AI Investment Boom, Jobs Report Accuracy, AI Investing Agents, Historical Market Trends, Consumer Spending Trends & More

The Trillion-Dollar Math of AI

We have officially entered the era of trillion-dollar math, and some of the numbers being thrown around should make investors stop and think.

A recent Barron’s article highlights the extraordinary financial assumptions behind the AI boom. Anthropic is reportedly targeting a valuation around $2 trillion while the company has projected enormous future infrastructure commitments. The business lost $42 B last year on $4.6 B in revenue. While revenue has improved substantially this year and as of July the company has an annual run rate of $65 billion, it expects to spend $518 B on cloud computing and infrastructure obligations in the coming years. At the same time, a Brookings estimate puts the total AI infrastructure buildout from 2025 to 2032 at roughly $10.3 trillion.

The question isn't whether artificial intelligence is going to change the economy. I believe it will. The question is whether investors are assuming that AI adoption, revenue growth, margins and cash flow will all grow fast enough to justify the valuations being placed on these companies.

This is where I think the discussion about regulation becomes important. A new Wall Street Journal poll found that more than half of those surveyed support safeguards on artificial intelligence, while nearly two-thirds believe the development of the technology should be paused.

We already accept that certain industries are too important, too complex and potentially too dangerous to simply tell companies to regulate themselves.

The FDA regulates about 3.6 trillion worth of food, medical, and consumer products using a risk-based approach because mistakes can directly affect people's health and safety. The FAA regulates aviation because an error involving an aircraft can have catastrophic consequences. NHTSA oversees vehicle safety and is now dealing directly with the challenges created by automated driving systems and robotaxis. The agency is developing new safety standards while also trying to allow autonomous vehicle technology to advance.

And look at the banking system. The Federal Reserve, OCC and FDIC require banks to have appropriate risk-management frameworks in place.

None of this means I believe the government should regulate every AI application or try to pick winners and losers. Far from it. But there is a difference between regulating an industry and regulating the specific high-risk applications of a technology.

If an AI system is helping write an email, the consequences of an error are relatively small. If an AI system is making a medical decision, controlling an aircraft, operating an autonomous vehicle, approving a loan or managing a critical financial system, the consequences can be dramatically different.

That is the type of regulation I think makes sense: risk-based oversight rather than blanket regulation. The financial side deserves the same scrutiny.

When a company is valued at $1 trillion or $2 trillion, investors aren't just betting on today's business. They are betting on years of extraordinary growth and future cash flows. If those assumptions are wrong, the consequences won't be limited to a handful of venture capital investors. AI has become such a large part of the public markets that a major repricing could affect the entire market.

AI is real. Technology works. The economic opportunity is enormous. But that doesn't mean every AI company is worth whatever valuation the market assigns to it.

We've seen this movie before with railroads, the dot-com boom, and other transformative technologies. Technology can be revolutionary while the investments made around it can still be terrible.

That's why I'm less interested in whether AI is the next great technological revolution and more interested in what investors are actually paying for that revolution. Technology can be real. The opportunity can be real. And the valuation can still be completely wrong.

How reliable is the jobs report? Look at the revisions

One of the biggest things investors should understand about the monthly jobs report is that the headline number is not a final number. It is an estimate, and in some cases, it is an estimate based on a surprisingly small amount of information.

The September establishment survey is a perfect example. The initial collection rate was just 53.1%. That means barely more than half of the reporting locations had provided information by the time the initial estimate was produced. BLS continues collecting responses after the first release, which is why the initial employment, hours and earnings numbers are revised over the following two months.

This is important because the establishment survey is what produces the headline nonfarm payroll number. BLS surveys roughly 119,000 businesses and government agencies representing about 622,000 worksites. The survey is designed to provide information on employment, hours and earnings, but the first estimate is necessarily based on incomplete responses.

So when you see an initial September payroll estimate, you shouldn't treat it as though BLS counted every job in the economy. It didn't. The data is collected from a sample and then used to estimate employment across the broader economy. BLS itself acknowledges that the most recent two months are based on incomplete returns and are considered preliminary.

The unemployment rate is calculated differently. It comes from the household survey, not the establishment survey. BLS and the Census Bureau survey roughly 60,000 households and determine whether individuals are employed, unemployed or outside the labor force. The unemployment rate is the number of unemployed people divided by the total labor force. To be counted as unemployed, someone generally must be without a job, available to work and actively looking for work.

