The debate about AI has gone tribal, and I don't think that serves anyone who actually manages money. On one side you've got the optimists, convinced AI will remake the economy (and take a lot of jobs with it on the way). On the other, we have the Ed Zitron’s of the world saying it’s all a longcon propped up by media laziness and cult-of-personality CEOs, with an imminent sonic boom-level bubble pop incoming.
The stakes are a good deal higher than they were a year or two ago. A surprising amount of the economy (and of everyday sentiment) now rides on how this all shakes out.
Certainty about bubbles is usually where smart investors get themselves into trouble. That’s why we’re not ready to call one either way… but here’s everything we’re watching that we think will prove instructive in that particular call.
In this piece:
- The two tribes
- What my first year in the business taught me
- Where this rhymes with the dot-com era
- What looks different this time
- The mileposts I'm watching
- Positioning without picking a side
- Where I actually land
For the full walk-through, the video below is the deep dive. If you'd rather have the cliff's notes, keep reading.
The two tribes, and a number worth watching
Start with one figure. Jason Furman, the Harvard economist, thinks something like 90% of US GDP growth right now is tied to the AI buildout. Morgan Stanley pegs combined AI CapEx near $800 billion this year, climbing past $1.1 trillion next year. That is enormous.
Most of that growth is capacity build: money spent against expected demand rather than productivity companies are already banking from AI use. And capacity build has a way of being one-time in nature. We think the demand keeps growing... but if supply ever outruns it, that spending often doesn't just dip a little. It can go away for a stretch.
Meanwhile the economy underneath looks K-shaped. AI-related valuations print new highs while University of Michigan consumer sentiment sits near the lowest readings on record. Sentiment toward AI itself has turned, too. Commencement speakers are getting booed for bringing it up, the Pope has weighed in (reminding us to bring our humanity into the AI conversation), and I hear about the power draw and the environmental cost from my own kids at the dinner table. Even though it might not always seem like it does, consumer sentiment does drive political outcomes, and elected officials often do listen to their constituents.
What my first year in the business taught me
I started in this business as an intern in the summer of 2000, right into the back end of the internet bubble and the recovery that followed. What I remember most is how obvious it all looks in hindsight, and how genuinely hard it was to see in the moment. The indicators of false demand were sitting in plain sight, and almost nobody could tell which one would turn.
Calling a bubble is close to impossible, and I'd be suspicious of anyone who tells you otherwise. Humans have a record going back to tulips of people getting excited about the next thing, inflating a bubble, and then getting the timing of the pop completely wrong. My favorite example is Stanley Druckenmiller, one of the great investors of his generation, who cut it wrong on the way up shorting, and then capitulated at almost exactly the wrong moment in 2000 and lost money as it burst too. Bubbles have this way of making even brilliant, patient, long-term investors look foolish.
Where this rhymes with the dot-com era
A few things do rhyme here...
- Any move toward AI, any announcement of a new initiative, drives an outsized swing in a company's valuation.
- Capital is being burned at a staggering pace, especially at OpenAI and Anthropic, on the promise of a business model that justifies the spend later.
- Usage is subsidized: a lot of today's demand is a response to pricing that doesn't cover the cost of delivering the service (the Uber playbook, where venture money buys you customers you monetize once they're locked in).
- And there's circular financing worth watching: Oracle’s leveraging its future with debt tied to AI customer contracts while Nvidia invests in nearly every AI startup that then turns around and buys Nvidia chips.
So one question genuinely stumps me: how do you tell a genuine demand curve from a subsidized one before the subsidy ends?
What looks different this time
Plenty rhymes, but the structure isn't identical to 2000. The most important difference is that the hyperscalers are funding much of this CapEx out of real cash flow, not just venture money. The majority of the investment in the space is coming from well-established, cash-generating businesses. That is legit different.
The effect on their legacy operations is noteworthy, too. I've followed Google since before its IPO (and was laughed out of the room when I proclaimed it would one day eclipse Yahoo), and the fear was that AI would eat the traditional ad business. Instead, Google's use of AI accelerated ad revenue growth, with that huge business growing almost 15% in a single quarter. That smacks of genuine demand here, with serious money behind it, at scale and speed. You can't simply sit it out, particularly if you're a financial advisor and this exposure already lives in your clients' portfolios.
The mileposts I'm watching
I'm not ready to call whether this is a bubble. What I can do is name the mileposts my team and I are tracking to eventually answer the question. There are three:
- Demand
- Supply
- Capital
On demand, the question is stickiness. Does enterprise usage hold once the inference subsidies end? Earlier this year Uber's CTO mentioned they'd blown through their entire annual AI budget in four months, and there are plenty of reports of individual employees spending more on tokens than their salaries. It isn't yet clear that any of that is turning into a return.
