleeting memory. Sooner or later, the narrative around AI will become less fantastical, and disappointment will set in.
The Dot.com top in 2000 developed and surprised the majority of investors because they were so caught up in the bullish narrative they weren’t watching the pot. The same thing occurred in 2007 prior to the collapse in the Housing bubble. There were plenty of warning signs, which I wrote about in 2007, but the majority of investors failed to see the problems that were hiding in plain sight. The same thing is setting up to occur with Artificial Intelligence. Most investors will hang on to the AI narrative and remain bullish even as the AI story unwinds.
As I wrote in the September Macro Tides, “I don’t think the Peak of Inflated Expectations for AI have been reached yet so the stocks involved can be expected to gain altitude.” I still think that applies, so in coming months AI can be expected to continue to add to GDP growth, S&P 500 earnings, and help the S&P 500 make higher highs. What I’m trying to do is identify the potential weak spots in the AI narrative knowing they will emerge sometime (probably in 2026) when most investors are least expecting it. I’m watching the AI pot since it is a big deal.
If You Build It, They Will Come
In the movie ‘Field of Dreams’ the owner of a farm growing corn is faced with a decision whether to sell his farm or let the bank foreclose the next day. As Ray Kinsella is trying to decide, Terence Mann a famous author tells him, "Ray, people will come Ray... they'll come to Iowa for reasons they can't even fathom. They'll turn up your driveway not knowing for sure why they're doing it." Ray decides to keep the farm and the final scene shows a line of car lights as far as can be seen coming to see the Field of Dreams.
I think the phrase ‘If you build it, they will come’ aptly describes where the Artificial Intelligence (AI) cycle is at. Companies will spend trillions of dollars to build the infrastructure needed to provide the computing power and electricity to achieve Artificial Intelligence’s potential. In the last three years the Big Four (Microsoft, Amazon, Meta, and Alphabet) have spent more than $800 billion on AI, but in 2025 only $20 billion of revenue has been generated. Despite the huge mismatch between expenditures and revenue, the race to build data centers and energy sources will continue. The belief underpinning this enormous spending is that customers and revenue will flow, if they build all the infrastructure first. As I wrote in the September Macro Tides, “I don’t think the Peak of Inflated Expectations for AI have been reached yet so the stocks involved can be expected to gain altitude.”
The surge in AI spending is so big that it is lifting the overall economy. Goldman Sachs estimates that $368 billion was spent on AI through August 2025, and will total $500 billion in 2025. The spending is concentrated on building data centers and power supply so the driver is construction and equipment orders. According to the Bureau of Economic analysis (BEA), GDP totaled $30.486 trillion as of June 30, 2025, and may have reached $31.5 trillion at the end of September. Spending on AI represented about 1.35% of GDP through the first nine months of 2025, and that contribution will increase since AI spending is growing much faster than GDP. As a share of GDP, investment in information processing equipment and software is 4%, but was responsible for 92% of GDP growth in the first half of this year, according to economist Jason Furman. GDP excluding these categories grew at a 0.1% annual rate in the first half of 2025.
Even if GDP growth excluding information processing equipment and software was higher than 0.1%, the concentration shows that GDP growth hasn’t broad based in 2025. This leaves the economy vulnerable when the growth rate in AI spending slows. At the end of 2023 the growth in annual AI spending was less than 10%. By the end of 2024 annual spending growth soared to 70%, and continued to grow by more than 70% through the third quarter, according to Goldman Sachs. As the law of big numbers takes hold, Goldman Sachs thinks the growth rate in spending will slow to 42% at the end of 2025, and drop to less than 20% by the end of 2026. Spending on AI will be running at an annual rate of more than $525 billion at the end of 2025 so an increase of almost 20% in 2026 will lift AI spending to more than $600 billion in 2026.
