Anduril rewrote defense contracting. Perplexity redefined search. Athletic Brewing invented a category that did not exist. NVIDIA hosts a conference where thirty-nine thousand people compete for position inside a game Jensen Huang designed. The most durable advantage in business is not winning the game — it is being the one who wrote the rules.
Anduril does not bid on defense contracts the way Lockheed Martin does. Lockheed plays cost-plus — respond to a government request for proposals, estimate costs, add a margin, and bill overruns when the program inevitably runs late. The game has existed since the Cold War. The rules reward program management, lobbying relationships, and the ability to stretch timelines without losing the contract. Every major defense prime plays this game. Some play it better than others. None of them questioned whether it was the right game.
Anduril did not try to play the cost-plus game better. It built weapons on its own money, at its own risk, and showed up with finished products — software-defined systems that receive capability upgrades over the air the way a phone receives app updates. The Pentagon did not change the rules for Anduril. Anduril changed what counted as a valid move. The result: revenue doubled to roughly two billion dollars in 2025, a twenty-billion-dollar Army contract followed, and the company is raising at a sixty-billion-dollar valuation in a market where the traditional primes trade at single-digit revenue multiples.
Lockheed Martin still makes more money. It still wins more contracts. By every metric that the cost-plus game measures, Lockheed is winning. But the game itself is shifting underneath it — and the company that defined the new game is capturing the economics of the transition.
Three Levels
There is a hierarchy to strategic engagement that most competitive analysis ignores because it looks at actors within a game rather than at the game itself.
The first level is playing. You accept the rules as given, study the other participants, and try to execute better than they do. In financial markets, this is trading — reading the same data as every other participant and trying to extract profit from the same price signals. In business, this is competing on execution within an established category: a better search engine, a cheaper cloud provider, a faster chip. Playing can be done brilliantly. It can be done profitably, for a time. But in any game with enough participants and transparent rules, the returns converge toward zero. The more efficiently the game is played, the less there is to win. This is not a metaphor. It is the definition of an efficient market.
The second level is modifying. You find edge cases — loopholes, asymmetries, temporary mispricings that the rules create but do not address. Hedge funds that exploit regulatory arbitrage are modifiers. Companies that find underserved niches within established markets are modifiers. The advantage is real but temporary. Once the modification is visible, the rules adjust or other participants copy it. Modifiers earn alpha, but the alpha decays.
The third level is creating. You define the rules, the scoring system, and the strategy space itself. You do not compete within the existing game. You build a new one and invite others to play. The advantage is architectural — it compounds rather than decays, because every new participant strengthens the game-maker's position rather than competing it away.
The distinction is not about size or ambition. It is about what kind of returns the activity generates. Playing produces returns that diminish as competition increases. Modifying produces returns that decay as the modification becomes known. Creating produces returns that compound as adoption grows. The same hour of attention, deployed at different levels, generates structurally different outcomes.
The Evidence
The pattern is visible across industries, and the data is specific enough to test.
Perplexity AI did not build a better search engine. It created what it calls an answer engine — a system whose output is a synthesized, cited response rather than a ranked list of links. This is not a feature improvement to search. It is a different game with a different scoring system. Google's game rewards the best-ranked links. Perplexity's game rewards the best-synthesized answer. The company crossed three hundred and thirty million dollars in annualized revenue last year, tripled its valuation to eleven billion dollars, and now processes thirty million queries per day. Google's response — AI Overviews, which embed synthesized answers at the top of search results — is Google playing Perplexity's game, not the other way around.
Block eliminated nearly forty percent of its workforce in February 2026 and the stock surged twenty-four percent. The Friction Tax documented the immediate market reaction. But the deeper move was not the layoff itself. Block announced that its open-source AI coding agent, Goose, could perform the work of the departed employees. The game being created is not a company that uses AI to cut costs — every company claims that. The game is a company that open-sources the AI tool that replaces its own workers, turning its internal productivity gains into an external platform that other companies build on. If Goose becomes the standard AI coding agent, Block captures the ecosystem economics, not just the headcount savings.
ElevenLabs did not compete with Siri, Alexa, or Google's text-to-speech. Those are voice features embedded in walled gardens — Amazon's game, Apple's game, Google's game. ElevenLabs created a standalone voice platform: cloning, dubbing, narration, and now music generation, sold as an API that any developer can build into anything. Revenue hit three hundred and thirty million dollars in annualized terms, forty-one percent of Fortune 500 companies use the platform, and the valuation reached eleven billion dollars in February 2026. The walled-garden voice assistants are playing a device game. ElevenLabs is playing an infrastructure game — and the infrastructure game has no ceiling on the number of players it can serve.
