The world is obsessed with AI demand, but quality investors should be focusing on AI supply.
If you listen to the quarterly earnings calls of the Magnificent 7 in 2026, the narrative is near-unanimous: the risk of under-investing is far greater than the risk of over-investing. This sentiment has fueled a historic arms race, with annual CAPEX across the tech giants now soaring toward the half-trillion-dollar mark.
To the average growth investor, this looks like a bold bet on the future. But to a student of the Capital Cycle, it looks like a classic siren song. History is littered with revolutionary technologies that followed a predictable & painful pattern. When capital floods a sector indiscriminately, the result is rarely a sustained moat. Instead, it creates a glut of supply that triggers commoditization, erodes pricing power, and ultimately collapses Return on Invested Capital (ROIC).
In this issue of Quality Stocks, we are stepping away from the excitement of what AI can do to look at the math of what the current investment binge will do. We are exploring the supply-side trap of the AI revolution and why the most innovative companies often make the poorest long-term investments.
To be clear: I am a firm believer in AIâs transformative power. The demand is undeniable. However, my background in Supply Chain management has taught me to spot the Bullwhip Effect long before it hits the balance sheet. Just as distorted demand signals can wreck a global supply chain, overinvestment in AI could create excess capacity. For over 2 years, this newsletter has been about avoiding traps. Today is no exception!
The engine of overinvestment: understanding the capital cycle
Here is the first section of your article. It establishes the âCapital Cycleâ framework and uses the historical parallel of the fiber-optic bubble to show your readers why âbelieving in the techâ isnât enough to justify the investment.
The Engine of Overinvestment â Understanding the Capital Cycle
To understand why massive investment can be a red flag, we must look at the âCapital Cycleâ, a framework made famous by financial historian Edward Chancellor.
Most analysts spend 90% of their time obsessing over demand: How many people will use ChatGPT? How many enterprises will integrate Copilot? But for a Quality investor, another important question is almost always about supply.
The Capital Cycle is a simple, recurring psychological and economic loop:
1. The flare. An industry enjoys high returns on capital (ROIC), often due to a new technology
The rush. These high returns act as a flare, attracting a flood of capital from CEO, VC and Wall Street
The glut: This capital is used to expand capacity (more factories, more chips, more data centers)
The crash. Eventually, supply outstrips demand. To fill that excess capacity, companies slash prices. Profits vanish, capital flees and only then (when the âweak handsâ are liquidated) does the cycle start over, leaving survivors more powerful
As Chancellor notes: âhigh returns tend to attract capital, just as low returns repel itâ. Today, we watch hundreds of billions pour into AI shovels.
Is AI spending excessive?
In supply chain management, we often talk about the Bullwhip Effect: how a small nudge in consumer demand can lead to massive, distorted over-ordering at the top of the chain. In 2026, we are not just seeing a nudge, we are witnessing a global scramble for capacity that has no historical equal.
To understand the scale, look at the CAPEX guidance for the âBig Fiveâ (Amazon, Microsoft, Google, Meta, and Oracle) for this year. Collectively, they are projected to spend a staggering $650B on infrastructure.
For over a decade, the âQualityâ thesis for Big Tech was built on their asset-light nature. They generated billions in cash because they did not need to build massive physical networks. That era is over. The new reality: by early 2026, these companies are spending nearly 65% of their gross profit on Capex, a level of capital intensity comparable to ExxonMobil or Ford.
Why are they spending this much? Because they have entered a prisonerâs dilemma. If Microsoft stops buying H200s or B200s, they risk losing their cloud lead to Google or Amazon. They are spending to stay relevant.
In a capital cycle, when investment is driven by the fear of being left behind rather than clear, incremental Return on Invested Capital (ROIC), the cycle is nearing its peak. As we move into 2026, the gap is widening: CAPEX is growing at 30â50% while the AI-enabled revenue at the software layer is struggling to keep pace.
The commoditization trap
The first sign of the trap is the Price War. As long as volume increases very fast, this is not an issue, but it could squeeze margins in the long-term. Efficiency gains have driven AI token costs down by over 70% annually (maybe just a sign that we are at the beginning of the learning curve).
Chinese models like DeepSeek and open-weight giants like Llama 4 have broken the premium paywall. Western incumbents are being forced to slash prices just to maintain market share: OpenAI recently implemented an 80% price cut on flagship APIs.