The slowing in the annual rate of change in AI spending will pose a test for AI investors next year, since valuations are so high and built on a growth rate that’s unsustainable. As the incremental increase in spending slows in 2026, investors will have to adjust to a different narrative that’s a little less positive. Rather than having a one-sided positive story investors will be provided a reason to buy less and maybe do a little selling. The slowing in AI spending will also provide GDP a smaller lift and only lift GDP by 0.8% in 2026 versus 1.5% in 2025. Some of the drag on GDP growth could be offset by an increase in productivity that could be as much as +0.3%. The slowing in spending on AI is more of a risk for investors in AI related stocks than it is for the economy.
The initial surge in AI spending by Mega Cap companies (Microsoft, Amazon, Meta, and Alphabet) was funded by the enormous cash flow these successful firms generate every year from their existing businesses. The average company in the S&P 500 will spend 10% to 15% on capital investments. Tech companies spend more, as was the case when these four companies were spending around 30% of their cash flow in the 5 years after 2012. In 2017 and 2018 Capital expenditure spending increased to 40% of cash flow and held at that level through 2021. Spending jumped to 55% in 2022 as the economy reopened after the Pandemic, and then dropped back to 40% in early 2024, after the Federal Reserve aggressively increased the policy rate. As of June 30, 2025 these 4 companies plus Oracle spent 60% of their cash flow on AI which is astounding.
As spending by these companies and other players ramped higher in 2025, cash flows weren’t enough to fund all the costs. According to Goldman Sachs, $141 billion of this year's $500 billion AI capital expenditure came from corporate debt. In the next 3 years Morgan Stanley estimates that $2.9 trillion will be spent globally by hyper scalers on AI, with $1.4 trillion coming from cash flow and $1.5 trillion having to be financed. As long as the long-term outlook for AI is positive, finding $500 billion each year for the next three years is probably doable from institutional investors. Pension funds, sovereign wealth funds, insurance companies, endowments, private equity, and high net worth retail investors will be willing to be a part of the next Big Thing.
The amount of money being invested in AI is unprecedented and some of the money will prove to have been misallocated. During the Dot-com bubble from 1995 to 2000, telecommunication companies spent an estimated $500 billion to $2 trillion building 80 to 90 million miles of fiber-optic networks. This massive, debt-fueled spending was based on speculative projections of explosive internet traffic growth that failed to materialize as quickly as forecast. Demand for internet traffic grew more slowly than anticipated, resulting in a devastating glut of unused fiber-optic cable. In 2001 there were estimates that up to 95% of the new fiber was unused. That changed as the growth in cloud computing, video streaming, and the high-speed internet absorbed all of the unused fiber optic in the next decade. The delay in the utilization of the fiber optic networks was devastating for investors. Telecom stocks lost an estimated $2 trillion in market value after investors fled this overvalued sector after it peaked in 2000.
A report from the Massachusetts Institute of Technology almost sounds like an echo of the fiber optic story, as discussed in the September Macro Tides. “In its report entitled “The GenAI Divide: State of AI in Business 2025”, MIT concluded that while AI continues to spark excitement and attract enterprise investment, the vast majority of initiatives aimed at driving rapid top-line growth are stalling. Although Adoption is high, transformation is rare and roughly 95% of organizations who are using AI are seeing no measurable impact on profits. Individual productivity has benefitted from general tools like ChatGPT and Copilot, but it hasn’t translated into an increase in productivity at the company level. Companies are launching internal pilot programs, but MIT found that only 5% actually make it into production.”
Roughly 3.5 billion people use Google Chrome as their internet browser, which gives Google a market share of about 65%. I use it and I’ve never paid a dime for the convenience and access Google Chrome provides in hunting down the charts and statistics used in the Weekly Technical Review and Macro Tides. In 2025 companies will spend $500 billion on the potential of Artificial Intelligence but will realize less than $20 billion in revenue. Hyper scalers are expected to spend another $500 billion in each of the next three years. In order to generate a return on the huge sums being spent on the AI infrastructure, someone will be asked to pay it. Unless more than 75% of companies can realize an improvement to their profits from using AI, fewer companies will use it. Eventually all of the fiber optic cable capacity was absorbed, but the delay in adoption was fatal for a number of the companies that had been fiber optic darlings.