That distinction matters. Someone who wants a job but has stopped looking is generally not counted as unemployed and instead is considered outside the labor force. That is one reason why looking only at the unemployment rate doesn't tell you everything about the health of the labor market.

The establishment survey also has another complication: businesses that are newly created cannot immediately be captured in the survey because they aren't yet on the sampling frame. BLS therefore uses a birth-death model to estimate the employment impact of new and closing businesses. Those estimates are eventually reconciled against much more complete unemployment-insurance tax records through the annual benchmark revision.

With numbers like these, it's no wonder we are seeing so many revisions.

The better approach is to look at the trend, multiple months of data, labor-force participation, the employment-population ratio, wages, hours worked and subsequent revisions rather than treating the first headline payroll number as the most important fact.

I still don't understand why, with all of the advancements in technology, there isn't a better way to collect this information and produce more accurate numbers from the start. We have technology that can process enormous amounts of data almost instantly, yet we're still relying on a system where nearly half of the establishments haven't responded when the initial jobs estimate is released. With the importance these numbers have for the economy, markets and Federal Reserve policy, it seems like there should be a better way.

What happens when everyone has an AI agent managing their investments?

A recent Barron’s highlighted the growing use of AI-driven investing. Since Robinhood launched agentic trading in May, more than 150,000 customers have opened agentic trading accounts, and AI agents are now using Robinhood’s tools nearly 30 million times a day. Public says customers created 20,000 AI agents within a month of making them available to all users, with more than $200 million in assets now overseen by AI agents.

The results from the article were kind of silly to me because it was over a 3-day period using just $500 on Claude Opus 5, ChatGPT-5.6 Sol and Gemini 3.1 Pro. Claude had 3-day return of 2.9%, ChatGPT was down 0.7% and Gemini was up 0.9%. Again, I don’t think the results tell us much given the short time period, but it made me think: If everyone uses AI to make investment decisions, won't everyone eventually be using the same strategy?

If millions of AI agents have access to the same earnings reports, financial statements, economic data, analyst estimates and market information, it seems reasonable that they will identify many of the same opportunities.

And that creates a fascinating problem. If every AI agent determines that a stock is undervalued at $50, investors could all try to buy it at $50. That buying pressure pushes the stock higher, eventually eliminating the very opportunity the AI identified.

This is actually similar to the argument surrounding index investing. Imagine if every investor simply bought and sold index funds and nobody made individual decisions about what companies were attractive or unattractive. In that world, the flow of money into and out of the index would essentially become the sole factor in determining where prices move.

Of course, we aren't there. Active investors, institutional investors, arbitrageurs, short sellers and other market participants still analyze individual companies and create price discovery.

But AI agents could create a similar dynamic if their use becomes widespread. Instead of everyone simply following an index, we could eventually have millions of AI agents analyzing the same information and reaching similar conclusions.

There's another interesting point from Barron's testing. When AI agents were asked to construct a traditional 60/40 portfolio, Claude, ChatGPT and Gemini all gravitated toward many of the same low-cost Vanguard and iShares index funds.

That shouldn't be surprising. If you ask several systems to find the most efficient way to build a diversified portfolio, they're going to find many of the same answers.

But investing isn't just about having more information. It's about how you interpret that information differently from everyone else.

Markets need disagreement. One investor thinks a company is worth $50, another thinks it's worth $75 and another thinks it's worth $35. That disagreement is what creates a market.

If everyone eventually has an AI agent that reaches the same conclusion, the question becomes: Who is left to disagree?

History is Helpful, But It Is Not a Guarantee

The four-year presidential cycle has historically been a useful guide for understanding market trends, and the numbers heading into the next few years are certainly interesting.

Since World War II, the pre-election year, which would be 2027, has been the strongest of the four years on average. In fact, the S&P 500, including dividends, has not experienced a negative pre-election year since the 1930s. One possible explanation is that policymakers tend to favor economic growth heading into an election, using fiscal policy to create more favorable economic conditions.

The midterm election year, by comparison, has historically been much more volatile. Larger-than-average market corrections are common, with many of those declines reaching their lows around September and October. While that can be uncomfortable for investors, history shows that these periods have often created attractive buying opportunities.

Since 1950, the S&P 500 has generated an average return of 16.6% during the 12 months following a midterm election, and there has not been a negative 12-month period following a midterm election.

Looking at the presidential cycle quarter by quarter makes the next period even more interesting. Historically, the nine months from October 1 of the midterm election year through June 30 of the following pre-election year have been the strongest stretch of the entire four-year cycle. During that period, the S&P 500 has gained an average of 20.2%.

