End Game Positions
What's actually scarce when intelligence isn't.
Yesterday Will Manidis wrote a great piece called End Game Play that I haven’t stopped thinking about. His point: across chess, war, and technology, the middlegame has collapsed. Everyone reasons backward from the terminal state. You play for the endgame. That framing applies to AI more than anything else right now.
Here’s my thought experiment. Whether it takes five years or fifty, assume we get there. AI works 24/7. Runs 100x faster. Costs 100x less. Can iterate toward decade-long goals. Can write any code, do any analysis, produce any content, solve any problem.
Intelligence is abundant. What positions do you want to hold in that endgame?
Energy. AI needs power to think. Factories need power to build. The more capable AI becomes, the more energy it consumes. You cannot think your way to more electrons. Energy is the ultimate bottleneck of the endgame, and it grows with every advance in capability. Endgame Position Examples: NextEra, Base Power, Oklo, anyone generating, transmitting, storing, or building new sources of cheap, abundant power.
Atoms. You can’t download steel. You can’t email a semiconductor. Physical matter requires physical processes, physical supply chains, and physical time. Even with perfect intelligence directing perfect robots, you need raw materials and factories. Endgame Positions: TSMC, ASML, Nvidia, Tesla, and anyone building in the physical world.
Capital. Physical things cost real money. A chip fab is $20B. A nuclear plant is $10B. A satellite constellation is billions more. When the endgame demands atoms and energy at massive scale, the ability to finance and deploy capital becomes a moat in itself. Endgame Positions: JPM, BlackRock, Goldman Sachs, any business that can marshal large-scale capital for physical infrastructure.
Regulatory permission. Governments move at the speed of politics, not technology. AI compresses everything except the regulatory timeline. If anything, regulators slow down as capability increases because the stakes get higher. Permission to operate becomes one of the most durable moats in the economy. Policy will evolve, but the principle persists: you need a license to touch money, health, and critical infrastructure. Endgame Positions: SpaceX, Anduril, Visa, Mastercard, Stripe, and anyone already licensed to operate in regulated industries.
Trust and accountability. When AI can do anything, the question becomes who’s responsible when it goes wrong. Someone has to be on the hook. AI does the work, but a human or entity bears the consequences. Doctors still need to be accredited. Lawyers still need to pass the bar. Auditors still need to sign off. The forms will change. The need for accountability won’t. Endgame Positions: Any profession or institution where a human must bear liability for outcomes.
Proprietary data. AI is only as good as its inputs. Unique datasets that took decades to accumulate or require privileged access to collect become more valuable, not less. When every model has the same architecture, the differentiator is what you feed it. Endgame Positions: Bloomberg, S&P Global, Epic, Palantir, credit bureaus, and anyone sitting on irreplaceable data.
Human attention. When the cost of generating content and outreach drops to zero, the bottleneck is the recipient’s ability to filter. Today you get five interesting cold emails a day. Soon it’ll be 500. Yes, your AI will help you filter. But the underlying resource, a human’s focus and time, is finite. Endgame Positions: The products and platforms that earn and hold attention as everything else gets noisier.
Network effects and liquidity. You cannot manufacture demand with intelligence. A marketplace wins on density, not software quality. Network effects are about human behavior and critical mass. Those take time and capital to build, not just intelligence. Endgame Positions: DoorDash, Uber, Airbnb, Waymo, and any marketplace with real liquidity.
Accumulated operational advantage. AI will make operationally complex businesses smarter and more efficient. But intelligence is replicable. The installed base is not. The driver network, the merchant relationships, the fleet, the decade of learnings baked into the system. A new entrant with perfect AI still has to build all of that. Endgame Positions: Amazon, DoorDash, Walmart, FedEx, Flexport, etc.
Security. The more capable AI becomes, the more capable AI-powered attacks become. Offense and defense both scale. There is no version of the endgame where security spending goes down. Endgame Positions: CrowdStrike, Palo Alto Networks, Flock Safety, Verkada, etc.
Physical space. You cannot create more land next to a power substation, more orbital slots, or more radio spectrum. Physical scarcity is permanent. As AI infrastructure scales, the value of the locations that support it only increases. Endgame Positions: Anyone who owns the physical locations where AI infrastructure gets built, data gets stored, or power gets delivered.
Intelligence itself. Even when intelligence is abundant, the companies producing it at the frontier will be few. Training frontier models requires billions in capital, the scarcest talent in the world, and compute infrastructure that only a handful of organizations can operate at scale. Endgame Positions: xAI, OpenAI, Anthropic, Google, Meta, etc.
These aren’t defensive positions. Every one of them benefits from better AI. They won't be replaced by it. That's the distinction.
If you’re building or investing right now, ask if an asset will get stronger or weaker as intelligence gets cheaper. If the answer is stronger, you’re playing for the endgame.