Analysis from Macro Strategy Partnership, using Swedish economists Knut Wicksell’s theories, calculated the scale of investment versus actual economic returns for AI. This allowed Macro Strategy Partners to compare the Dot.Com and Housing bubbles and the money being spent on Artificial Intelligence. They found that the AI bubble is 17 time larger than the Dot.com bubble and 4 times the Housing bubble that led to the 2008 Financial Crisis. At a minimum, this analysis indicates that the there is an Artificial Intelligence bubble and a misallocation of capital.
The Gold Rush took hold in 1848 and 1849 and attracted almost 300,000 people who hoped to strike it rich panning for gold. Few of the dreamers made money from mining gold, but the firms that sold the picks, shovels, and pans did great. Levi Strauss made a fortune selling the pants that could hold up to the harsh conditions miners endured. Nvidia is a great company and is at the center of the AI Gold Rush, since data centers need the computing power Nvidia’s equipment provide. Nvidia’s success has rewarded investors. Nvidia's $4.6 trillion market cap now exceeds the combined value of every publicly traded bank in the US and Canada at $4.2 trillion. According to Deutsche Bank, Nvidia’s market capitalization is now bigger than the entire stock market capitalizations of Britain, France, and Germany. Only China, India, and Japan have stock market capitalizations larger than Nvidia. High valuations have been afforded to many companies associated with AI. According to CB Insights, more than 1,300 AI startups are valued at more than $100 million, and 498 companies sport a valuation of $1 billion or more. After the coming shakeout only a few of these companies will become long time players.
The cost of building out data centers is so high that competitors of Nvidia and the other major chip majors will endeavor to come up with a less expensive approach that still provides users what they need. On October 20 Alibaba announced it had developed a new approach to maximize the potential of existing and older chips. During a multi-month beta test inside its Model Studio marketplace, Alibaba Cloud claims its new Aegaeon pooling system reduces the number of Nvidia GPUs required to serve Large Language Models by 82%. Aegaeon is an inference-time scheduler designed to maximize GPU utilization across many models and virtualizes GPU access at the token level. This allows it to schedule tiny slices of work across a shared pool, and allows one H20 could serve several different models simultaneously. The effective output rose by as much as nine times compared to older serverless systems. During the beta test, the number of GPUs needed to support dozens of different LLMs, ranging in size up to 72 billion parameters, fell from 1,192 to just 213, a decline of 82%. It will take time but this is just the first example of what competitors will do to compete by saving money for users of high-end GPUs and computers.
Open AI launched its Chat GBT program in November 2022 which arguably also launched the fervor surrounding Artificial Intelligence. OpenAI is a private company and is valued at $500 billion. In the first half of 2025 OpenAI generated $4.3 billion in revenue and lost $13.5 billion.
Despite such dismal finances OpenAI has initiated $1 trillion in deals in 2025. This is how a technology analyst framed the OpenAI spending spree after OpenAI announced its deal to buy $300 billion of services from Oracle over the next 5 years. “Michael Cembalest from JP Morgan follows Oracle offered this appraisal. “Oracle’s stock jumped by 25% after being promised $60 billion a year from OpenAI, an amount of money OpenAI doesn’t earn yet, to provide cloud computing facilities that Oracle hasn’t built yet, and which will require 4.5 GW of power (the equivalent of 2.25 Hoover Dams or four nuclear plants), as well as increased borrowing by Oracle whose debt to equity ratio is already 500% compared to 50% for Amazon, 30% for Microsoft and even less at Meta and Google."
