The Divergence Between Hyperscalers and Chips
A defining theme in markets right now is the split between the so-called Mag 7 (the megacap hyperscalers) and the chipmakers. The conventional framing pits hyperscalers against chips as if capital must flow to one or the other. But this theme has not fully played out yet, and there is a genuine risk that the hyperscalers take their foot off the capex pedal in the coming quarters.
The central question is whether both groups — the Mag 7/hyperscalers and the chips — can rise together and both be rewarded simultaneously. The answer is yes, but only under a specific condition: if the return on investment (ROI) were concrete and established, if money were pouring in, and if end-user demand were as insatiable as the demand for chips and memory has been on the infrastructure side. In other words, both can win at once only if the downstream economics justify the upstream spending.
The Warning Signs in End-User Demand
The trouble is that the recent reads on end-user demand are not encouraging, and they feed directly into concerns about commoditization and ROI. Two specific data points illustrate the problem:
- A Wall Street Journal report on June 10th indicated that OpenAI is going to be slashing prices to compete with Anthropic, in anticipation that Anthropic will lower its prices.
- On June 16th, reporting suggested Microsoft might be looking into DeepSeek as a way to offer a lower-cost option, apparently because adoption of existing offerings is not that strong.
If these are the genuine signals on end-user demand, then the picture is not great. Price-cutting and the hunt for cheaper alternatives point toward commoditization, which undermines the ROI thesis that has been supporting the entire infrastructure buildout.
Why Capex May Come Down
This leads to a clear logical chain. If the hyperscalers are not going to be able to get the prices they modeled — the revenue they assumed they would earn from end-user demand at the enterprise level, the consumer level, or wherever — then they are likely going to take capex down. And if they take capex down, that immediately raises the question: what happens to the chips? This is precisely why the money may end up flowing to one group or the other rather than lifting both.
Importantly, this does not mean one side is going to crash or that either group is "dead." Rather, it suggests there needs to be some kind of reset of expectations. The reason this matters so much is the sheer scale of the numbers analysts are currently modeling: roughly $750 billion in capex this year and $1.1 trillion next year. The pressing question is what happens to the entire trade if those numbers get challenged.
The Car Maker and Toll Booth Analogy
A useful way to understand the structure of this market is to think of the AI labs as the car makers and the hyperscalers as the toll booth operators. What we are witnessing is the car makers downgrading from the Ferrari to a twelve-year-old Honda — cutting their prices and racing toward cheaper models. The toll booth operator, by contrast, does not care what car you drive; it charges everybody essentially the same. As a result, the toll booth operators' revenues keep going up even as the car makers compete their margins away. This ties into the idea of "token maxing" — the dynamics of token pricing and consumption.
The Race for the Best Model — and Whether It Matters
Underlying all of this is the assumption embedded in the industry's spending race. Mark Zuckerberg's comments at Meta capture it: he would rather overspend than miss out on this opportunity. That is a bold statement on Wall Street, which generally dislikes heavy spending. The whole point of the race is to have the best model.
But you do not necessarily need a Ferrari. Maybe you just need a bike — something that gets you through. This raises the question of where the pricing equilibrium lies, the point where demand and supply meet. It may turn out that frontier models do not need to be the absolute best for most use cases.
There is an important caveat here, because AI is a very broad topic. In certain domains — biotech, coding, and other specialized industries — the best frontier model genuinely matters. But at the enterprise level (for example, a Copilot-style product) or the consumer level, the key variable is what the customer is willing to pay. The difference between a customer paying a $100 subscription and a $20 subscription is enormous when it comes to modeling future revenue projections. That gap is what makes the revenue assumptions behind the capex so fragile.
Something Has to Give
Putting it together, something has to give. The market already experienced its DeepSeek scare and has weathered ongoing discussions about ROI and commoditization. Up to now, investors have been willing to wave these concerns away — to say the returns are coming, they are down the line, let's back-burner that discussion. But at some point, the persistent underperformance of the Mag 7 forces the issue, and the market is beginning to get a little antsy around these topics. (The Mag 7 has, this quarter, arguably earned the nickname the "lag seven.")
The Extremes and Divergences of the Quarter
This period has been defined by striking milestones and divergences, even within the tech space:
- The chips (the SOX/semiconductor index) are up over 80% — its best ever.
- Microsoft posted its worst performance since 2020.
- The Russell 2000 could be looking at its best first half since 1991.
- Gold, Bitcoin, and dollar-yen have all been heading in the other direction.
The explanation begins with the largest capex binge in history. That spending acts as a fundamental driver, putting a bid underneath the market and supporting EPS growth, which keeps the bias tilted to the upside.
Conditioned to Ignore the Headlines
On top of this fundamental support sits a behavioral pattern: markets have been conditioned to ignore the headlines as long as EPS growth continues to track at around 20%. Whether it was DeepSeek, the tariff tantrum of last year, or the Iran war, the market has learned to look past the noise. Even now, with tensions in the Middle East and uncertainty over whether parties are even meeting, oil prices remain very well behaved — which is precisely why the market continues to perform.
The open worry is whether investors have grown too complacent, waiting only for earnings or hard data to validate something while ignoring all headlines in the meantime.
Are We in a Bubble?
It would not be surprising if the market is somewhere in the third stage of a bubble. Financial professionals are trained on the five stages of a bubble:
1. The breakthrough in technology or infrastructure.
2. The capital investment / capex binge.
3. FOMO (fear of missing out).
4. Insider selling.
5. Capitulation — everybody gets out, and that is the bust.
On this framework, the market appears to be somewhere in the FOMO stage. The catch is that, just like the dot-com bubble, this phase can run on for a couple of years, so investors have to be mindful of that timing risk rather than assuming an imminent collapse.
A Unique and Strong Period
Despite the bubble concerns, this is a very unique and very strong economic period, underpinned by a genuine technical renaissance. That renaissance is not just AI — it spans autonomous systems, robotics, and space, among other areas. It is, in short, an interesting and consequential time to be living through, even as the fundamental question of AI ROI remains unresolved and increasingly pressing for the megacap names and their investors.
