What Was Impossible Is Now Within Reach

Not every analogy holds, but the parallels to the Dotcom Boom are hard to ignore.

In the mid-1990s, the internet was primarily infrastructure. Many companies understood it would matter, but few could explain concretely which products, services, or business models would emerge from it. That uncertainty is exactly why enormous capital flowed into everything remotely connected to the internet. The underlying technological thesis was correct. Nearly every industry was eventually reshaped by digital transformation. But the actual advantage emerged only where companies built concrete applications that people genuinely used.

Infrastructure doesn't win

The biggest winners weren't automatically those with the largest technical infrastructure, but those who built the strongest platforms, interfaces, and user habits. Amazon didn't invent the internet; it built a new commerce model on top of it. Meta scaled digital networks. Uber connected smartphones, GPS, and platform logic. The real value creation came not from the technology itself, but from the ability to turn it into working products, ecosystems, and new behavioral patterns.

AI as a new infrastructure and capital cycle

That's where the actual strength of AI lies today. In his essay "AI eats the world", Benedict Evans describes AI less as a finished product and more as a new infrastructure and capital cycle. According to Evans, the major tech companies plan to spend around $700 billion in capex on AI infrastructure in 2026 alone. Datacenters, chips, energy supply, and compute capacity are currently getting more attention than concrete end products. At the same time, Evans raises the question of whether long-term value creation will actually happen at the model level, or rather where concrete applications get built on top. This is exactly why he draws the analogy to earlier platform shifts: mainframes, PCs, the web, smartphones.

Models are becoming a commodity

His most interesting thesis is that models themselves may increasingly become interchangeable. Evans describes foundation models as a commodity without strong network effects. The real value therefore moves up the stack: into interfaces, workflows, vertical solutions, processes, proprietary data, and concrete use cases. As he puts it: "Innovation will move up the stack." The most interesting companies of the next few years probably won't just build better models. They'll build new kinds of software, new ways of working, and new user experiences.

What was impossible is becoming cheap

The most important question is what new possibilities emerge when certain tasks suddenly become nearly free, infinitely scalable, and always available. Evans describes AI at one point as "infinite interns." That's the origin of new forms of software, new decision processes, and new interfaces. Companies can suddenly analyze millions of data points, generate individualized content, make complex knowledge accessible in real time, or dynamically orchestrate entire workflows. Things that were previously too expensive, too slow, or operationally impossible become economically viable.

The real value doesn't come from automating individual tasks, but from the ability to build new products and user experiences from that capacity. Evans returns to the same question throughout: "What was impossible that now becomes cheap?" That's where the most relevant companies of the next few years will likely emerge. Not through more tokens or larger models, but through new interfaces, new habits, and new forms of value creation that simply weren't possible before.