Man, A Plan, A Canal: Panama!

In addition to providing the inspiration for an excellent palindrome, the construction of the Panama Canal was one of the most consequential infrastructure projects in human history. The idea for a canal connecting the Atlantic and Pacific Oceans through Central America has existed for almost as long as Europeans knew there was a Central America. Yet more than 350 years would pass between Spanish conquistador Francisco Nunez de Balboa noting in his journal that the Isthmus of Panama would be fine spot for a canal and the first serious attempt at constructing one.

In the mid-1870s, French diplomat and entrepreneur Ferdinand de Lesseps became the driving force behind this first attempt to carve a path through Panama. With a stellar reputation based largely on the massive profits he generated for investors from leading the construction of the Suez Canal connecting the Mediterranean and Red Seas, de Lesseps and his partners were able to quickly determine a route, acquire a concession from Colombia to build a canal along that route, and, most importantly, raise a significant amount of capital from investors.

Yet, despite the obvious economic benefits, ample political will, the advent of steam engines to move and power heavy equipment, and investors eager to provide capital to fund the project, the French contingent failed to account for one tiny complication: the mosquito.

The Panamanian jungle was a far more difficult terrain to build in than the flat, Egyptian desert, but this could have been overcome. The greater challenge was the catastrophic death rate of the workers from yellow fever, malaria, and other tropical diseases. At the time it was unknown that such bloodborne diseases were spread by mosquitoes, thus all attempts to stop the spread of these diseases proved ineffective. All told, some 22,000 workers died from disease and accidents in the eight years of the French attempt.

With the project running hopelessly behind schedule, de Lesseps drew on his surplus of charisma to keep the funds flowing from French investors. However, an inability to retain workers, or keep a sufficient number of them alive, ultimately doomed the project, and the money eventually ran out in 1889. The de Lesseps venture reportedly spent about $10 billion in today’s dollars and the project’s bankruptcy wiped out hundreds of thousands of small investors.

Eventually, the United States would take over the project and, benefiting from the discovery that mosquitoes were the source of spread for various tropical diseases, completed the canal in 1914, 33 years after the French first put shovels into the ground.[1]

In the end, the Panama Canal was completed along the same route as originally planned, the economic benefits to maritime trade were as immense as expected, and the original canal remains in use to this day. Yet, the Panama Canal saga serves as a potent reminder that large infrastructure projects can often be delayed by unforeseen obstacles and the economic benefits, even if correctly predicted, may not arrive on the timetable investors require to produce a satisfactory return on their investment.

ChatGPT, Can You Think of a Palindrome Related to AI? 

The estimated $25 billion expenditure (in today’s dollars) to construct the Panama Canal seems positively quaint in comparison to the sums currently projected to be spent on data centers and infrastructure to meet the compute demands of AI. Morgan Stanley estimates that $2.9 trillion will be spent just on data centers in the next three years.[2] This doesn’t include the additional investments in electricity generation and transmission or local infrastructure to bring these data centers online. The five largest public cloud companies, Amazon, Microsoft, Alphabet, Meta Platforms, and Oracle, have increased capital expenditures from about $160 billion per year before investment in AI became a priority to an expected $400 billion this year; the amount is expected to continue to grow past half a trillion dollars in subsequent years.

Figure 1: Historical and Projected US Hyperscaler Capital Expenditures ($ millions)

Source: Consensus Wall Street estimates via Bloomberg.
The justification for this unprecedented level of investment was summarized by Sam Altman, the founder and CEO of OpenAI, in a blog[3] post earlier this year:

“Here are three observations about the economics of AI:

1.The intelligence of an AI model roughly equals the log of the resources used to train and run it.

2.The cost to use a given level of AI falls about 10x every 12 months, and lower prices lead to much more use

3.The socioeconomic value of linearly increasing intelligence is super-exponential in nature.

If these three observations continue to hold true, the impacts on society will be significant.”

In layman’s terms, he posits that the more capital companies throw at AI, the better and more efficient on a per unit basis it will become, which will encourage adoption and transform the way humanity works and plays. It is hard to argue with that logic, but a keen observer may have noticed the “if” in the last quoted sentence. We’ll come back to that.

For the time being, the market has enthusiastically embraced this potential. Using the iShares AI Innovation and Tech Active ETF as a proxy for the difficult-to-define universe of “AI stocks”, this basket has gained 26% year-to-date vs. 15% for the overall market, in which many of the same stocks are outsized contributors to that return. In the third quarter alone, “AI stocks” gained 18% vs. 8% for the overall market.

