Who Gets to Own the AI Economy? Krishan Rattan of Terra Invest on AI IPOs

I am interested in AI IPOs because artificial intelligence could create important industries that our existing market categories do not yet describe well. A company that designs new molecules, runs experiments and helps manufacture the resulting products might contain elements of software, biotechnology and industrial infrastructure. The economic opportunity lies in the whole capability. Splitting it into familiar labels can obscure what has become possible.

At Terra Invest, the question I want to pursue is how far that expansion could go. When useful intelligence becomes cheaper and more widely available, businesses can attempt work they previously rejected before it reached a budget. There are scientific questions without research programmes and engineering designs without prototypes. They represent suppressed demand: projects that cannot become markets until the means of pursuing them change.

An existing company’s spending is consequently an incomplete measure of AI’s opportunity. A pharmaceutical research budget tells us what that organisation can currently investigate. It cannot tell us the value of all the plausible therapeutic approaches that better modelling and experimentation might make worth testing. Similar possibilities exist in materials, robotics and infrastructure design. My bullishness comes from that increase in the range of feasible work.

Stanford’s 2025 AI Index provides one historical marker: inference at roughly GPT-3.5 capability became more than 280 times cheaper between November 2022 and October 2024. This does not measure every task or today’s frontier models. It does demonstrate how sharply the cost of a useful capability can move. Once a task becomes affordable, people can attempt it more often, for smaller problems and in entirely different settings.

The important next step is to connect those attempts to reality. A model can propose a material; someone must synthesise it, measure its properties and discover whether it survives the conditions that matter. The resulting data should improve the next proposal. I expect some compelling AI companies to be built around that repeated movement between prediction and experiment. Their advantage will accumulate in reliable outcomes and the knowledge of how to obtain them.

Tesla’s 2023 Master Plan Part 3 is a useful reference here, whatever one makes of Elon Musk’s wider public persona. It considered generation, storage and electrification within one system, and that willingness to reason across an entire production problem is worth learning from. An AI founder should be able to ask which neighbouring constraints must change before a remarkable model can produce a remarkable real-world result.

Integration should follow that reasoning. Sometimes the company needs its own machines and facilities. Sometimes strong partners make it faster and more capable. The vital question is where learning accumulates. If the business cannot see the consequences of its decisions, it will struggle to improve them. If every successful experiment makes its next decision better, it can develop an advantage that extends beyond the model it happens to use.

Google DeepMind offered a useful example in its May 2025 description of AlphaEvolve, which combines generated programs with automated evaluation. The company reported improvements in parts of its computing infrastructure and AI training. I see this as an early instance of AI helping improve the means of producing AI. One successful optimisation can improve the resources available for the next round of work.

That is where the IPO debate becomes interesting to me. Laboratories, specialised equipment and new production systems need substantial capital. Successful businesses may eventually use public markets to scale capabilities that began in research. Investors would then be helping finance the industrial consequences of cheap intelligence, including companies whose assets and products look quite different from a conventional software business.

Ownership deserves attention at the same time. If AI contributes to extraordinary productive gains, more people should have ways to participate in the companies making those gains possible. Public equity can play a role. Access to useful models, scientific infrastructure and opportunities for new founders matters too; those channels determine who can create value in the first place.

My interest in AI IPOs is therefore tied to a much larger ambition. We should build a capital system capable of supporting new forms of discovery and production. The eventual listings will be significant when they give more people a stake in industries that could barely be imagined when those companies began.