What the Internet Era Teaches Us About Investing in AI

A few weeks ago, we published a newsletter on the SpaceX IPO, which turned out to be the biggest in history (the IPO, not the newsletter, although we appreciated your kind words). In the piece, we pointed out that SpaceX is much more a bet on AI than it is a bet on space. SpaceX, however, is not the only way to bet on the future of AI. There are microchip makers like Nvidia, of course, and a host of other companies, such as Microsoft and Google, that are also investing billions of dollars in developing AI capabilities. And coming around the bend are two more mega-IPOs slated for later this year — Claude maker Anthropic and ChatGPT maker OpenAI.

Of the six companies mentioned, only Nvidia's AI efforts are profitable. Does this mean that our only two options are buying companies that are already very expensive or are losing money? Or is this simply the risk we take to participate in the incredible future upside of AI? In this newsletter, we will explore what history suggests about the best way to invest in a new technology long-term.

The thesis can be right, and the investment can still be wrong

Let's begin with the lesson the last technology revolution taught most painfully: you could be completely correct about where the technology was going and still lose your shirt.

The signature example is the late-1990s internet buildout. Telecom companies laid enormous amounts of fiber-optic cable on the thesis that internet traffic would explode. It did — beyond almost anyone's imagination. And yet most of that fiber sat unused for years because of the ‘last mile’ problem; the companies that had financed it with debt collapsed; and household names like Global Crossing and WorldCom went bankrupt. The destination was exactly as advertised; the path ruined the people who bet on it most aggressively.

Notice that the biggest failures were overwhelmingly money-losing pure-plays, whose entire value rested on a future that had not yet arrived. That is the first answer to our opening question: history does not say you must buy unprofitable front-runners to own a revolution. If anything, it warns that they are the most dangerous way in, because they have the least margin for error when the reckoning comes.

And reckonings are not abstractions. Because speculative businesses burn cash, the danger arrives the moment the spigot runs dry — choked off by rising rates or by a loss of confidence that leaves investors unwilling to keep funding losses. That is what happened when the dot-com bubble burst. Exodus Communications, which built the data centers that hosted the early web — the infrastructure layer of its day — burned through enormous sums and was bankrupt by 2001. Even Netscape, whose browser launched the commercial internet, was outcompeted and absorbed within a few years: building the foundational technology was no guarantee of surviving to profit from it.

We share this not to dampen enthusiasm but to frame it. Even the era's eventual winners could be poor investments at the wrong price — Cisco did not reclaim its 2000 peak for two decades, and Microsoft's stock went nowhere for fourteen years even as its earnings climbed, because it first had to grow into a bubble valuation. So far this year, Nvidia has been the worst performer in the PHLX Semiconductor Index, according to Dow Jones Market Data. The right response to a revolution is neither to chase it nor hide from it, but to participate with discipline — and to assume that, over any ten-year horizon, at least one painful valuation reset lies ahead.

The many ways to "own AI" — and how their risks differ

When clients picture investing in AI, they usually picture a few famous names. In reality, the opportunity is layered, and each layer carries a very different risk.

The first layer is infrastructure

The semiconductors, networking, and data centers that train and run these models. This is Nvidia's world, where the revenue and profits are most real today; the chip and memory makers are selling everything they can produce, and this build-out likely has years left to run. But notice where the risk truly sits. The chipmakers prosper only as long as their customers keep buying — and it is those customers, the large technology companies pouring hundreds of billions into capacity ahead of proven demand, who will be left holding the rapidly depreciating hardware if revenue is slow to arrive. Here the fiber era's difference cuts the wrong way: dark fiber could be "lit up" years later, patient and durable in the ground, whereas today's AI chips may be largely obsolete within a few years, with no second life to redeem them.