The recent outperformance of all things AI-related combined with lofty valuations assigned to many of these companies has effectively converted the broader US stock market into a wager on whether AI will live up to the hype. At quarter-end, the Information Technology sector made up 35% of the total large cap market weight, which is higher than the weight of this sector at the height of the late-90s Dot-Com Bubble. Subtract a few companies in that sector not exposed to AI and add in Tech/AI-adjacent companies from other sectors, including Amazon, Meta, Alphabet, Tesla, and others, and you can quickly get to well over 40% of the overall US market trading in some fashion on AI-related sentiment.

This phenomenon is not solely limited to US markets either. The much-maligned emerging markets[4], which have trailed US market returns by 10% per year for the decade prior to 2025, have had a banner year with a 28% year-to-date return, well ahead of the 15% return of the domestic market. However, here too, the top five contributors to this return are all AI-related stocks, including semiconductor manufacturer Taiwan Semiconductor, Chinese Internet and cloud giants Tencent and Alibaba, and Korean memory manufacturers Samsung and SK Hynix. These five companies now comprise 26% of the Emerging Markets index, which contains nearly 1,200 stocks.

Returning to Sam Altman’s “if” statement, there is ongoing fervent debate over whether all of this investment will be worth it. Per the Wall Street Journal, quoting consultancy Bain & Co., “the wave of AI infrastructure spending will require $2 trillion in annual AI revenue by 2030. By comparison, that is more than the combined 2024 revenue of Amazon, Apple, Alphabet, Microsoft, Meta, and Nvidia, and more than five times the size of the entire global subscription software market.”

Furthermore, even if AI does live up to the most bullish expectations, there are many “mosquitoes” that could possibly derail the timing or path linearity of AI’s economic success. These include difficulties in continuing to improve advanced chips on the necessary time scale and building the required electricity generation and transmission infrastructure needed to power data centers. OpenAI’s plans alone would require the equivalent of 17 nuclear power plants to be built in the next three years. Something as simple as a farmer refusing to allow a transmission line to cross her property could delay bringing a data center online for years. 

Despite these potential obstacles, as of this writing, OpenAI alone has made multi-year commitments to various suppliers and cloud hosting vendors totaling over $1 trillion. This is an astonishing sum relative to the company’s current annualized revenue run-rate of $13 billion. However, OpenAI’s ability to raise capital to fund its ambitions in spite of a lack of operating cash flow, and the willingness of investors and creditors to provide that capital, is also evidence of the confidence in the world-changing potential of AI possessed by both OpenAI and its various funding partners.

If that seems like circular logic, that’s because it is. This same circularity is present in the mechanisms being used to fund the AI infrastructure buildout. For example, in late September, Nvidia announced it would be making a $100 billion investment into OpenAI. Earlier in September, Oracle had announced a $320 billion increase in remaining performance obligation (software-speak for future revenue commitments) that was later learned to be primarily from a single customer: OpenAI. Putting the pieces together, Nvidia is investing in OpenAI, who is committing to pay Oracle for compute, who is purchasing GPUs from Nvidia in order to provide this compute capacity to OpenAI. And the valuations of all three companies have soared following these announcements.

The below chart recently published by Bloomberg[5] does an excellent job showing the interrelated nature of many AI-related businesses.

Figure 2: How Nvidia and OpenAI Fuel the AI Money Machine

Source: Bloomberg News Reporting

So You're Saying This is a Bubble?

None of this is to say this is a bubble. Some ingredients of a bubble are certainly present: unbridled enthusiasm for an emerging business model, rapidly appreciating stock prices, and creative funding arrangements that involve significant use of debt. But it is unclear if expectations have outstripped the yet-to-be-revealed reality of AI. Betting against human ingenuity has historically been a losing proposition in the long run and Altman’s vision of exponentially improving AI models delivered at decreasing unit costs and providing massive economic benefits to society could very well come to pass. Some of the smartest and most successful business leaders in the world are planning to spend trillions of dollars based on that vision of the future. And even if it is a bubble, the primary lesson of the late-1990s is that an asset bubble can continue to inflate long beyond the point where traditional fundamental analysis can justify underlying asset prices.

