With the rapid development of artificial intelligence technology, the training cost of AI models has reached an astonishing $1 billion. According to Anthropic CEO Dario Amodei, that cost could soar to $10 billion or even $100 billion in the next three years. The sharp rise in this number has sparked widespread concerns about whether the AI bubble is about to burst.
Dario Amodei further pointed out that with the continuous evolution of AI models, hardware demand will grow exponentially, becoming the main driver of training costs. He also emphasized the gradual process of artificial intelligence from generative artificial intelligence to general artificial intelligence, which is similar to the learning methods of human children.
In the upward trend of AI training costs, hardware and energy consumption are two key cost expenditures. GPU delivery volumes from suppliers such as Nvidia will increase geometrically, and the power demand in data centers will also increase dramatically. These factors together push up the training cost of AI models.
In addition, human resources, data collection, operation and maintenance will also cost a lot of money. If these problems are not solved, the training cost of AI models is likely to reach US$100 billion in 2027. This prediction has led the industry to think deeply about the sustainable development of AI technology.
Key points:
⭐ The current AI model training cost is as high as US$1 billion, and it is expected to rise to US$10 billion or even US$100 billion in the next three years.
⭐ Hardware and energy consumption have become the main drivers of training costs, and spending on both aspects will increase dramatically as the AI model develops.
⭐ Human resources, data collection, operation and maintenance will also cost a lot of costs. If these problems are not solved, the cost of AI training in 2027 may reach US$100 billion.