For the Hyperscalers (Microsoft, Google, Amazon), this is a nightmare for margins. They are spending record sums on infrastructure while the unit price of their output is plummeting. When margins are squeezed at the top, the first thing to be cut is the next round of capital investment.
Note: we are likely witnessing a fundamental shift in the AI value chain. The capital cycle suggests that money will not be made in the models (which are being commoditized), but in the application. We will therefore see the revenge of the SaaS. The ultimate winners might be the software companies that leverage cheap, standardized AI to deliver massive value to end-users.
The Whiplash for Equipment Providers
The real first danger of the capital cycle lies with the companies that provide the tools for the shovels:the semiconductor equipment manufacturers like ASML, Applied Materials (AMAT), Lam Research (LRCX) and KLA Corporation (KLAC).
For now, orders are massive, driven by the massive capacity expansion of TSMC and Intel. On the surface, this looks like a secular bull market. But through a supply-chain lens, it looks like a classic shortage gaming peak.
When a chipmaker like TSMC or Samsung fears they wonât have enough capacity for the next generation of Nvidiaâs Rubin or Blackwell chips, they donât just order what they need, they order for future needs based on current forecast (that are rather optimistic). This creates a demand cliff.
The equipment providers (at the very top of the whip) will feel the snap first and hardest. History tells us that in a capital cycle, the companies with the most impressive backlogs today are often the ones with the most painful cancellations tomorrow.
The cyclical nature of equipment providers (ASML, AMAT, etc) is a masterclass in operational leverage. A 20% decline in end-user demand does not lead to a 20% revenue dip for the toolmakers. It leads to a total stop of new investment as customers digest their over-ordered capacity.
Spotting the survivors
To find the stocks that can weather the capital cycle whiplash, I am looking for 3 specific traits.
The first trait to look for is deep vertical integration, a strategy exemplified by giants like Alphabet and Microsoft. Rather than being at the mercy of the capital cycleâs whiplash, these companies have spent the last few years building closed loops. By designing their own silicon, such as Googleâs TPUs and owning the massive distribution networks of Search, Office and Cloud, they have effectively reduced their dependency from where the price wars are happening: at the model layer.
The second trait marks what I call the real value providers. While the market was obsessed with who would build the biggest LLM, the true quality survivors may be the software companies that will successfully moved into the deployment phase. These firms donât need to win the intelligence race because they will win the workflow race. By embedding cheap, standardized AI into existing enterprise systems like SAP, ServiceNow or Dassault Systemes, they turn a volatile technology into a sticky & high-margin utility. Another value provider comes from the physical bottlenecks of the cycle, specifically the energy and ressource providers. While GPU orders can be canceled in a heartbeat due to the bullwhip effect, the structural demand for electricity to run the worldâs data centers is only moving in one direction
Conclusion
We are currently standing at a crossroads where technological potential meets economic reality. There is little doubt that AI represents a generational shift: a transformative force that will reshape how we live, work and create. However, as investors, we must decouple our enthusiasm for the technology from our discipline regarding the capital cycle. History suggests that the more revolutionary a technology feels, the more likely it is to attract the kind of undisciplined capital that destroys long-term returns and create risks.
Risks are high that the next few years will be defined by disillusionment for the some actors. As the Bullwhip Effect snaps and the Price War at the model layer intensifies, many of todayâs market darlings will find that their massive CAPEX bills have become anchors rather than engines. The winners of the infrastructure phase will probably not be the winners of the deployment phase: the risk being on the equipment providers.
To succeed in this environment requires a contrarian mindset: the courage to look past the current order backlogs and focus on who actually owns the customer, the energy and the workflow.
The capital cycle never lies. Belief in the technology and over-investment in supply are perfectly compatible. Where most investors go wrong is assuming that revolutionary potential justifies indiscriminate capital deployment. The survivors won't be the ones with the biggest backlogs. They'll be the ones who understood that conviction without discipline breeds casualties, not returns.
Great capital cycle framing. I think there's a second-order effect here: if AI capex commoditizes returns at the infrastructure layer while simultaneously repricing the software layer downward, where does capital go? The quiet beneficiary might be boring physical-world compounders â waste, utilities, rail â that neither AI builders nor AI disruptors can touch. Both sides of the cycle push capital toward the same place.