Nevertheless, one does not need to have a view on the ultimate economic return on the current and forthcoming investment in AI infrastructure to hold that risk is increasing and the valuations of AI-related companies have become far more demanding than they were just five months ago.

One must also acknowledge that the S&P 500 cannot currently be considered a broadly diversified investment. The concentration and return contribution of AI-related stocks has morphed the index into something akin to an actively managed AI fund.

When advising our clients, our preferred stance is always one of cautious optimism. That stance remains unchanged, although the current environment perhaps warrants placing a thumb on the cautious side of the scale.

That said, in all environments, there is no cause for drastic action when it comes to one’s comprehensive financial plan. The whole point of having a plan is to incorporate an asset allocation that takes into account one’s ability and willingness to accept risk over a long time horizon. Here risk is expressed as the inevitable tendency for financial markets to oscillate between exuberance and despair. To combat the psychological urge to take action in response to volatility, populating that asset allocation with a diverse mix of investments that are prudently managed and appropriately balance risk and reward is the best way to avoid overreactive blunders that can permanently impair capital. With that framework in place, whether or not AI lives up to current market-implied expectations, just stick to the plan and all should be well.

---------------------------------------

[1] Despite much-improved ability to combat mosquito-borne illnesses, another 5,000 or so workers still died during the American effort from disease and accidents. If you ever find yourself asking, “Why is it so hard to build things these days?”, a significant reason is that the death of thousands of workers in major construction projects is no longer considered an acceptable cost of doing business. 

[2] Ren, Shuli. Bloomberg. “AI Data Centers Give Private Credit Its Mojo Back”. October 2, 2025.

[3] https://blog.samaltman.com/three-observations.

[4] “Emerging Markets” stocks are represented by the MSCI Emerging Markets Index.

[5] https://www.bloomberg.com/news/features/2025-10-07/openai-s-nvidia-amd-deals-boost-1-trillion-ai-boom-with-circular-deals.

DISCLOSURES

Fees: Gross-of-fees composite returns incorporate the effects of all realized and unrealized gains and losses and the receipt, though not necessarily the direct reinvestment, of all dividends and income.  Gross-of-fees returns are presented before management fees, but after all trading expenses.  From inception through December 31, 2020 and after July 1, 2024, the Beginning Value Method (BVM) method was used to calculate returns. From January 1, 2021 through June 30, 2023, the Average Capital Base (ACB) method is used. Beginning on October 1, 2020, the net-of-fees returns are calculated by deducting model investment management fees, which are defined as the highest, generally applicable fees for the strategy of 1.00% of all composite assets. Prior to that, generally applicable fees were 1.25% for equity assets and 0.50% for cash assets. The firm's current management fee schedule is as follows: 1.25% on assets below $1 million, 1.0% per annum for assets from $1 million to $5 million, 0.85% per annum on assets from $5 million to $10 million, 0.75% per annum for assets from $10 million to $20 million, 0.65% per annum for assets from $20 million to $35 million, 0.55% per annum for assets from $35 million to $50 million, and 0.50% per annum for assets over $50 million. Such fees are negotiable. Where applicable, the total bundled or wrap fee charged to each portfolio is dependent on the end client’s financial advisor and wrap sponsor. The composite includes accounts that do not pay trading fees.

Definition of The Firm: Kovitz Investment Group Partners, LLC (Kovitz) is an investment adviser registered with the Securities Exchange Commission under the Investment Advisers Act of 1940 that provides investment management services to individual and institutional clients. From October 1, 2003 to December 31, 2015, the Firm was defined as Kovitz Investment Group, LLC. Effective January 1, 2016, Kovitz Investment Group, LLC underwent an organizational change and all persons responsible for portfolio management became employees of Kovitz. From January 1, 1997 to September 30, 2003, all persons responsible for portfolio management comprised the Kovitz Group, an independent division of Rothschild Investment Corp (Rothschild).

The description of products, services, and performance results of Kovitz contained herein is not an offering or a solicitation of any kind. Past performance is not an indication of future results. Securities investments are subject to risk and may lose value.

Kovitz does not guarantee the timeliness, sequence, accuracy or completeness of information included. The information contained in this material, including, without limitation, forward-looking statements, portfolio construction and parameters, markets and instruments traded, and strategies employed, reflects Kovitz’s views as of the date hereof and may be changed in response to Kovitz’s perception of changing market conditions, or otherwise, without further notice to you.