OpenAI has more than 800 million weekly users up from 500 million in March which sounds impressive. However, only 5% of its current user base actually pays OpenAI $20 a month. Furthermore, Deutsche Bank’s dbDataInsights team tracks consumer spending data across France, Germany, Italy, Spain, and the United Kingdom. Since May 2025, monthly spending growth on ChatGPT has slowed sharply. OpenAI has committed $1 trillion in spending to create the infrastructure it believes it needs, while losing billions as it services 95% of its users who pay nothing. This business plan epitomizes the philosophy “If you build it, they will come.”
Revenues will have to grow so much in the next few years to justify all the spending on AI that a valuation hangover seems unavoidable. I have no idea if the coming AI hangover will mimic the Dot.com and fiber optic wipeout, or merely be a take two aspirin and an Alka Selzer type of hangover. What I am confident about is that at some point investors will realize that a gap exists between where the AI stocks are trading and an economic reality that is less than all the hype currently accepted without question. It’s not if, but when AI companies and investors experience Déjà vu all over again. That realization is coming, but in the meantime I’ll repeat what I wrote in the September Macro Tides. “I don’t think the Peak of Inflated Expectations for AI have been reached yet so the stocks involved can be expected to gain altitude.”
The valuation risk, profitability risk, and excess capacity risk reminiscent of the Dot.com bubble aren’t the only challenges.
Depreciation
Depreciation is the accounting method of allocating the cost of a tangible asset over its useful life, reflecting the decrease in its value due to wear and tear or obsolescence. In the Gold AI Rush, companies are spending a lot of money on GPU’s (General Processing Units) and will depreciate the cost over time. The amortization period for GPU chips varies significantly, typically ranging from 1 to 6 years. Earnings are reduced by the cost of amortization, so earnings can be boosted if GPU chips are amortized over a longer time frame. Datacenter GPUs may only last from one to three years, depending on their utilization rate, which is normally between 60% to 70%. At that rate a GPU will last about 2 years. Hyper scalers like Google, Microsoft, and Amazon use GPUs for a wider range of workloads and can amortize them over longer periods, typically 3 to 4 years.
Nvidia has been doubling the computing power of its GPU chips every 10 months, which makes older less powerful chips worth less. Usage may make a GPU last less than 3 years, but technological obsolescence can cause the value of GPU chips to fall faster. The speed at which GPU chips will depreciate in value compared to an amortization schedule creates an accounting problem that can lower earnings. In order for a data center to remain functional at a high level (and attract paying customers), the owners will need to spend more money on new chips before the old chips have produced an investment return. If the demand for data centers slows at some point in the future, some owners may consider walking away from the data center just as commercial landlords give the keys back to the back once vacancy and cash flow weakens and results in a decline in the valuation of an office building. This risk may not manifest for another 2-4 years but it’s real and underappreciated today. Compare the AI data center amortization treadmill to the lifespan of fiber optic cables that last at least 40 years.
Electricity Needs
The demand for energy to power data centers has already impacted the cost of energy. Since 2021 the average cost of electricity in US cities has increased 35.7% rising from $.14 per kilowatt to $.19. One data center consumes as much electricity as 1,000 Walmart stores, while an AI search uses 10x more energy than a Google search. The demand for more electricity can be expected to increase the cost of electricity in coming years. The US Department of Energy estimates that US data centers will use 325,000–580,000 Gigawatt-hour (GWh) of power per year by 2028. That’s equivalent to the power used by 40 million homes. Data centers consumed 4% of US electricity in 2023 but could reach 12% within three years.
President Trump is trying to use trade policy to bring manufacturing back to the US. The cost of energy could prove problematic. Alcoa says aluminum producers need electricity at $30 per megawatt hour to be globally competitive, but industrial rates have averaged over $80 for four years. This pricing gap makes domestic manufacturing increasingly uneconomical compared to countries with less expensive electricity and fewer data centers.
