You’ve probably been losing sleep over the wrong things. We all have. We look at the AI space and obsess over who has the most parameters, who topped the LMSYS leaderboard, and who is closest to AGI. We assume technical superiority equals commercial victory.
But after analyzing 34 AI companies across the globe, the data paints a radically different picture. The best model doesn’t win. In fact, chasing model supremacy might be the fastest way to go broke.
Having the smartest AI doesn’t guarantee you make money; it just guarantees you have a massive burn rate.
Look at OpenAI. They defined this entire generative AI era. ChatGPT is synonymous with artificial intelligence. But building frontier models is like running a Michelin-star restaurant where the truffles cost more than the tasting menu. The compute required to train and run these models is staggering. Yes, they generate revenue through subscriptions and APIs, but does that revenue cover the astronomical inferencing costs?
Furthermore, people confuse Microsoft Copilot with OpenAI’s enterprise success. Microsoft owns the product and the customer relationship; OpenAI provides the brainpower. It’s a dangerous dynamic. You might be the genius in the room, but the guy who owns the building is collecting all the rent.
In the AI gold rush, the guys digging for gold are going broke, while the guys selling the shovels are buying yachts.
Enter Nvidia. While OpenAI burns billions to push the frontier, Nvidia doesn’t care whose model ranks #1 this week. They just know that everyone—OpenAI, Anthropic, Google, Meta—needs GPUs to train, run, and scale. Nvidia isn’t just selling hardware; they’ve built an inescapable ecosystem with CUDA. Developers build on it, enterprises standardize on it, and the switching costs become insurmountable.
Nvidia isn’t the sexiest player in the AI space. They are the unavoidable bottleneck. And bottlenecks are where the money is.
You don’t need to win the AI race if you own the road every racer has to drive on.
But what if you can’t build a monopoly on infrastructure? What if you’re a startup trying to carve out a living in the application layer? The lesson from Midjourney is your blueprint.
Midjourney didn’t try to build a general-purpose AI platform. They didn’t pretend they were going to achieve AGI. They picked one specific thing—AI image generation—and built an unbeatable user experience around it. They wrapped it in a simple subscription model and let the product speak for itself. They don’t have the strongest foundational model in the world, but they have a product people gladly pay for every month.
Users don’t pay for parameters; they pay for problems to disappear.
If you’re a founder, an investor, or a product team, you need to stop obsessing over benchmark scores and start asking three brutal questions:
First, how do you actually make money? Are you selling compute, API calls, SaaS subscriptions, or enterprise implementations? The path to revenue dictates your entire strategy.
Second, what is your actual moat? If your only defense is ‘our model is slightly better,’ you have no defense. Models get commoditized. Real moats are built on proprietary data, deep industry know-how, embedded workflows, and delivery capabilities.
Third, what painful problem are you solving? AI is just a tool. If a customer can’t articulate why they need your product to survive, you don’t have a business.
Technical superiority is an illusion if your customer has no reason to write you a check every month.
The AI race is not a single track. It’s a sprawling, messy value chain. The frontier labs are fighting for the future, burning capital at an unprecedented rate. The infrastructure players are taxing the entire industry. And the vertical players are quietly building profitable empires by solving specific, painful problems.
Stop trying to build the smartest model. Start building the most defensible business.
The model is the entry ticket, but the business model is the exit strategy.
FAQ
Q: Doesn't the company with the best model eventually just crush everyone else?
A: No. Frontier models require astronomical compute costs that often outpace revenue. Without a defensible ecosystem or vertical lock-in, superior tech just means a superior burn rate.
Q: What's the practical takeaway for a startup founder right now?
A: Stop trying to build a general-purpose AI platform. Pick a specific, painful industry problem, embed your tech into their daily workflow, and make it impossible to rip out.
Q: Is investing in foundation model companies like OpenAI a bad idea?
A: It's a high-risk, high-reward bet, but it's not where the guaranteed margins are. The safer, more lucrative bets are on the infrastructure layer (the bottleneck) or highly specialized application layers.