The AI ​​frontier is crowded. There will be big winners and big losers.

It’s a stunning collapse. In about 18 months, the cost of 1 million tokens went from $180 to just 75 cents.

Winners

Think of it as a benchmark for how much it costs to use AI models. The winners will be those who are heavy users. This revolutionary new resource is about 240 times cheaper now. This saves a lot of money for startups, researchers and others.

Martin Casado, a respected artificial intelligence expert at Andreessen Horowitz, compared this to the rapid decline in technology costs during previous waves of innovation.

“A nice frame of reference overlays the cost per bit over time for the Internet. Or the computational cost during microchip growth,” he tweeted. “Very clearly a supercycle where the marginal costs of a fundamental resource will go to zero.”

losers

There is another flat that is more worrying.

Part of the reason token prices have fallen is that the supply of capable AI models has grown substantially. And the rules of supply and demand dictate that when this happens, prices often fall.

Having a top AI model isn’t nearly as special as it was 18 months ago. During that time, OpenAI’s GPT-4 emerged as a clear leader. Since then, Google has released a new Gemini model that is practically as good. Anthropic has great designs. X has Grok. And there are more than 100 more available on a careful monitor industry ranking.

All these models are mostly trained on data that is publicly available on the Internet. So they are not that different from each other. Standing out in this crowd is increasingly difficult, so it goes without saying that the service they provide – token processing – is worth much less.

“Frontita is now a commodity”, Guillermo Rauch, CEO of AI startup Vercel told me. “AAt one point, GPT-4 was in a class of its own, right? And they were charging $180 per million tokens. Now everyone else has a frontier model.”

Zuck’s influence

Mark Zuckerberg has already taken this to the extreme by making Meta’s Llama models free as open-source services. Startup Mistral did the same.

“I think the one who disrupted all of this is Zuck,” Rauch said, noting that Meta has succeeded with previous open-source efforts. One example is Facebook’s Open Compute Project, which standardized and reduced the price of data center infrastructure.

Zuckerberg has been dropping not-so-subtle hints about the crowded market for AI models lately.

“I expect AI development to continue to be very competitive, meaning that open-sourcing any given model does not provide a massive advantage over the next best models,” he wrote in a recent blog.

Translation: No AI model is going to be that much better than others, so why not throw these things away.

Every time the price of something crashes, there are losers. This time it could be any companies that are still trying to build best Model AI and earn money directly from it. An example is the implosion of the startup AI inflection this year, who spent a lot of time and money developing their own frontier model that failed to stand out from the crowd.

The answer and the problem

The answer is to make a new AI model again that is much better than everyone else, right? If you can achieve this, then you can again charge more for your token processing services.

The problem is that the cost of developing the next almighty model is skyrocketing and likely rising. Nvidia GPUs are still very expensive. Building or leasing the giant AI data centers needed only gets more expensive. The price of electricity to power and cool these facilities is increasing.

The return on investment from pouring massive capital into a new AI model is good – if you get a much better brain. If it’s only slightly better than what already exists, it’s not that good, Rauch explained.

The Big GPT-5 Question

Beyond costs, AI experts are beginning to debate whether it’s even harder to make new models that are significantly better.

OpenAI released its GPT-1 AI model in June 2018. After just 8 months, GPT-2 came out. Then it took 16 months to launch GPT-3. And it took about 33 months for GPT-4 to appear.

“These huge leaps in brain power, in reasoning power. It’s getting harder and harder to achieve,” Rauch told me. — That is a fact.

There are high expectations for GPT-5 from OpenAI. Many in the industry expected this to come out in the summer. Summer is over and hasn’t come yet.

Maybe GPT-5 will crush the competition when it finally comes out. And OpenAI will be able to charge much higher prices per token. We will see.