Hyperscale facilities typically host at least 5,000 servers, but many contain far more and have footprints measuring tens of thousands to hundreds of thousands of square feet – with the largest hitting 1 million-plus. By comparison, the average data center usually has 2,000 to 5,000 servers. Hyper scalers have realized that local utilities won’t be able to provide the electricity they need and have pursued a multi- faceted approach. Hyper scalers like Amazon, Microsoft, and Google are the world's largest corporate buyers of clean energy. To bypass multi-year grid delays, hyper scalers are building their own on-site power. Hyper scalers have purchased natural gas turbines and fuel cells and located them next to renewable power sources, independent of the local utility. Data centers are now being planned and built adjacent to existing power plants which reduces transmission costs.
In 2024, natural gas supplied over 40% of electricity for U.S. data centers, according to the International Energy Agency (IEA). Renewables such as wind and solar supplied about 24% of electricity at data centers, nuclear power supplied 20%, and coal supplied around 15%. Natural gas is projected to continue supplying the largest share of energy at data centers through 2030, but nuclear power could eventually play a larger role.
To address the surge in electricity demand attention has shifted to nuclear power, but the near-term need for electricity is beyond what is possible in the next 5 years. The average nuclear produces power plant can produce 8,000 GWh per year. The US would need to build 73 new nuclear plants to satisfy the 580,000 GWh of power data centers will need by 2030. There’s just one minor detail. The US has added just 2 nuclear reactors since 1995 and it takes over 10 years to build one. Reaching the required energy output to keep data centers running in 2028 and beyond will be a serious ongoing challenge until nuclear power arrives.
Public Opinion
Data centers are often built in communities and sometimes require a vote by elected officials, which can be swayed by voter’s views. As people learn more about data center’s impact on communities, a groundswell of opposition has emerged.
Data centers are loud, with noise levels of 55 to 80 decibels (dB), which is comparable to listening to a lawn mower. The constant 24 hours a day low-frequency hum from cooling systems can be heard for two to three miles depending on topography. Data centers use a lot of water. In 2023, data centers used 17 billion gallons of water for cooling, but also used 211 billion gallons of water indirectly for power generation, according to Jeffries Research. Although construction of data centers creates jobs, the number of permanent jobs created can be underwhelming. A $7.3 billion Microsoft data center project in Wisconsin is expected to create 800 permanent jobs, but a $7.6 billion Hyundai car plant in Georgia is expected to lead to 8,500 jobs.
Data Center Watch, a research firm backed by AI security company 10a Labs, has found that $64 billion of data center projects in 28 states were delayed or blocked between May 2024 and March 2025 because of community opposition. Since then, several more projects have hit speed bumps. In September, Google backed out of a planned 468-acre data center in Indianapolis before the city council vote against it. The backlash against data centers will make it more difficult and more expensive for the hyper scalers to build the planned AI infrastructure.
In the last week of October, Amazon, Microsoft, Google, and Meta reported revenue and earnings for the quarter ending September 30 and every company beat estimates for revenue and earnings. The reaction to Meta’s report may be providing the first hint of AI fatigue. Revenue rose 26% to $51.2 billion and earnings were $7.25 per share versus Wall Street’s estimate of $6.69, which is a big beat. However, Meta’s stock declined by -15% in the next two days after the quarterly report was released on October 29. Meta’s stock has done well in 2025 so some profit taking was understandable, and there was a one-time tax charge of $15.93 billion from the One Big Beautiful Bill Act. Spending on AI though was the largest focus since Meta intends to spend much more on AI in 2026 as CEO Mark Zuckerberg noted. Meta will spend $70 - $72 billion in 2025 (above previous estimates) and could spend more than $100 billion in 2026.
Meta’s stock didn’t sell off because it was spending less on AI, it sold off because there is concern it might be spending too much. For the first time we’re seeing the question of return on investment enter the conversation. The overspending concern will spread to more of the AI related stocks at some point in 2026. On October 30 Meta planned to sell $25 billion in debt to fund its spending spree, but was able to sell $30 billion of bonds, since there was demand for $125 billion. This is why the Meta’s stock reaction is merely a hint since Meta’s advertising business still provides 98% of its revenue and advertising grew by 26% in the third quarter, and investors were eager to buy Meta’s bond offering. Of the $30 billion of bonds offered, $4.5 billion were 40-year bonds with a yield of just 1.1% above the benchmark Treasury bond!
