There’s a temptation right now to look at the valuations of companies like NVIDIA, AMD, Broadcom, TSMC, Microsoft, and the younger AI platform players and assume that the market has slipped back into some overcaffeinated version of 1999. People see fast-rising charts and instinctively reach for the word “bubble.” But if you scratch beneath the surface, the comparison falls apart almost immediately. AI today is not a speculative promise built on imaginative future usage. It is already embedded in enterprise workflows, reshaping the semiconductor supply chain, changing software business models, and reorganizing cloud infrastructure economics. The revenue streams tied to AI are material, recurring, and expanding. And more importantly, we are still in the first innings of infrastructure build-out. The investments to support global-scale inference and training have barely begun to mature.
An easy way to see why this is not a bubble is to follow the physical world. Every major AI breakthrough ends up constrained by compute, not imagination. The world is currently bottlenecked by GPUs, networking fabric, advanced packaging, and power delivery. Hyperscalers are hunting for land near hydro and nuclear power sources. Nations are securing supply chains with the seriousness usually reserved for energy or defense procurement. These are not behaviors consistent with speculative tech froth. They are the moves of governments and corporations preparing for a generational technological platform shift. When the bottleneck is physical, not psychological, it changes the story. The value is grounded in scarcity and capacity, not hype. If anything, the market is still underestimating the scale of the build-out needed for AI to reach full commercial utility across sectors like healthcare, transportation, finance, manufacturing, and energy grids.
And then there’s the demand side. We are only beginning to understand what happens when language models become reasoning models, when multi-agent systems move from toy experiments to autonomous workflows, when robotics integrates with foundation models, when enterprise software becomes orchestration rather than manual interface. Every time the tooling gets better, new categories of economic activity appear. Electricity didn’t just improve lamps; it created entire industries that didn’t previously exist. The internet didn’t just digitize newspapers; it rewrote the logic of distribution and discovery. AI won’t merely improve existing workflows. It will replace them, collapse them, expand them, and generate entirely new configurations of labor. That kind of shift takes decades to play out. We’re barely past year two.
Valuation anxiety usually comes from anchoring to historical comps, but there is no historical comp that maps cleanly. The closest might be the PC revolution plus cloud computing plus the smartphone era, all happening at once but moving faster. Productivity gains from AI are not theoretical; even the messy early-stage tools used today are cutting man-hours and accelerating cycle times in software, chip design, drug discovery, logistics modeling, and real-time industrial monitoring. Once the interfaces smooth out and models become more reliable, adoption will accelerate across every knowledge-heavy sector. The total addressable market is effectively “the global economy,” which is an uncomfortable sentence, but it is what it is. Markets have trouble pricing something that rewrites the denominator.
So when people say AI stocks are inflated, what they’re really signaling is that their frame of reference is too small. They’re comparing this moment to the top of something, when it is actually the bottom. The phase we are in now is the wiring of the foundation: data centers, custom silicon, model architectures, distributed training networks, inference optimization. The returns so far are just the early glimmers of what happens when that foundation becomes reliable, cheap, and ubiquitous. The supercycle is not about this quarter or even this decade—it’s about the restructuring of economic production. If investors think today’s valuations are the final form, they are misunderstanding the scale of the shift. We are not witnessing the peak of AI. We are watching the basement being poured.
