3rd Qtr 2025 West of the Hudson™

 Dear Clients: 

As we turn the quarterly calendar, there is one overarching investment theme to unpack. But before we do so, we first want to point out that investors have proved far more resilient to the uncertainties posed in the first quarter of the year than most experts expected. Since Liberation Day (4/2), the S&P’s is +35%, stocks and margins are at all-time highs, core inflation is moderating, and we have entered a rate-cutting cycle. Even the real matter of import taxes (tariffs) have proved to have a silver lining as repeated implementation delays provide a tailwind to what was expected to be an immediate margin-narrowing levy. This key trade policy has settled into a drawn-out, wait-and see muddle through and, while higher costs are expected and planned for in 2026 budgets, they remain mostly at arm’s length for many. The longer tariff policy remains in flux and tactical carve-outs are granted, the more flexibility managements have. Corporations are seemingly doing more with less and it’s flowing through to the bottom line. 

Markets are often driven by the possibilities of changing policy, and deregulation is everywhere and always a primary factor in driving sentiment. While certain policies loom large over the impacted industries, strategic directives are not driving market momentum. Today’s market propulsion is found in the enthusiastic support of all things artificial intelligence — especially the select few who foresee unprecedented profitability that results from the immense AI-supporting investments they make today. 

The sums a small handful of leading businesses devote to building and operating enormous, automated data centers that provide vast, global on-demand cloud computing power and storage are substantial. And the market supports their “hyperscale” efforts. This select few lead the world in utilizing semiconductors, computer hardware, energy, and real estate required to power the large language models at the heart of their operations. With the launch of ChatGPT (11/2022), the appreciation for AI’s power and utility by the masses went parabolic. Since then, stocks of companies participating in the growth of AI-data centers have accounted for ~75% of S&P500 returns, ~80% of earnings growth and ~90% of capital spending growth.(1) 

Trillions in equity valuations now rely on those hundreds of billions of investments to deliver, and it’s safe to say that the stock market economy is inordinately leveraged to the AI narrative. But so too is the broad U.S. economy, which has been driven this year-to-date not by the consumer, but data center investments.(2) While perhaps transitory, this is a historic shift with few serviceable comparisons. 

Artificial intelligence has evolved from a buzzword and meme to an all-encompassing capex technology boom powering the U.S. economy and stock market. Tech-infrastructure spending is rising at breakneck speed with forecasts for continued multi-year expansion. As we know well, past investment booms in technology and capital goods can end in memorable fashion, but investors see several factors suggesting this cycle may have room to run. We, along with many of you, invested through the tech bubble and are always on our toes when it comes to the profitability claims of future tech-enabled productivity. But as far as the market is concerned, AI is an innovative technology that must be adopted as quickly as possible. 

This AI capex boom today is contributing more to the U.S. growth rate than everything else in the entire economy. That’s different and worth unpacking. And it here that we jump off for this, our third letter West of the HudsonTM in 2025. 2 

• The Trillion Dollar Question 

We have previously cited the unique characteristics of the S&P500s Magnificent 7 relative to the remaining stocks in the index, but it requires updating. Today, the combined Mag 7 market-cap continues to swell, totaling 35% of the S&P500, 28% of its profits, and 42% of its YTD gain (6.2% of 14.8%).(3) The starter fluid for this magnificent rally is the extreme, intensive capital expenditures (capex) committed to meet the future expectations that support the all-encompassing AI narrative. The bulk of these investments are confined to just four of the seven companies — Alphabet, Amazon, Meta, Microsoft — all immensely large innovation leaders that continue to defy the law of large numbers and generate substantial, historic profits. Meeting future expectations requires colossal growth in cloud-related expenses, and these companies are meeting this need and being rewarded with higher valuations accordingly. 