The second layer is software and applications

The tools, like those from Microsoft and Google, that put AI to work inside a business. Here we are more optimistic than the prevailing worry suggests. The fear is that AI "agents" that complete tasks on their own will replace workers and make the software they use obsolete. But a powerful countervailing force, first described by the economist William Stanley Jevons in 1865, points the other way: when a technology makes a resource far cheaper to use, we tend to consume much more of it, not less — steam engines made coal more efficient, and Britain burned more of it, not less. Apollo's chief economist Torsten Slok has applied this logic to AI: as the cost of knowledge work falls, demand for it tends to expand rather than collapse. This is the heart of what is sometimes called augmented intelligence — human judgment paired with machine speed — and it may be the most direct and democratic benefit of the revolution: better tools in the hands of ordinary workers, making them more capable rather than redundant. Where that benefit lands as profit — with the software vendors or the far larger universe of businesses using their tools — is an open question, though history suggests much of it accrues to the users.

A third layer most investors overlook: electricity and the grid.

AI runs on power, and its binding constraint is increasingly not chips but electrons — generation, transmission, storage and cooling. Here our sustainability and governance lens becomes an investment lens rather than a values overlay: the most durable participation tends to come through carbon-free baseload and grid modernization — nuclear, transmission, electrification — and favors disciplined operators with long-term contracts over speculative builders. This layer shares the buildout risk of the others, too: capacity is being committed on the assumption that AI's appetite for power only grows, and if that demand disappoints, some of these investments will look premature. It is also where the affordability and reliability of power for ordinary households is becoming a public-policy question — exactly the kind of governance issue we believe requires attention.

The fourth layer is the frontier model labs

Anthropic, OpenAI, and their peers, several now preparing to go public at extraordinary valuations. These may yet become profitable, but today they are the most high-stakes way to own the theme: venture-stage businesses dependent on continuous outside capital, several governed by structures that explicitly place their mission ahead of shareholder returns. Jay Ritter's decades of research on IPOs is a useful check — the most anticipated new listings have, on average, been disappointing long-term holdings. Exciting and investable are not the same thing, a point we previously made about SpaceX.

Where the value actually accrues

Consider now who actually won in the dot-com era. The companies that came to define the internet economy were, for the most part, not the stars of the pre-1999 boom — they were forged in the wreckage that followed. Google did not go public until 2004, two years after the market bottomed; Facebook did not yet exist when the bubble burst; Amazon's world-conquering decade came after its near-death experience, not before. The lesson is humility: the durable winners were not knowable at the peak of enthusiasm. They emerged afterward.

The wealth the internet ultimately created came in two forms. Some was spectacularly concentrated — a handful of technology companies grew into the most valuable enterprises on earth. But, as we just saw, those were not the names you could have bought at the peak; they were a different cohort, small or not yet born, that emerged after the bust, and really took off in the third internet decade. The second form is the one investors most often miss: a vast pool of value that diffused into the broad economy, as thousands of ordinary, profitable companies across every industry used the technology to widen their margins, serve customers better, and out-compete slower rivals. That value never looked like an "internet investment." It looked like good businesses becoming more productive.

If AI spreads faster and deeper than the internet did — and we believe it will — both pools should be larger, not smaller. And here is the part that resolves the tension: you do not have to choose between the concentrated winners and the diffuse ones. Which leads to a conclusion almost too plain to be thrilling: over the long term, the most powerful AI portfolio may not look like an AI portfolio. It may look like broad, disciplined ownership of high-quality, profitable businesses purchased at sensible prices. We don't have to choose between money-losing companies and missing out — profitability is not the obstacle to owning AI, but one of the surest ways to own it well.

Owning the market broadly, with a deliberate tilt toward profitability and value — the premiums documented for decades by Eugene Fama and Kenneth French, and central to our core portfolios — is a way to own the eventual winners and the diffuse adopters without naming them in advance. To be clear about what that tilt does and doesn't do: we still own today's biggest winners — they're enormously profitable, exactly the kind of company the approach favors — but we won't overpay for them. As a stock grows more expensive, our fund partners tend to hold a little less of it rather than dropping it altogether.

An honest word about the current leaders

We owe you candor about the other side. The companies powering this market are not the profitless dreamers of 1999 — the mega-cap leaders are extraordinarily profitable, generating a large share of all American corporate profit, and that performance is real. A value-disciplined approach has trailed them for much of the past fifteen years, and we won't pretend that being early to discipline is costless.

