TL;DR: The dominant narratives around AI—Silicon Valley techno-optimism, open-source utopianism, accelerationist maximalism—are built on sand. They assume that artificial intelligence is another iteration of the consumer internet, to be conquered by founders, turbocharged by venture capital, and scaled through APIs. But this vision is collapsing under the weight of geopolitical reality and infrastructural constraint.
Here is a contrarian but increasingly correct model of where AI is actually heading:
1. The Predator Platform
Large language models are not platforms in the way AWS or iOS are platforms. They are agglomerative engines: they absorb functionality, collapse verticals, and eat the very ecosystem that tries to build on top of them. They are intelligent interfaces that continuously cannibalize the cognitive labor of their users.
This makes startups structurally fragile. Wrapper apps and thin UX layers built atop GPT-4 or Claude will be disintermediated the moment the model itself replicates their functionality. The venture thesis that you can simply build a “better frontend” to the model is mostly a mirage.
Just as Google in the 2000s centralized attention and monetized discovery, the frontier model providers today are predatory intelligence infrastructures. They’re not platforms to build on. They’re organisms that consume and subsume.
2. AI as Dual-Use: The State Awakens
AI is no longer just a general-purpose technology. It is a dual-use technology, increasingly seen as strategic infrastructure. As U.S.–China AI parity tightens, expect a rapid convergence of compute, data, and model control under the auspices of national security.
In practice, this means:
Tight restrictions on export of high-end compute (already underway)
Classification of certain models as controlled technologies
Alignment of frontier AI development with defense priorities
Startups hoping to thrive in this environment will face the same dilemma that biotech startups faced during wartime or that space startups face under ITAR: either work directly with the government or be frozen out.
The U.S. state, once hands-off and deferential to tech, is asserting itself as sovereign arbiter. And it's just getting started.
3. The Vance Doctrine: Postliberal AI
The 2028 U.S. presidential election is, as of this writing, J.D. Vance’s to lose. His brand of nationalist-populism implies a much different orientation to AI:
State-led development in the name of national unity
Suppression of open models under the banner of social cohesion
Protection of American labor from mass automation
Integration of AI with the defense-industrial base
This is not a retreat from AI, but its domestication by the state. Expect a fusion of Silicon Valley technical talent and D.C. strategic priorities. DARPA becomes the new Y Combinator. Military alignment replaces product-market fit.
And the venture capital model? Collateral damage.
4. Power: The Final Boss
The true constraint on AI isn’t talent. It’s not even GPUs. It’s electricity.
Training frontier models and hosting them for real-time inference at scale requires a new class of infrastructure: hyperscale data centers plugged into multi-gigawatt baseload power. That means:
Building transmission lines across state borders
Accelerating nuclear, geothermal, and advanced fossil projects
Breaking the regulatory death grip of local environmental veto players
States that treat AI as a power-and-concrete problem, not a software abstraction, will win. That means Texas, not California. Saudi Arabia, not Germany. The Dakotas and Appalachia, not Silicon Valley.
The AI wars will not be fought in code. They’ll be fought in zoning boards, utility commissions, and substation permits.
Toward an AI Realist Agenda
If this framework holds, we can expect the following over the next 5–10 years:
Decoupling from the VC ecosystem: As sovereign interests dominate, the venture model becomes increasingly irrelevant outside of state-adjacent projects.
Collapse of open-source frontier AI: The most capable models will become state secrets. Local LLMs may proliferate, but frontier performance will be gated.
Rise of defense-tech industrial policy: The center of gravity shifts to DC. Think Lockheed + OpenAI hybrids. Think sovereign chips and sovereign data.
Infrastructure nationalism: The winners of AI are those who master power, land, cooling, and bandwidth. The rest are tenants of empires they do not control.
Red state ascendancy: Jurisdictions with low regulatory friction, cheap land, and abundant power will become the new centers of AI development.
Most of Silicon Valley is still sleepwalking through this shift. They are optimists building in a world that increasingly demands realists. The AI future will not be a playground. It will be a fortress.
Founders who want to survive must adapt. Those who want to win must align with sovereignty. And those who want to pretend that code is apolitical should start preparing for obsolescence.
Welcome to the age of AI realism.
I like this article and your prognosis is spot on