My own prediction: where enterprise software is easily rebuilt with AI, we'll see price cuts on those contracts rather than wholesale replacement, and where the tools are genuinely hard to replicate (CRM is the example I keep using), companies will try, bang their heads against the wall for a while, and keep paying for the thing that works.
On supply, the binding constraint is power. The IEA figures roughly 20% of the next couple of years' data center builds are at risk from grid capacity alone. A joint Ohio State and Meta study puts required generation growth from just over 100 terawatt-hours in 2024 to nearly 300 by 2030. Closer to home for us, the state of New Mexico uses a little over two gigawatts of power for everything, residential and industrial. OpenAI already uses about that much, and its own projections get it to 30 gigawatts by 2030. That is a staggering ask, and it sits on top of a semiconductor cycle that has always been brutally cyclical (and the equipment that builds the chips is several times more cyclical than that).
On capital, it comes down to whether the money is there. A year ago the AI 2027 project made predictions I thought were absurd, including CapEx approaching a trillion dollars and a leading lab valued near a trillion. We're nearly at both. Sarah Friar, OpenAI's CFO (and one of my favorite tech analysts back in her Goldman days), has been open about the difficulty of closing the funding gap, and has even questioned whether the business is ready to go public. Watch the IPOs. If demand for the first big one out of the gate cracks, it will tell you a lot about whether the window stays open for trillion-dollar listings behind it.
Positioning without picking a side
What does this mean for a portfolio? First, if you own the major indices, you already own a lot of this. The top ten names in the S&P 500 make up around 40% of the index by weight, so you're covered in the optimistic case where these companies keep running.
The harder half is the downside. That argues, in my opinion, for real diversification: fixed income that stays plain rather than reaching into private credit, and a hard look at international and small- and mid-cap names, where good businesses that aren't AI plays have seen their valuations compress. We think that compression is an opportunity that savvy active managers can potentially capitalize on.
The analysis isn’t the hardest part of this though… it’s the psychology.
There's a YouTube clip of two capuchin monkeys doing the same task. They both get a walnut for completing the task, and both are thrilled to keep doing it. That is until one of the monkeys watches the other one get a grape for identical work. Suddenly the walnut is an insult, and the first monkey is throwing it back at the researcher. That's keeping up with the Joneses, and it's exactly what happens when a client's neighbor is bragging about his AI gains. Sticking to diversification gets harder the higher valuations climb, and the pressure to pile in peaks right at the top.
Where I actually land
My own view, without hedging: I'm an AI optimist. What I see around me is talented people using these tools to get more done, not less, and doing better work with the tools than the tools do alone. Nobody I know who has really dug in is working a 20-hour week. They're chipping away at an infinite backlog they could never reach before. I think we're on the cusp of real advances in healthcare and elsewhere, and I think people need meaningful work to be happy, so I expect new and different jobs on the other side of this, not a hollowed-out economy.
That doesn't mean the ride is smooth. There will be a period where it feels nasty, I promise you that. I just don't believe anyone can reliably time it. So the work is to stay curious and skeptical, to keep informed, and to refuse the tribalism, because the truth almost always lives in the nuance in the middle. Picking a tribe was never the job. Doing the work is, understanding the nuance and managing the money with our eyes open.
IMPORTANT DISCLOSURES
The information contained in this communication has been designed for general informational, illustrative, and educational purposes only and does not constitute an offer to sell or a solicitation of an offer to buy any security. Moreover, the information provided is not intended to provide any investment advice whatsoever. Different types of investments involve varying degrees of risk, and there can be no assurance that the future performance of any specific investment, investment strategy, or product, or any non-investment related content, made reference to directly or indirectly in this communication will be profitable, equal any corresponding indicated historical performance level(s), be suitable for your portfolio or individual situation or prove successful. Due to various factors, including changing market conditions and/or applicable laws, the content may no longer be reflective of current opinions or positions. No discussion or information contained herein serves as the provision of, or as a substitute for, personalized investment advice. To the extent that a reader has any questions regarding the applicability above to his/her individual situation of any specific issue discussed, he/she is encouraged to consult with the professional advisor of his/her choosing. City Different Investments is neither a law firm nor a certified public accounting firm and no portion of this content should be construed as legal, tax, or accounting advice.
The presented information and statistics have been obtained from sources we believe to be reliable but cannot be guaranteed. Any projections, market outlooks or forecasts discussed herein are forward-looking statements and are based upon certain assumptions. Other events that were not taken into account may occur and may significantly affect the returns or performance of these investments. Any projections, outlooks or assumptions should not be construed to be indicative of the actual events which will occur. These projections, market outlooks or estimates are subject to change without notice. Please keep in mind that past performance may not be indicative of future results.