These investments are so large that AI-related capital expenses alone contributed 1.1% to US GDP growth.(4) While GDP has long been driven by consumer spending (+70%), more than 50% of the first half of 2025 contribution to U.S. growth was derived from data center investments.(5) Widening the lens, since Q1’23 investment in overall tech infrastructure — graphics/memory chips, servers, networking gear to train and run the large language models at the heart of the boom — has grown ~23% while total GDP expanded just ~6% over the same interval.(6) All that computing power also needs land, buildings, power generation, and oodles of engineers. The real estate sector reports $40B worth of data centers currently under construction, a gawdy ~400% growth rate in since 2022 and expected to soon to exceed construction in office buildings (which has declined ~50% in the same period).(7) AI spending has, essentially, propped up the economy as consumer spending moderates. 

The AI boom has ushered in an exorbitant building spree with reports that, in just the past three years, the largest tech companies have committed more toward data centers than it cost to build the interstate highway system over four decades (inflation adjusted).(8) When it comes to technology and/or capital goods investment booms, there is a long history of valuation risks — they are well-established and ever-present (though not always obvious) — and investors are always ready to accept them. 

Today is no exception. As always, “it’s different this time” leads the sentiment and investors cite several factors suggesting this cycle has room to run. The latest entrant in the narrative is that U.S. government shares a sense of urgency to prevail in this space and is investing accordingly (what the WSJ dubbed “state capitalism with American characteristics”).(9) AI is treated as both a strategic priority and a national security issue, and this alignment of public and private incentives adds a powerful tailwind. 

As stewards of capital invested in these securities, our interest was nonetheless piqued with the following back and forth between the CEOs of Microsoft and Meta (Satya Nadella and Mark Zuckerberg) in an off-script exchange in regard to the weak link (electricity growth): 

“I hope we don’t take 50 years to meet our goals” (Nadella). “Yeah, well, we’re all investing as if it’s not going to take 50 years” (Zuckerberg).(10) 

This helps explain why the combined Mag 4 spent more than $100B on data center projects in Q2’25 alone and are forecasting nearly ~$400B of capital investment for the full year. They are essentially acting as a private-sector stimulus program.(11) Pre-pandemic, such spending would be on a par with your modern national fiscal stimulus package. 

Investors believe that these new technologies will rapidly improve and transform the economy and produce steady, annuity-like profits for the companies with the largest footprints. Data center vacancy rates are very low(12) and, while a virtual tidal wave of new plans are at various stages of development, new grid connections currently average four years in much of the country.(13) Google’s CEO Sundar Pichai has said, “lack of available, reliable power has become the biggest bottleneck for us”.(14) 3 

In this context, these CEOs are telling us what keeps them up at night as the rapid growth in “compute intensity” translates directly into higher electricity consumption. Data centers alone consume ~ 4% of the Nation’s electricity and that share is modeled to triple by the decade’s end.(15) Several states are already reporting AI demand above 10% of total electricity use and the fast-rising demand is causing a sharp jump in prices, with certain regions hit particularly hard.(16) So even as markets rise on the shoulders of data center investment, strategic importance and plentiful capital, it comes back to the emphasis on electrical grid and supporting infrastructure as the main limiting factor for data center growth.(17) 

Excitement does indeed breed momentum, but the view through our valuation prism also sees the transformative nature of these expenses. Yes, net income (NI) continues to soar, but free cash flow (FCF) has fallen precipitously since 2023 while, according to the data, from 2016 to 2023, NI and FCF of these companies as a group grew roughly in tandem. Since 2023, combined income grew +73% while FCF declined -30%.(18) It’s not across the board and there are outliers but it’s notable. For years, investors loved those companies because they were “asset-light” and earned their profits on intangible assets such as intellectual property, software, and digital platforms. Adding revenue required little in the way of more buildings and equipment, making them cash-generating machines. 

