X

Training your favorite AI models is getting more expensive

Featured image for Training your favorite AI models is getting more expensive

As you may know, AI is an expensive business. It costs tons of money to secure GPUs, develop tools, and maintain them. One of the biggest costs is training the AI models powering our AI world. According to Anthropic AI’s CEO, training AI models is getting more expensive.

Any person looking to develop their own AI startup will need to come to terms with the fact that they’ll need millions, if not billions, just to get off the ground. This is because developing an AI model is no cheap affair.

During a recent episode of the In Good Company podcast, Anthropic AI CEO Dario Amodei spoke about some of the costs associated with AI. During the podcast, he mentioned that it took $100 million just to train the GPT-4o model. That’s just one model. This should come as no surprise, as OpenAI burned through $520 million last year.

Training AI models is getting more expensive

Unless you haven’t gone to the grocery store or tried to buy anything in the last couple of years, you’d know that prices are going up! Well, training AI models is not very different. Right now, companies like Google, OpenAI, Anthropic, Perplexity, Meta, xAI, etc. are pouring money into training their latest and greatest models. While we don’t know how much money each company is shelling out, we can guess that they’re paying in the range of hundreds of millions of dollars.

Amodei said during the podcast that models that cost $1 billion to train are coming. According to him, this shouldn’t take more than a few years to happen. He even mentioned models that cost $10 billion or even $100 billion to train. “Right now, 100 million. There are models in training today that are more like a billion… I think if we go to ten or a hundred billion, and I think that will happen in 2025, 2026, maybe 2027,” he said, “we’ll be able to get models that are better than most humans at most things.”

What does this mean for AI in general? It could mean general AI. Right now, companies are racing to achieve AGI (Artificial General Intelligence). This is an AI that’s smarter than a human being in all respects. Most industry experts believe that we’ll be able to reach AGI within the next decade. When that happens, we can only imagine how much money it will cost to train and maintain it.