Athletic Brewing may be the most instructive example precisely because it is the least technological. Seven years ago, non-alcoholic beer held three-tenths of one percent of total beer sales in the United States. The product category was a punchline. Athletic did not try to make a better NA beer within the existing beer market. It created a different game: premium craft beverages for health-conscious adults who want the social ritual without the alcohol. Dedicated non-alcoholic-only breweries, fitness and wellness positioning, a brand built around an identity rather than a compromise. The category has grown from one hundred million to over one billion dollars. Athletic holds eighteen percent of the overall non-alcoholic beer market and more than half the market among dedicated NA brewers. Heineken 0.0 and Budweiser Zero — products launched by established brewers — are playing Athletic's game. Athletic wrote the rules.
The Asymmetry
The question is not why game-creation produces better returns. That is almost tautological — defining the rules gives you architectural advantage. The question is why the returns are structurally different in kind, not just in magnitude.
When you play a game, every additional player competes away your returns. An efficient poker table with nine skilled players has a negative expected value for everyone after the rake. A stock market with a million informed participants prices information into securities before any individual can extract it consistently. The more participants, the thinner the edge. This is the fundamental property of games with fixed rules: the value of participation declines with the number of participants.
When you create a game, every additional player increases your returns. NVIDIA's GTC conference — happening today, thirty-nine thousand attendees, five hundred and fifty sessions — is a demonstration of this inversion. Every company presenting is building on NVIDIA's CUDA platform, training on NVIDIA's GPUs, optimizing for NVIDIA's architecture. The more companies that attend GTC, the more entrenched NVIDIA's position becomes. Jensen Huang did not win the GPU market by making faster chips than AMD. He created the AI compute game by investing in CUDA when it had no customers, building developer tools when there were no developers, and establishing a software ecosystem that made the hardware decision automatic. The chip is the product. The game is the ecosystem. And the ecosystem has the property that every new player strengthens the game-maker.
This asymmetry — diminishing returns for players, compounding returns for creators — is not a metaphor. It is a structural consequence of who controls the rules. Players compete within a fixed strategy space. Creators expand the strategy space itself. When the space expands, the creator captures the expansion. When the space is fixed, the players divide a static prize.
The Recognition Problem
The hardest part of the framework is not understanding it. It is recognizing which level you are operating at.
Most companies that believe they are creating a new game are actually modifying an existing one. Most people who believe they are investing are actually gambling — playing someone else's game with the rules stacked toward the house. The distinction is not always obvious from the inside. You can feel the urgency of competition, the thrill of a good trade, the satisfaction of outperforming a benchmark, and still be operating entirely within someone else's rules.
There are diagnostic questions. Who defined the scoring system you are optimizing for? If someone else defined it, you are playing their game. Can new participants enter your activity and compete away your advantage? If yes, you are playing. Does your position strengthen or weaken as more participants join? If it weakens, you are playing. If it strengthens, you may be creating.
Prediction markets illustrate the distinction with uncomfortable clarity. Kalshi and Polymarket created the game — the platform, the contracts, the settlement infrastructure, the liquidity incentives. The platforms earn fees on every trade regardless of which side wins. The traders play the game — analyzing probabilities, sizing positions, competing against other informed participants for edges that the market prices away. The platform captures a fee on every transaction. The traders, collectively, break even minus fees. This is the structural arithmetic of every well-designed game: the creator extracts value from the activity itself, while the players compete for shares of a shrinking surplus.
The same logic applies to conferences, marketplaces, standards, protocols, and operating systems. Apple's App Store is a game. The developers are players. Amazon's marketplace is a game. The merchants are players. NVIDIA's CUDA is a game. The AI companies are players. In each case, the creator earns a toll on every transaction, every interaction, every participant — and the toll grows as the game grows.
The Choice
There is nothing wrong with playing. Most of the world's economic activity is playing — executing within established rules, providing services within existing markets, competing on quality and price within defined categories. Playing is how goods get produced, services get delivered, and the economy functions. The point is not that playing is bad. The point is that playing is a fundamentally different economic activity from creating, and confusing the two leads to misallocated attention.
An hour spent playing a well-designed game generates value that is divided among all the players, minus the creator's toll. An hour spent creating a game generates value that compounds with every future player who enters. The input is the same — one hour of human attention. The output is structurally different. The returns from playing are bounded by the competition. The returns from creating are bounded by the size of the game.
The strategic hierarchy — playing, modifying, creating — is not a value judgment. It is an observation about how returns accrue. Anduril's two billion dollars in revenue comes not from being a better defense contractor but from defining what counts as defense contracting. Perplexity's eleven-billion-dollar valuation comes not from being a better search engine but from redefining what search means. Athletic Brewing's market share comes not from making a better beer but from creating a category that turned non-alcoholic brewing from a compromise into an identity.
In each case, the game-maker did not compete harder. The game-maker competed differently — by refusing to accept the existing rules as the only rules, and building something that made the old game optional.
Thirty-nine thousand people will gather at the SAP Center in San Jose today to hear Jensen Huang describe what NVIDIA has built. Most of them will be there to learn how to play NVIDIA's game better — how to optimize for the latest architecture, how to use the newest tools, how to build on the platform. A few of them will be there to study how the game was made — and to think about what game they might create next.
The returns from the conference will be distributed accordingly.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)