This is the final installment in a three-part series. In Parts 1 and 2, I trace how compute, data, distribution and regulation are already forming decisive moats in AI. Catch up here:
Read Part 1: The Compute Chokehold & Who Owns the Corpus
Read Part 2: The Model Is Not the Moat: Distribution, Regulation & Open-Source Mirage
The Great Bifurcation
The endgame is a two-tier market: a handful of companies who own the infrastructure, and everyone else paying rent. This is reinforced by a circular economy of compute, where the same handful of frontier AI labs, hyperscalers and chipmakers make reciprocal commitments to one another: equity stakes, multi-billion-dollar GPU orders, cloud contracts, and coinvestment deals that keep capital, chips and revenue cycling inside the same small loop. In that loop, Nvidia finances the labs that commit to buy its accelerators, hyperscalers finance and host those labs on their clouds, and all three jointly lock up the power, data and distribution needed to stay permanently in Tier 1. As those loops tighten, the outcome is inevitable. Competition slows, rent-seeking starts, prices rise, and the platforms extract more value than they create.
We’ve seen this movie before. The precursor already played out with cloud computing, app stores, and social media. But AI is different. AI is becoming the substrate for everything: decision-making, knowledge work, creative output, scientific research. If three or four companies control that layer, they don’t just control a market. They control the infrastructure of thought.
The existential risk here is something more banal than killer robots. A technofeudal economy where the rest of us are serfs working on someone else’s digital estate. We will till the fields, generating data, building applications, creating content, while the platform lords collect the rent. Indefinitely. As The Wall Street Journal writes in “Why the Tech World Thinks the American Dream Is Dying,” many Silicon Valley executives now openly frame this AI boom as the last chance to amass generational wealth before AI becomes the dominant substrate of economic life and makes traditional paths to prosperity effectively unreachable for everyone else. That logic turns the American dream into a closing window: if you are not building and owning the platforms now, you risk being permanently relegated to working their estates later.
What I’d Tell a Boardroom
So what do you do with all of this if you’re a founder or business leader?
First, for founders building AI companies: Don’t compete on model quality. You’ll lose. Compete on data, on vertical and domain expertise, on distribution, on speed. Find a niche the big labs don’t care about yet. If you’re in healthcare or life sciences, that usually means competing on privileged data, clinical workflow depth, or distribution into existing provider and payer systems, not on raw model performance. Assume the current leaders will stay leaders. Build in a way that works even if OpenAI, Anthropic, and Google continue to dominate. That might mean building on their APIs. That might mean targeting markets they’re ignoring. But don’t bet your company on them stumbling.
Second, for leaders adopting AI into their organizations: Own your data layer and build for portability. The structural dependency this essay describes isn’t just a macro trend; it’s a procurement risk. If your workflows, your clinical data, your decision-making infrastructure all route through a single provider’s stack, you’ve handed them leverage you won’t easily claw back. Diversify your vendor relationships, keep proprietary data yours, and architect for switching costs you can live with. The platforms are architecting for switching costs you can’t.
Third: Act now. The current generation of frontier models (GPT-5.4, Claude Opus 4.6, Gemini 3 Pro) can handle complex, multi-step workflows across software engineering, legal analysis, clinical operations, and finance. Anthropic’s CEO confirmed that over 90% of new code at the company is now authored autonomously by AI agents. This is not a technology to evaluate next quarter. Organizations building AI into their operations today are the ones that will retain control over how they adopt, which platforms they depend on, and what data stays theirs. Wait, and you adopt on someone else’s terms.
Fourth: Be skeptical of narratives that benefit the people telling them. When a company that just raised $110 billion at a post-money valuation of $840 billion tells you models are commoditizing, ask yourself who benefits from you believing that.
The Bottom Line
The model is not the moat. That much is true. But the implication most people draw from it, that the market will remain open and competitive, is the part that’s misleading.
Follow the money. It is already concentrating in a small set of firms whose executives, as Tim Higgins notes in the WSJ, increasingly frame this AI wave as the last opportunity to lock in generational wealth before the ladder is pulled up and AI becomes the dominant substrate for everything else. Worldwide AI spending is projected to hit $2.5 trillion this year alone, more than thirteen times the inflation-adjusted cost of the entire Apollo program. That capital is flowing overwhelmingly to a handful of companies building what they believe will be the operating system of the next economy.
The moats are being dug right now. The question should no longer be whether AI will reshape the economy, but whether that reshaping will serve the many or the few. What we are witnessing is a civilizational-scale restructuring of how knowledge is created, accessed, and controlled. Right now, that architecture is being built to serve the few. While most of us are stuck debating whether AI is smart enough, the few who know it is are already locking up control over who benefits.
This is the third and final part of a three-part series.
This concludes my three-part series on AI's real competitive barriers: where they're being built, who benefits, and what founders and business leaders should do about it.
Part 1: The Compute Chokehold & Who Owns the Corpus
Part 2: The Model Is Not the Moat: Distribution, Regulation & Open-Source Mirage
If you found this article valuable, share it with a founder, investor or executive navigating the AI landscape.
The sharpest point here is that AI concentration is not just a technology story but a market structure story. Even if models commoditise, the firms that control infrastructure, distribution, and workflow integration will still capture most of the economics. Boards should treat platform dependency, switching costs, and data control as strategic risks now, because later may be too late.