The Cost of Training AI Models Has Skyrocketed—Up 3,550% Since 2020

Since the launch of ChatGPT two years ago, there’s been a sharp increase in the mainstream adoption of AI, which is now reshaping various industries. But behind its success is a challenge that’s hard to ignore: the skyrocketing cost of training AI models. According to research firm Epoch AI, the cost of training AI models has exploded in the past few years. Since 2020, the price tag for developing AI systems, particularly large language models, has jumped by more than 3,500%.

As these costs continue to climb, concerns are growing about whether this rapid pace of innovation is sustainable. While AI’s influence is undeniable, the real question is whether the industry can keep up—or if only the wealthiest players will be able to stay in the game.

Companies like OpenAI and Google have taken AI to new heights, but the price tag has gotten out of control. According to Statista’s Katharina Buchholz, both Google and OpenAI AIU systems have come with staggering price tags, potentially costing hundreds of millions of dollars to train, according to data from Epoch AI. This leap in complexity and cost underscores the massive resources required to stay on the cutting edge of AI development.

The Extreme Cost of Training AI Models

OpenAI’s GPT-4 alone cost up to $78 million to develop, while Google’s Gemini could have cost as much as $191 million. Those are just the computing costs, too—once you add in salaries, the total is even higher. Back in 2020, it cost between $2 million and $4 million to train ChatGPT-3. That number seems tiny now.

“The cost of training Gemini, which is a large language model that can be inputted with text, voice commands and images, reportedly stood between $30 and $191 million even before taking staff salaries into consideration,” Buchholz wrote.

She added, “ChatGPT-4, the latest edition, had a technical creation cost of $41 million to $78 million, according to the source. Sam Altman, CEO of OpenAI, has in the past said that the model has cost more than $100 million, confirming the calculations.”

Based on the Epoch AI data:

• In 2020, training ChatGPT-3 cost around $2 million to $4 million.
• In 2023, ChatGPT-4’s training cost increased from $41 million to $78 million, which represents an increase of 1,925% to 3,800%.

The cost of training AI models has grown by 1,925% to 3,800% since 2020.

Is AI Sustainable?

With costs soaring, smaller organizations and research institutions are finding it harder to keep up. This raises a huge question: can AI innovation keep moving forward at this rate, or will the costs eventually force some to step back?

Epoch AI data highlights this rapid jump in expenses. It’s partly driven by how much more advanced models have become in just a few years. ChatGPT-4, which started with text input, now also supports voice and images. Google’s Gemini was designed from day one to be a multimodal model, which likely pushed its costs higher.

As AI evolves, the financial strain might end up limiting who can stay in the race. For many, this creates a tension between the desire to innovate and the reality of how expensive it’s becoming to stay competitive.