FOMC
On October 29 the FOMC voted to lower the Funds rate by 0.25% to a range of 3.75% - 4.0%. This was widely expected so attention shifted quickly to the prospect of another 0.25% at the December 17 meeting. Prior to Chair Powell’s post meeting press conference, the odds of a cut in December were more than 90%. In Chair Powell’s opening statement, he addressed the prospect of a December meeting rate cut. “In the Committee’s discussions at this meeting, there were strongly differing views about how to proceed in December. A further reduction in the policy rate at the December meeting is not a forgone conclusion—far from it. Policy is not on a preset course.” After Chair Powell said this, the odds of a rate cut in December dropped to 62%.
Chair Powell’s statement wasn’t in response to a reporter’s question, it was premeditated, so it was important to Chair Powell and was likely added after the meeting. As he noted, ‘at this meeting there were strongly differing views’, which shouldn’t have been a surprise given the message from the September Summary of Economic Projections (SEP). I reviewed the SEP in detail in the September 22 Weekly Technical Review and concluded that the SEP was more supportive of a single cut before year end rather the than the 2 cuts widely forecast.
“The Median projection for the Funds rate at the end of 2025 was 3.6% and 3.4% at the end of 2026. These projections suggest there will be two more cuts in 2025 (October and December) and just 1 cut in 2026. In the June SEP the projections were for the Funds rate to be 3.9% at the end of 2025 and 3.6% in 2026. Wall Street thinks that the September SEP projection of 3.6% by the end of 2025 almost guarantees the FOMC will lower the Funds rate by 0.25% at the meetings in October and December. I don’t think it’s that cut and dried, and Chair Powell noted that there is a real split within the FOMC. “We have 10 participants out of 19 who wrote down two or more cuts for the remainder of the year and nine who wrote down fewer than two cuts.” The Median is the mid-point of the 19 members which is why the projection is 3.6%. However, if the 18 older member votes are weighted the average is 3.859% for the end of 2025, barely above what one cut would be (3.875%). When new member Steven Miran’s target of 2.875% is included (5 more 0.25% cuts before year end) the weighted average is 3.807%, which is still closer to 3.875% than 3.625% after two cuts.”
The divide within FOMC was reinforced by the two dissenting votes – One member (Miran) wanted a cut of 0.50%, and one member (Schmid – Kansas City President) didn’t think another cut was warranted. Interestingly, Lorie Logan (Dallas Fed president) gave a speech on October 31 and revealed she didn’t favor the cut this week. “I did not see a need to cut rates this week. And I’d find it difficult to cut rates again in December unless there is clear evidence that inflation will fall faster than expected or that the labor market will cool more rapidly.” On October 31 Kansas City Fed President Jeffrey Schmid explained why he dissented. “As it is, I see the stance of policy as being only modestly restrictive. Financial market conditions appear to be easy across many metrics. Equity markets are near record highs, corporate bond spreads are very narrow, and high-yield bond issuance is high. None of this suggests that financial conditions are particularly tight or that the stance of policy is restrictive. Likewise, the economy is showing continued momentum. Consumption remains solid, and data for July and August suggest an acceleration through the summer. With inflation still too high, monetary policy should lean against demand growth to allow the space for supply to expand and relieve price pressures in the economy.”
The FOMC has lowered the Funds rate by 0.50% since September 16. The 2-year note yield was 3.547% on September 16 and closed at 3.60% on October 31. The 10-year yield has ticked up from 4.076% to 4.101%. The bond market seems to be agreeing with Kansas City Fed President Jeffrey Schmid.
