Last week Google released Gemini 3. This morning Anthropic dropped Opus 4.5. By the time you finish reading this paragraph, OpenAI or Meta or xAI will probably have announced something else that “changes everything.”

The tech twitter sphere is, as always, exhausted. We treat these releases like rounds in a boxing match. “Gemini KO’d Opus!” “Claude is back on the ropes!” We are obsessed with finding the winner. We always think we need to find one model to rule them all.

It is arguably the stupidest conversation in technology. Except maybe vim vs. emacs. Because it’s obviously vim.

If you look at the history of industrial innovation, “winning” doesn’t look like a knockout. It looks like a messy, decades-long stalemate where everyone gets a little better, and the only people who actually lose are the ones who make unforced errors.

The Ford vs. Chevy Fallacy

In 1963, one out of every ten cars sold in the United States was a Chevrolet. It was dominance on a scale that is hard to comprehend today. General Motors and Ford were locked in a death struggle for the American driveway.

Did one of them win?

No. A new Ford didn’t mean Chevy went out of business. It meant the next Chevy had to be a little better. It meant they had to differentiate. Ford doubled down on trucks (the F-Series has been the best-selling vehicle for over 40 years). Chevy focused on performance and value.

Then the market shifted. Toyota didn’t “beat” Ford by making a better Mustang. They entered the market with the Corolla and the Hilux—vehicles that promised to outlast the heat death of the universe. They didn’t try to win the existing game; they found a niche that eventually became a massive segment.

The AI race isn’t a Highlander “there can be only one” scenario. It’s the auto industry in 1950. There is room for the truck (massive context windows), the sports car (reasoning speed), and the reliable commuter (cost efficiency).

Don’t Look for Winners, Look for Losers

While there won’t be a single winner, there will absolutely be losers. History is littered with them, and they almost always fail for the same reasons. If you want to know which AI labs are doomed, don’t look at their benchmarks. Look at their strategy.

1. The “DeSoto” Problem: Redundancy

DeSoto was a mid-priced car brand from Chrysler that died in 1961. Why? Because it didn’t have a reason to exist. It was squeezed between Dodge (cheaper) and Chrysler (fancier). It had an identity crisis that no marketing campaign could fix.

The AI Risk: If a model provider is just “OpenAI but slightly worse and slightly more expensive,” they are the DeSoto of this generation. Being a wrapper without a soul is a death sentence.

2. The “Gateway” Problem: Missing the Shift

In the 90s, Gateway 2000 was iconic. You couldn’t miss their cow-spotted boxes. My brief college career was during their heyday. I remember move in day looking like a cattle stampede of cow boxes. But they died because they were too slow to pivot to laptops. They were the kings of the desktop right as the world decided they wanted portability.

The AI Risk: We are currently moving from “Chat” to “Agents.” The Gateway of AI will be the company that has the best chatbot on the market exactly one year after everyone has stopped chatting and started assigning tasks.

3. The “AMC” Problem: Weird for Weird’s Sake

American Motors Corporation (AMC) gave us the Gremlin and the Pacer. They were innovative, sure, but they were often weird because they lacked the capital to compete head-on with the Big Three. They used old tech (straight-six engines) in new bodies because they couldn’t afford to develop new ones.

The AI Risk: Watch out for labs that claim their lack of compute is a “strategic choice” for “efficiency.” Sometimes it is. Usually, it just means they are bringing a Gremlin to a Formula 1 race.

The Rising Tide

We aren’t watching a race to the finish line. We are watching the formation of a new ecosystem.

GPUs and TPUs wouldn’t exist today without the oddly-shaped PCs of the 2000s gaming market. The failures of Gateway and the struggles of AMD in the early 2010s paved the way for the hardware we have now. The hardware that is making AI possible.

So stop waiting for Gemini to kill GPT, or for Claude to kill Gemini. It’s not going to happen. Instead, look for the ones who are losing their identity, missing the platform shifts, or running out of gas.

That’s where the history lesson is.