Today’s tech giants are no longer asset light, and a once unfathomable depreciation line has appeared on their cashflow statements. And we can expect that to swell as current AI hardware is estimated to depreciate at an accelerating rate with a useful average life of 2-3 years.(19) 

By any measure, today’s capital spending dwarfs the dot-com bubble and implies a massive shift in consumption is needed to make these investments worthwhile. Leading VC firms estimate that the money invested in AI infrastructure since 2023 alone requires consumers and companies to buy ~$800B in AI products over the lifespan of these chips and data centers to produce a good ROI.(20) Another leading industry heavyweight breaks down the ROI requirements further, estimating that the wave of spending will necessitate $2 trillion in annual AI revenue by 2030.(21) For context, that is more than the combined 2024 revenue of Amazon, Apple, Alphabet, Microsoft, Meta and Nvidia. As of now, the AI sector makes money from a combination of subscription fees for chatbots (ChatGPT) and money paid to use these companies’ data centers. A leading investment bank chimed in with an estimate that there was ~$45B of AI product revenue generated in 2024.(22) $45B is 2% of $2 trillion, and how the tech sector will cover the gap is the trillion dollar question. Consumers have been quick to use AI, but most are using free versions, while businesses have been slow to adapt (beyond the ~$30/month fee for Copilot or similar products).(23) 

Two emergent parallels to the last great tech cycle is vendor financing and the significant growth in the leverage supporting the boom. As has been widely reported, the leading stock in the index, Nvidia, invested $100B in start-up OpenAI, which then signed a cloud deal for a similar amount with Oracle that, in turn, led Oracle to purchase a similar value of Nvidia chips for its data centers.(24) Oracle jumped over 30% in a single session after the OpenAI announcement for an amount of money OpenAI doesn’t earn yet, to provide cloud computing facilities that Oracle hasn’t built yet — requiring 4.5 GW of power (the equivalent of 2.25 Hoover Dams or four nuclear plants) — that isn’t available yet.(25) We are not suggesting that this momentum is about to end, but at a minimum, it’s worth noting that these deals are not firm contracts, but rather “memorandums of understanding” and therefore carry “less commitment”.(26) 

Nonetheless, OpenAI’s latest capital round values them at $500B, making it the most valuable startup ever, far ahead of SpaceX and ByteDance (TikTok’s parent).(27) Among other things, it’s ambition is a tech holy grail: to convert conversational queries into direct purchases and fill in the blank between vague user intent to specific product — a place where today’s social discovery feeds still struggle. 

But this vision is gated by physical reality, requiring an AI power grid of unprecedented scale that doesn’t exist and/or is spoken for. This has forced sophisticated partnerships built on complex financial engineering and a high-stakes bet on the future of energy. This is a story about megawatts, money, and moats, and BIG ideas don’t boot up a 10GW grid. Cash does. Multi-gigawatt buildouts pull billions of dollars in capital forward, well before a single dollar of revenue shows up. This creates a massive cash chasm. 4 

So, how do you finance a project with a colossal upfront cost and a long-tail payoff? It requires a sophisticated piece of financial engineering designed to de-risk the venture. The escalating OpenAI deals achieve this in theory but still, circular (vendor) financing is a high-wire act. The supplier takes an equity stake in the customer, who in turn signs a massive purchase commitment with that same supplier. It’s aligned by design. But the structure is not without peril. 

All the risks are manageable if the grid is built on schedule and the GPUs are running at a high load factor, turning capex into cash flow. If utilization lags, the deal’s logic flips. Fixed costs become a deadweight, and the equity marks turn volatile, creating a financial headache. 

The tech infrastructure boom had been largely self-funded, but Oracle has broken the pattern. They generated $11B in 2024 FCF and there is no way to pay for their Nvidia deal other than to raise debt. But the deal is significant enough to force Amazon, Microsoft and Google to respond. AI infrastructure investment is no longer a discretionary investment, and they will defend their turf. What had been disciplined, high-margin, asset-light, cashflow machines, has moved towards a debt-fueled arms race.(28) 

No one (as yet) questions AIs potential to raise growth and productivity but financing this boom is a strain and business must invest to “keep up”. As Meta stated on its earnings call, “we are early in the life cycle of investments, and we don’t expect that we are going to be realizing significant revenue from any of those things in the near term”.(29) In other words, the profitability payoff is very much in the future. And not that many business will be able to keep up. 

• Cycle Comparison 

The overwhelming investor belief today is that data center growth is dissimilar to the 1990s telco investment glut. Nonetheless, the most common question floated remains “is this AI cycle comparable to the dot.com cycle of the late 90s”? The short, if maddening, answer is “yes and no”. 