Economy
As of October 27, the Atlanta Fed’s GDP Now model estimates that the economy grew 3.9% in the third quarter, with the New York Nowcast at 2.4%. The economy is being supported by Federal government spending (6.0% of GDP budget deficit), spending on Artificial Intelligence, and the top 10% of earners who account for 50% of consumer spending. Just as the concentration of the 41 AI related stocks have contributed an outsized portion of S&P 500 investment returns, the decent GDP estimates are likely masking some underlying weakness. According to Moody’s Analytics, 22 states are in recession and 11 states are just treading water. I’ve written about the bifurcation within the economy and Moody’s is providing evidence. Spending on AI is supporting economic growth but it creating an economic vulnerability. The economy will slow when spending on AI slows materially, but the much of the economy may be unable to pick up the slack.
Stocks
In the October Macro Tides, I reviewed that the Advance – Decline Line has been invaluable in identifying when the stock market is vulnerable to an increase in selling pressure. “The Advance – Decline Line has telegraphed when the stock market is vulnerable to intermediate and major declines consistently since 1928.” I noted that the NYSE Advance – Decline Line recorded a new high on September 18, which pretty much ruled out that the S&P 500 would experience a decline of more than -7%. There were technical signs that the S&P 500 was likely to experience a pullback of -3% to -5%. “In recent weeks the 21-day Advances – Declines Oscillator has failed to confirm the higher highs in the S&P 500. This is why a -3% to -5% pullback has been expected.” On October 10 President Trump announced he would levy a 100% tariff on China, after China indicated it would suspend exports of rare earths to the US. The S&P 500 fell -2.7%. I thought there would be little more weakness before the S&P 500 rallied to a new high.
In the October 27 Weekly Technical Review, I laid out what the S&P 500 needed to do to validate the expectation for a pullback or confirm a rally to 6850 – 6900. “If a decline to test 6551 is coming, the S&P 500 shouldn’t close above 6772 and should turn down by October 22. S&P 500 has now closed above its 50-day moving average for 116 consecutive trading days, the 3rd longest streak going back to 1990 (1995 and 2007 were longer). Just because a drop below the 50-day average (6570) is overdue, it doesn’t mean its guaranteed soon. If the S&P 500 closes above 6772 a rally to 6850 – 6900 could happen before the Chinese turbulence hits.” On October 24 the S&P 500 opened at 6772, closed at 6691, and on October 29 traded up to 6920.
The Advance – Decline Line recorded a new high on October 24 and on October 27. However, as noted earlier, “On October 28 the S&P 500 closed at a new all-time high after gaining +0.23%, but only 104 stocks were up and 396 closed lower. Since 1990, the S&P 500 has never had weaker breadth on a day that it closed positive.” In addition, the number of stocks making a new low has jumped from 24 on October 27, to 50 on October 28, 110, 131, and 96 on October 31. Historically, an increase in the number of new lows above 40 is a warning sign of coming weakness.
The stock market is almost exactly where it was last month. The A-D Line suggests any correction will be followed by a rally to a new high, but is vulnerable to a -3% to -5% correction in the short term. Aggressive traders can establish a modest short position if the S&P 500 trades above 6950 using 7050 as a stop.
Treasury yields
The Federal Reserve has lowered the Funds rate by 0.50% since September 17 and Treasury yields have move up instead. As you may recall, the 10-year Treasury yield jumped from 3.6% in September 2024 to 4.8% in January 2025, even though the FOMC lowered the Funds rate by 1.0%. The bond market is suggesting that the rate cuts aren’t needed since the economy is doing OK at this point. The FOMC’s focus is wanting to act before real weakness materializes to avoid an unwanted increase in unemployment. That’s why Chair Powell has called the two cuts in 2025 “risk management.” The bond market may also be concerned about the amount of supply it must absorb since there is no sign that the budget deficit will fall materially from the $2.0 trillion in 2024 and 2025. The EU has signaled that larger deficits will be tolerated as military spending is increased, so the global supply of sovereign debt will be increasing. Treasury bond yields will have to compete for global capital flows which may require higher yields to increase demand for US debt. This is one reason why I think the long-term trend in Treasury yields is higher.