But I would be careful about giving 100% credence to these statistics. Historical patterns are interesting, but they don't know what the economy, interest rates, inflation, earnings, or geopolitical environment will look like in the future. A pattern that worked repeatedly in the past doesn't mean it has to work the next time around.

Why Does Consumer Spending Look So Good?

The final revision of second-quarter GDP showed the U.S. economy growing at an inflation-adjusted annual rate of 2.2%, up from the previously reported 1.5%. Much of the upward revision came from stronger personal spending, which increased at a 3.8% annualized rate, compared with the earlier estimate of 3.2%. Overall GDP growth was weighed down by spending on imports and a drawdown in business inventories. Excluding those factors, domestic final sales grew at an impressive real 3.9% annualized rate. That represents a significant acceleration from the 2.4% growth recorded in the previous quarter and just 0.5% in the quarter before that. While this sounds like the consumer is feeling good, it seems to conflict with what consumer surveys are telling us.

The University of Michigan’s recent consumer sentiment survey posted its second-lowest reading going back to 1952. We saw a similar trend in the Conference Board’s consumer confidence data, which came in lower than expected. You have to go back 12 years to find a reading that was as weak.

So, why does consumer spending continue to look so strong despite such poor consumer sentiment?

Based on what I’m seeing, I still believe higher-income households are accounting for a disproportionate share of consumer spending. While there are signs that lower-income consumers are doing better, it is worth looking at the wealth being accumulated at the top. The top 10% of U.S. households saw their wealth increase by $38.2 trillion in the year ended June 30.

Another interesting statistic comes from economists at Bank of America, who noted that U.S. household equity wealth is now roughly 10 times consumer liabilities. That is a significant amount of wealth relative to what households owe, and I believe it helps explain why spending has remained resilient. When people feel wealthier, they are generally more willing to spend, even when consumer sentiment surveys suggest that confidence is weak.

The question is how sustainable this trend will be. A meaningful reversal in technology stocks, particularly if it develops into an extended downturn, could reduce household wealth and eventually weigh on consumer spending. Given how much wealth has accumulated in the stock market, this is a risk worth watching closely.

Financial Planning: Big Changes Coming to Social Security…

The Claiming Age Clarity Act seeks to simplify Social Security terminology by renaming certain claiming milestones. Currently, the terms “early eligibility age,” “full retirement age,” and “delayed retirement age” are used for claiming benefits at ages 62, 67, and 70, respectively. The legislation would replace these terms with “minimum monthly benefit age” at 62, “standard monthly benefit age” at 67, and “maximum monthly benefit age” at 70... that’s it.

Unfortunately, this legislation does nothing to address the far more significant issue facing Social Security, its long-term financial sustainability. Changing terminology does not increase funding, reduce the projected shortfall, or prevent future benefit cuts. While clearer communication may have some educational value, it is difficult to view this as meaningful progress when Social Security’s underlying financial challenges remain unresolved. At a time when insolvency is only a few years away, we would be better served by policymakers focusing on substantive reforms to address the program’s long-term solvency rather than spending time on what amounts to little more than a rebranding exercise.

Selling a House? This Is Not 2021!

Some people selling a house are still living in the past, thinking they can list their home on the market and have it sell very quickly. Welcome to 2026, where mortgage rates are well over 7% and existing home sales have fallen to their lowest level in a year. The next existing-home sales report will reflect the impact of the Federal Reserve’s recent interest-rate hike, and we could see sales decline even further.

The good news is that if you’re selling a luxury home, there are still cash buyers and the market remains competitive. However, buyers in the beginning and middle portions of the market are generally much more sensitive to interest rates. Higher mortgage rates have reduced what these buyers can afford to pay because of the higher monthly payment.

If you’re selling a home today, there are some mistakes you need to avoid. One of the biggest is thinking your home is worth more simply because you believe it is better than everything else on the market.

Remember 2021. There were very few homes for sale, buyers were aggressively competing for properties, and mortgage rates were extremely low. Sellers had much more leverage and could be aggressive with their asking prices.

That is not the market we are in today. If you’re selling a home today, you should think carefully before rejecting an offer simply because the price is below what you want or because the buyer is asking for concessions. It could be a long time before you receive another offer, and there is no guarantee the next offer will be better than the one you just rejected.

Another thing you may be asked to do as a seller is contribute money toward buying down the buyer’s mortgage rate. While that may not have been something sellers had to consider in 2021, it can make a significant difference today by lowering the buyer’s monthly payment and making the home more affordable.