We have cited a few examples, but they bear repeating. The similarities are much fewer but still substantial due to the absolute size of these companies and resulting extreme S&P500 index concentration. Market returns have broadened out from 2024 (a positive for active managers!) but the YTD results remain bifurcated: S&P500 (+14.8%), S&P Mag 7 +19.5%, S&P 493 +13.5%.(30) This reflects the potency of Mag 7 earnings growth (32% YOY). Outside of these companies, earnings growth across the rest of the market is 6% YOY. (31) We run a highly concentrated portfolio (~30 positions), and while our portfolio companies have generated higher than average earnings growth, we are anxious to see that spread across a host of industries beyond the AI-adjacent. There is no time in history were such levels of market concentration prevail for long. 

Financing (especially off‑balance‑sheet and light covenant quality structures) is one to watch in this cycle as this too is a massive front‑loaded capital buildout. It is estimated that U.S. broadband/telco capex peaked at ~$100B in 2000. As noted above, 2025 hyperscale capex alone is tracking above $400B. There is also a similar refrain: “the Internet changes everything” and “AI changes everything”. In terms of market concentration, the top 10 S&P500 tech names represented ~22% of the index at the Dot‑Com peak; the top‑10 tech slice is ~40% (and growing) today.(32) 

So, what is different? The 1990s networking and fiber optic age involved hundreds of telco providers with highly-levered balance sheets — network growth was almost entirely debt-driven. Today’ participants are a small handful of highly-rated corporate behemoths with extraordinary operating cashflows to fund growth internally. Still, we mustn’t lose sight that funding has been mostly cash-driven (until the Oracle/Nvidia deal) but that cant likely last as the cost of investment is going up, and the ROI of these projects can’t all be profitable. 5 

Equipment scarcity is also different. The 1990s was defined by a global demand for immense amounts of fiber optics, which weren’t available in scale and delayed the growth narrative even as debt ballooned. Today’s scale-up is in computation and power, and there are few anecdotes blaming a lack of availability for hardware, software, or capital. Again, we must account for supply chain functionality, and the war tariff needs to be monitored across availability, pricing, transportation, and lead-times. 

Still, the visible early payoff is market nirvana and today’s market leaders in this arms race are reporting growth in AI-associated profits already.(33) Buoying this are that financial conditions are much different: the Fed was hiking rates in the late 1990s — making debt-service much more arduous — whereas seem likely to continue cutting rates today. 

Valuations are elevated but do not sit alone on a higher plateau from the rest as existed then.(34) There are doubts — as evidenced by the occasional enhanced volatility around these stock prices from time to time — but the overwhelming investor belief is that data center growth is dissimilar to the telco investment glut. And this is reflected in the dominance of these securities in equity markets. 

Interestingly, the first step onto the internet required consumers and businesses to get wired for high-speed access. That took years. However, because of that heavy lifting, cloud computing and by extension, AI chatbots, etc., are immediately available for use today. The over-deployed fiber optics that laid during formant for a decade is today fully integrated into our wired world. In that sense, it has a useful life measured in decades. 

• Final Thought 

The discourse of the moment understandably revolves around the familiar refrain of all unbridled investor fixations: is this a bubble? 

Bubbles come about through the divergence between an asset’s price and its lifetime use or cash flow. To be sure (in financial circles at least), the term “bubble” is readily thrown around whenever any kind of over-zealous frivolity is on display (or whenever a bull market unfolds in which one is not participating!). This is frustrating as the term “bubble” can actually describe different phenomena, starting with productive vs. unproductive.(35) 

Is the AI craze a bubble? Going back to our earlier response to an equally difficult set of choices –yes and no. The spending has enough attributes of a bubble: massive capital spending plans with very limited visibility of future returns, connected party transactions, and very stretched valuations for most key actors. But still, the hope is that all the money spent building LLMs will unleash a productivity boom and that’s the impetus to “pay up and own the market”. As long as investors remain receptive to the strategy, these tech giants will one-up each other with their spending commitments. But if AI does not start to deliver returns in the next few years, what will the data centers being built today be worth in 2028 and beyond? That’s not dissimilar to the telecoms, who spent billions blanketing the country with fiber optic cables in the belief that the internet’s growth would be so explosive, any investment was justified. The result was a massive overbuilding that went bust and industry giants toppled like dominoes.(36) Ultimately, the growth of video streaming in the early 2010s helped soak up the oversupply, but that wasn’t the use case the investment narrative was originally driven by. 

