For decades, investors treated “tech stock” and “growth stock” as synonyms. The assumption was so embedded it stopped feeling like an assumption.
Tech companies innovate. Innovation drives revenue expansion. Revenue expansion drives multiple expansion. Therefore, tech equals growth.
That framework worked in the PC era. It worked through internet, mobile, and cloud. Each cycle reinforced the label until the label became the thesis — and investors stopped questioning whether it still applied.
Here is the inversion most portfolios have not priced in: the best growth stocks of the AI era are not tech stocks. They are infrastructure stocks. And what the market still calls “tech” — the software layer — is increasingly the defensive, compressible, at-risk part of the stack.
NVDA, ASML, TSMC, TSLA. These are not tech stocks in the way that term has been used for thirty years. They are the new growth stocks. They just happen to involve hardware.
How the Conflation Happened
Growth stocks historically had a specific structural profile: expanding addressable market, pricing power, recurring revenue, near-zero marginal cost at scale, and a reinvestment flywheel that compounded returns over time.
Software-as-a-service was the cleanest expression of this. Add users. Expand contracts. Add modules. The incremental cost of serving the next customer was close to nothing. Investors paid high multiples not because current earnings were large, but because the duration and scalability of growth felt durable.
Tech was growth because tech created new demand curves — and could serve them cheaply. Software was the medium. Growth was the outcome. The two became synonymous because for twenty years they were.
AI breaks the equation. Not because tech stops growing — but because the thing doing the growing is no longer software.
What AI Actually Is
AI is not another software layer. It is compute-intensive, energy-intensive, and manufacturing-intensive. It pulls the system downward into physical constraints: semiconductors, advanced lithography, packaging capacity, power generation, cooling infrastructure, data center real estate.
When growth becomes constrained by physical inputs rather than user adoption, the assets that matter change.
The marginal dollar of global capital expenditure is no longer buying ad impressions or SaaS seats. It is buying advanced accelerators, EUV lithography equipment, leading-edge foundry capacity, and power infrastructure. That capital is flowing toward a small group of companies that cannot be replicated quickly, cannot be substituted easily, and cannot be scaled without years of investment and physical buildout.
That is the structural profile of a growth stock. It just does not look like one because the sector labels have not caught up.
The New Growth Stocks Look Like Industrials
The companies named here are illustrative of a structural position, not a price recommendation — the thesis is about which layer of the stack matters, not whether any individual name is cheap or expensive today.
NVDA designs the chips that gate AI expansion. ASML builds the only machines capable of producing leading-edge semiconductors at scale — there is no alternative at the frontier. TSMC fabricates the chips that the entire AI stack depends on. TSLA is building the robotics and autonomy layer that sits at the intersection of AI and physical labor.
These companies share the three characteristics that define a structural chokepoint: limited substitutability, long lead times, and systemic dependence. Multiple industries rely on the same inputs. Capacity cannot be expanded quickly. There are few viable alternatives at the frontier.
When those conditions hold, pricing power increases. Margins become durable. Returns concentrate. That is exactly what growth investors were buying in SaaS — except SaaS had low capital intensity and these companies require massive physical investment to maintain their positions.
The difference is that these moats are harder, not softer. A competitor cannot spin up a new ASML in eighteen months. They cannot replicate TSMC’s process technology with a Series A. The barriers are physical, geopolitical, and generational.
If global AI investment doubles but leading-edge chip manufacturing can only expand incrementally, incremental demand pushes against fixed supply. The constraint captures value. This is where equity outperformance concentrates during an industrial buildout — and that is what we are in.
What Happened to Software
Meanwhile, the asset class investors have called “tech” for thirty years — software — is facing a different structural reality.
In prior cycles, building software required engineering teams, sales infrastructure, distribution strategy, and time. Those inputs created moats. Specialization had value because replication was expensive.
Foundational model providers have now collapsed the marginal cost of code generation, workflow automation, content production, and knowledge extraction. A small team with AI assistance can produce what once required a full engineering organization. The moat was always replication cost. When replication cost drops, the moat narrows.
The result is not that software disappears. It polarizes.
The commodity application layer — workflow tools and point solutions that can be replicated or embedded by AI platforms — compresses first. These are the companies still trading on SaaS multiples that the market will slowly reprice.
The systems of record — databases, identity layers, deeply integrated enterprise platforms — retain power because data gravity and switching costs persist. Enterprises do not migrate these lightly.
The control planes — security, orchestration, compliance, governance — remain critical because AI reduces the cost of generating features but does not eliminate the need for accountability, auditability, or regulatory compliance.
Two of those three categories survive. One compresses. But even the survivors are not growing the way infrastructure names are growing. They are enduring. That is a different investment proposition.
The Classification Error in Practice
Markets often misprice assets because they use outdated labels.
Tesla spent years being analyzed as a car company. If it were simply a car company, it would trade like one. The persistent valuation debate around Tesla was largely a classification debate — and investors who resolved it correctly earlier did better than those who did not.
The same dynamic is playing out now, at a broader scale, across the entire stack.
Investors are asking which SaaS company wins AI. That is the wrong question. The better question is which layer becomes structurally indispensable — and whether it is priced accordingly. When semiconductor fabrication capacity determines the pace of AI adoption, advanced manufacturing is part of the technology stack. When electric grid capacity determines data center viability, utilities are strategic infrastructure. The line between tech and industrial is not blurring. It has already moved.
The most important growth assets of the AI era may never appear in a tech ETF. They will appear in industrial indices, in materials, in energy. Investors anchored to the old label will look at ASML and see a manufacturing company. Investors using the right framework will see the only company on earth that can produce the machines required to build the chips that run AI at scale.
Those are not the same asset.
The Counterargument Worth Taking Seriously
AI efficiency is improving faster than most forecasts predicted. Inference costs have dropped dramatically over the past two years. Models are doing more with less compute. If that trajectory continues, physical scarcity at the compute layer loosens faster than this framework implies — and the constraint migrates before the capital cycle fully plays out.
This is how industrial transitions have always worked. The constraint does not hold permanently. Oil replaced coal. Fiber replaced copper. The chokepoint moves.
This thesis is a regime claim, not a permanent one. It argues that during the current buildout phase, physical infrastructure is the binding constraint and capital is concentrating accordingly. The signal to watch is whether AI efficiency gains are outpacing demand growth. So far, demand is winning. But that is the variable that matters, and it deserves to be tracked rather than assumed.
What This Means for Portfolio Construction
For retail investors, the confusion is narrative-based. The “tech equals growth” label is still embedded in index construction, in financial media, in the default mental model. A tech ETF bought as growth exposure may actually be concentrated in the compressible software layer — the part of the stack most at risk from AI commoditization.
For advisors, the confusion is structural. Infrastructure names behave differently from SaaS names. The return profile is different. The duration is different. The risk factors are different. Clients allocated to “tech” as a growth proxy deserve to know which kind of tech they actually own.
The goal is not to abandon quality investing. It is to understand that quality means something different in this regime than it did in the last one.
A good software company executing well may underperform a hardware infrastructure name that simply sits at the constraint. That outcome does not require anyone to make mistakes. It only requires the regime to remain.
The Inversion
The previous decade’s growth stocks were asset-light, software-driven, and globally scalable at near-zero marginal cost. The next decade’s growth stocks are capital-intensive, physically constrained, and impossible to replicate quickly.
One category looks like tech. The other looks like industrials.
The market is slowly figuring out which is which. Portfolios built on the old label — tech equals growth — will feel that repricing before they understand it.
The category “tech stock” is not wrong.
It is just no longer the thing you want to own.