The Zero-Sum Tax of Frontier Models: Why Your Company’s AI Budget Is Wrong

You’ve probably felt it: the crushing anxiety of AI budget FOMO. You’re terrified that if you don’t buy the most expensive frontier models, your competitors will leave you in the dust. Meanwhile, your CFO is hyperventilating over the monthly token bill, secretly suspecting you’re overpaying for a parlor trick.

Look at the massive split happening right now in Big Tech. Nvidia is handing its engineers a $250,000 annual token budget per person. They’re allocating $2 billion like it’s growth capital. Uber, on the other hand, burned through its entire 2026 AI coding budget by April and slapped a hard cap on its engineers: $1,500 a month.

Why are two elite tech companies having completely opposite reactions to the exact same technology? Because they are running two completely different economics. And one of them is making a fatal mistake.

AI doesn’t create value out of thin air; it simply amplifies the bottleneck that already exists.

In an expansion market, intelligence directly converts to growth. A hedge fund using AI to trade alpha—where better intelligence literally equals more money—will buy the best model available, at almost any price. It’s like Formula 1 racing: if a faster lap time wins the race, you don’t argue about the cost of the engine.

But in an efficiency market, output is choked by non-intelligence constraints. Uber’s ride-sharing platform is bottlenecked by passenger demand, not code complexity. Smarter code doesn’t magically summon more riders. Once AI crosses the threshold of “good enough” to replace human labor, any additional intelligence is worthless. You’re just replacing labor.

If you’re using frontier models for efficiency tasks, you’re not investing in the future—you’re putting a Ferrari engine in a commuter car.

This is the massive trap most Fortune 500 companies are falling into right now. They are using GPT-4-level models to extract fields, summarize documents, and handle basic customer support. They are paying a massive premium for intelligence they don’t actually need. Once an open-source model can do the job, paying frontier prices isn’t strategy—it’s corporate vanity.

But the real dirty secret of the AI industry isn’t just that efficiency buyers are overpaying. It’s that a massive chunk of frontier model revenue is actually a zero-sum tax.

Think about it. If one hedge fund buys a frontier AI model to gain an edge, every other fund must buy it too, or lose. Once everyone has the same model, the advantage zeroes out. The total pie hasn’t grown, but the entire industry’s cost baseline is permanently elevated.

Frontier labs are taxing competitors to stay in the exact same place, enriching themselves without expanding total value.

This happened before. In 2010, a firm spent $300 million laying an 827-mile fiber optic cable through the mountains just to shave 4 milliseconds off a trading route. Everyone had to buy access, or lose. Then the exchanges changed the rules, and the cable became worthless. AI is the new fiber optic cable. The labs are the exchange, and you’re the trader paying the toll.

The metric that actually matters isn’t a lab’s total revenue. It’s the mix of zero-sum versus positive-sum customers. Zero-sum revenue is a tax—intense but incredibly fragile. Positive-sum revenue (like drug discovery, where better AI actually cures more diseases and grows the market) is an investment.

So how do you stop wasting money and start actually investing? You ask yourself one brutal question.

For the specific output your dollar is buying, is intelligence actually the bottleneck?

If yes—like in drug discovery or alpha trading—frontier AI is cheap at almost any price. If no—which is true for 99% of enterprise efficiency tasks—the cheapest model that clears the competency bar is your only rational choice.

The AI narrative is splitting. Stop letting FOMO drive your budget. If you’re an enterprise constrained by demand, distribution, or regulations, stop paying the zero-sum tax on frontier models. Embrace your identity as an efficiency buyer, deploy open-source, and stop pretending you’re buying a race car.

The real frontier isn’t about who spends the most on tokens. It’s about who is smart enough to know when not to buy.