But something beneath those earnings is quietly changing. Two years ago, these were classic software businesses — asset-light, high-margin, and richly valued for exactly that reason. Today, racing to build AI, they are being pulled into capital-heavy infrastructure, spending hundreds of billions a year on data centers, chips, and power — the financial signature of a utility, not a software firm. Free cash flow is consumed by construction, depreciation climbs on fast-aging hardware, and the returns are unproven. It may pay off handsomely, but it is a bet — and a different one than the market previously rewarded.

The risk, then, is not insolvency; these companies are far too profitable for that. It is a re-rating, as the market gradually realizes it paid software prices for what may become infrastructure businesses — the Cisco and Microsoft pattern again, a fine company whose stock goes nowhere for years while its valuation normalizes. Which sharpens the case for diffusion: the ordinary profitable companies that simply use AI to widen their margins reap the benefit without carrying the capital burden.

For clients who want a deliberate tilt

That said, some clients want, with eyes fully open, an explicit overweight to today's AI leaders — and we can build that thoughtfully, as a satellite alongside the disciplined core rather than in place of it. Such a sleeve would draw on the layers above: the software franchises that monetize AI as an enhancement of their users; the platforms that pair proprietary data with the distribution to deploy AI at scale; broad exposure to the semiconductor and infrastructure leaders; and the clean-energy and grid build-out that AI's power demand requires, consistent with our sustainability mandate. All of it would be tailored to your circumstances and reviewed for suitability. If a deliberate AI tilt is something you'd like to explore, we'd welcome the conversation; the value of the satellite approach is that it lets you express conviction in the leaders while the core does its quieter, more durable work.

The bottom line: discipline is participation

We'll leave you with what we find most freeing: being bullish on AI and disciplined about price are not in tension — they are the same thesis. The best "internet portfolio," for someone who lived through it, was a broad book of excellent businesses held through the crash, with reserves to buy when the headlines called it all a bubble. We intend to own the AI future the same way — broad, disciplined ownership, treating any reckoning as an opportunity rather than a threat. The technology will be transformational; that was never the hard part. The hard part is staying invested and disciplined at once — exactly the work we are here to do with you.

Sources

This material is intended for general informational and educational purposes only and does not constitute investment, tax, or legal advice, nor a recommendation to buy or sell any security. Any securities or companies mentioned are referenced solely to illustrate concepts discussed and should not be construed as recommendations; they may or may not be suitable for any particular investor. Investing involves risk, including the potential loss of principal. Diversification does not assure a profit or protect against loss in a declining market. Past performance is not indicative of future results, and forward-looking statements are subject to change. Any tilt toward individual securities or sectors increases concentration risk and is appropriate only as part of a strategy tailored to an individual's goals, risk tolerance, and circumstances. Securities and advisory services offered through Commonwealth Financial Network®, Member FINRA/SIPC, a Registered Investment Adviser. Longwave Financial Partners is a [marketing name / d.b.a.] — please conform to your approved Commonwealth disclosure language before distribution.

Author: Nathan Munits, CRPC®, AIF®

Founder & Senior Financial Advisor, Longwave Financial

Nathan Munits, CRPC®, AIF®, is the Founder and Senior Financial Advisor at Longwave Financial, where he provides independent, fee-based financial planning and investment advice as a fiduciary. Before founding Longwave, Nathan spent 10 years at Ameriprise Financial, where he became one of the youngest advisors to reach Private Wealth Advisor status, the firm's highest tier of recognition.

Nathan holds the Chartered Retirement Planning Counselor (CRPC®) designation from the College for Financial Planning and is an Accredited Investment Fiduciary® (AIF®). He also holds a BA in Philosophy from Queens College. His approach centers on comprehensive financial planning grounded in three principles: independent advice, fiduciary commitment, and long-term client collaboration.

Born in Riga, Latvia, and raised in the US after his family immigrated when he was five, Nathan brings a first-generation perspective to helping clients build lasting financial security.

Maria Andreina Perez