If your home needs repairs, don’t assume the buyer will simply take care of those costs. A buyer may look at the cost of the repairs, combine that with a mortgage payment based on today’s higher interest rates, and decide they simply cannot afford the house. They may walk away altogether.

Before putting your home on the market, give it a facelift. Paint the walls, replace worn or outdated baseboards, make the small repairs you have been putting off, and make sure the landscaping looks attractive. These relatively small improvements can make a big difference in how buyers perceive your home.

The bottom line is simple: If you’re selling a home in 2026, you need to price it for the market you are actually in, not the market we had in 2021.

You may still be able to sell your home, but you may have to be patient, negotiate on price, offer concessions, or help the buyer with their mortgage rate.

Colleges Are Finally Cutting Back on Tuition Costs

Finally, colleges are starting to get the hint that people will not pay just any price they charge for tuition. I have a master’s degree myself, and it has served me well. However, times have changed, and a degree from certain colleges does not always translate into higher earnings or justify the cost.

Some colleges have cut their tuition in half, while others are promising loan-free financial aid to families who meet certain criteria. Some experts argue that these tuition cuts are largely a marketing tool because relatively few students were paying the full sticker price in the first place. They also point out that students often feel better receiving a scholarship because it feels like they are getting something, rather than simply paying a lower tuition price.

Regardless of how you look at it, colleges are facing growing pressure to bring down the cost of higher education, and in my opinion, that is a good thing. The price of a college education needs to make sense relative to the financial benefits a student can reasonably expect to receive.

My hope is that colleges can achieve these cost reductions by cutting back on expensive administrative staff and bloated overhead rather than reducing teachers' and professors' pay or compromising the quality of education.

Why the Difference Between the Market-Cap-Weighted and Equal-Weight S&P 500 Matters

When people invest in the S&P 500, they often think they are sufficiently diversified because they are investing across 500 companies. While they technically own exposure to all 500 stocks, they may not realize how unbalanced the index has become. Today, the 10 largest companies account for roughly 40% of the S&P 500 based on market capitalization.

This concentration becomes even more apparent when looking at market performance. Year to date in 2026, the S&P 500 has gained roughly 13%, but just four companies—AMD, Apple, Micron, and Nvidia—have accounted for nearly half of that gain.

This is why the difference between the traditional market-cap-weighted S&P 500 and the equal-weight S&P 500 matters. In a market-cap-weighted index, the largest companies have the greatest influence on performance. In an equal-weight index, each company receives approximately the same weighting, giving smaller companies within the index a much greater influence on returns.

We are beginning to see opportunities emerge outside the largest technology companies. In September, the equal-weight S&P 500 fell roughly 5%, while the traditional S&P 500 declined only about 0.5%. The difference highlights how the strength of a handful of large companies can mask weakness across the broader market.

There are plenty of warning signs suggesting that the market carries significant risk and could experience a major pullback. However, that does not necessarily mean every stock will decline equally. The market's heavy concentration in a handful of companies creates the potential for a very different outcome among individual stocks and sectors.

If you own an S&P 500 index fund or have accumulated substantial gains in technology stocks, now may be a good time to review your portfolio. Consider whether you are adequately diversified, whether your investments are priced reasonably, and whether you have too much exposure to a small group of expensive companies. In some cases, reducing exposure to overvalued stocks or even selling them entirely may make sense, even if that means trimming your position in the S&P 500 itself.

And that brings us to another major question facing investors today: the AI trade. Michael Burry recently argued that the market is entering the first stage of grief over the potential fate of the AI trade: denial. He stated in a social media post, “Per 2000 and 2008, this stage lasts 6-9 months.” Burry, who became famous for correctly betting against the housing market before the financial crisis, is currently shorting several companies involved in AI.

He has also criticized Nvidia's depreciation schedule, arguing that it understates the true economic cost of the company's rapidly changing GPU technology. He has warned that institutional lenders exposed to data centers could eventually face problems as well, particularly through data-center loans and bonds.

Whether Burry is right about AI is still very much up for debate.

The interesting part is that we could potentially have two very different forces working against each other. On one hand, historical market patterns suggest the next nine months could be one of the strongest periods of the presidential cycle. On the other, investors are facing questions about AI valuations, massive capital spending, financing, depreciation and whether the expected returns on all of that investment will ultimately justify the enormous amount of money being spent.

This is exactly why I think historical market cycles are best used as a guide rather than a prediction. History can tell us what has happened before. It cannot tell us what is going to happen next.