For a number of reasons of which we have discussed, it’s not the 1990s and the potential innovation curve behind todays exuberance is well-grounded. Still, the combination of investor enthusiasm backing a really big theme and resulting market concentration just as leverage comes to the fore…it’s not the same, but it is similar nonetheless. The index concentration is obviously not good for investors seeking to diversify their investments and investing in the S&P500 index is a vote for the Mag 7 stocks to continue on their trajectory (in truth, its actually 3 stocks — Nvidia, Microsoft, and Apple represent ~22% of the index).(37) If profitability sustains, then markets are very much priced accordingly. If not, it’s another “wash, rinse repeat” moment. 6 

We are not pushing back on the narrative but are mindful of the broadband boom as historical context. They were right that the internet would drive productivity, but wrong about the timing of the financial payoff. History is replete with game-changing advancements and the investor optimism they provide. Be they canals, railroads, electricity — investors imagine explosive, unbridled growth. And they are usually right for a while. But overbuilding always follows and the growth fails to materialize before the debt comes due. Business fail as their financial apparatus disappears, and investors pay the price…even as the new technology slowly permeates the economy for the benefit of future generations. 

There are a small handful of stocks that are getting much attention, and investors spend a disproportionate amount of time asking about them. But numerous other stocks (and investment strategies) can deliver stable returns with fewer sleepless nights. Some enjoy the adrenaline rush that comes with investing in the latest shiny toy, but for investors looking for steady and stable returns, boring is often a better strategy than fancy. A dot-com-style bust looks far-fetched as todays tech companies are mature and profitable. But history isn’t as binary with “winner/loser” generalities. Still, investors expect tech’s new asset-heavy business will be as profitable as their asset-light models, but this space bears watching as the spending rolls on. After investing through the tech bubble, it is natural to consider that AI is blowing its own bubble as it dominates capital investment. Afterall, the companies chasing AI can’t all win. 

Yours truly, 

Hirschel B. Abelson 

Chairman, LCES 

Adam S. Abelson 

Chief Investment Officer, LCES 

This information is intended for the recipient’s information only. It may not be reproduced or redistributed without the prior written consent of Fischer Stralem Advisors & HighTower Advisors LLC, an SEC-registered investment advisor. Securities offered through HighTower Securities, LLC, member FINRA/SIPC. This information is intended for the recipient’s information only. It may not be reproduced or redistributed without the prior written consent of Fischer Stralem Advisors. This commentary reflects our current views and opinions. These views are subject to change at any time based upon market or other conditions. Past performance is not indicative of future results. Sources: (1, 4) JP Morgan 9/24, (2, 5) Apollo 8/17, (3) Fischer Stralem Advisors 10/1, (6, 11) Wall Street Journal 8/2, (7) Goldman Sachs 8/29, (8, 10, 23, 36) Wall Street Journal 9/25, (9) Wall Street Journal 8/11, (12) CBRE 9/29, (13, 15, 17) BCA Research 9/25, (14) Google 7/25, (16) NYT 9/29, (18) FactSet 9/18, (19, 35) GaveKal 10/3, (20) Sequoia Capital 9/25, (21) Bain & Co 9/23, (22) Morgan Stanley 9/25, (24, 25) Bloomberg 9/16, (26) Strategas 9/23, (27, 28) Wall Street Journal 9/29, (29) Meta 7/30, (30) Bloomberg 10/1, (31) Bloomberg 10/2, (32, 33, 34) Evercore ISI 10/3, (37) Apollo 7/10