In the short term, I thought the 10-year Treasury yield could approach 3.886% after falling below 3.992%, unless it closed above 4.09% (red trend line). On Friday October 31, the 10-year yield closed at 4.101%. The down trend from the July high of 4.493% appears to have been broken. This suggests a move up to 4.20% is likely. A close above 4.201% would open the door for increase to 4.351%
Gold
As Gold ramped higher in September and October, I thought the pattern from the September 2022 low indicated that Gold was in Wave 5 of Wave 3. The vertical increase left Gold open to a quick hard decline once Wave 5 of Wave 3 ended. After topping at 4381 on October 20, Gold shed $494 in 6 trading days. The sharpness of the decline indicates that Wave 4 from the major low of $1046 in December 2015 has begun and will last for months.
Gold traded up to 4381 on October 20, so Wave 3 was $2765 (4381 – 1616 September 2022 Wave 2 low). A 23.6% retracement is $653 and the 38.2% is $1056 from 4381 the Wave 3 high. This generates targets of 3728 if Gold retraces 23.6% and 3325 if Gold retraces 38.2% of Wave 3. I think Gold will trade below 3750 in coming months. Chart wise the range between 3300 and 3500 is a target since the recent breakout occurred at 3403 and the April high was 3496. Markets frequently return to prior resistance before resuming an uptrend. This suggests there will be an opportunity to buy Gold below 3750 at some point as Wave 4 progresses.
Gold was expected to rally above 3496 once it traded above 3403, and a spike above 3700 was expected since Gold tends to zoom in a Wave 5. After Gold traded above 3800 and the blue trend line connecting the highs of Wave 1 and Wave 3, I thought it would reverse and close below 3800 and register a fake breakout. I adopted a strategy of selling into strength since Gold often spikes in Wave 5 and then experiences a quick sharp decline after the spike high is reached. Gold followed the pattern, but my execution was poor and left a lot of money on the table. The average price of the 5 sell was $3750.
Wave 4 is unlikely to be a straight line to the targets discussed above, since the narrative supporting the spike in price isn’t going to disappear overnight. There will be trading opportunities as Wave 4 unfolds. The breakout line (green) connecting the high of Wave 1 and Wave 3 high is near 3850, so Gold could work its way down to test this line before a more sustained rally takes hold.
Dollar
In early September I thought the Dollar would trade below the July 1 low of 96.37 and it fell to 96.21 on September 17 after the FOMC lowered the Funds rate by 0.25%. In the September 22 WTR, I noted, “Given the cut in the Funds rate and projection for two more reductions in 2025, one would have thought the Dollar would have continued to decline. Instead, it reversed up by almost 1%. The reversal is encouraging but the Dollar needs to do more to confirm that the low at 96.21 is the bottom.” We’ve been waiting since September 17 for the Dollar to show enough strength to confirm that the Dollar bottomed at 96.21.
The Dollar rallied after Chair Powell said, “A further reduction in the policy rate at the December meeting is not a forgone conclusion—far from it.” As noted previously, “The Dollar needs to close above 100.25 to provide more confirmation that the low at 96.21 is in place. If the Dollar does close above 100.25, it will need to hold above 97.46 on any pullback.” The Dollar strengthened more after the comments by District Presidents Logan and Schmid on October 31, which added to Chair Powell’s statement.
Major Trend Indicator
The Major Trend Indicator is comfortably above the blue horizontal line. As you can see the Bear market declines in 2022 and 2025 didn’t occur until after the MTI fell below the blue horizontal blue line, as well as the Intermediate decline in 2023 during September and October. The MTI is providing the same message of the Advance – Decline Line – a decline of more than 7% isn’t likely.
The MTI has been chopping sideways which indicates that upward momentum has stalled. However, the MTI hasn’t weakened as it did in January and February when the S&P 500 made a new high. The S&P 500 is expected to experience a correction that will be less than -7.0%.
