According to the in-depth analysis of the AI trend report released by the "Internet Queen," the computational economics of artificial intelligence (AI) models are undergoing a critical transformation. The report points out that training the most powerful large language models (LLMs) has become one of the most expensive and capital-intensive investments in human history, with the cost of training each model often exceeding $100 million. Dario Amodei, CEO of Anthropic, noted in mid-2024 that some models under development are nearing a cost of $1 billion, and he predicts that models with training costs as high as $10 billion may emerge by 2025.
However, the Internet Queen's AI trend report also emphasizes that the cost of inference for running models (i.e., generating predictions, answers, or content) is rapidly declining. The report cites NVIDIA data showing that the Blackwell GPUs released in 2024 require 105,000 times less energy per token than the Kepler GPUs from 2014. Stanford HAI data further reveals that the customer price for AI inference (per million tokens) dropped by 99.7% over two years. This improvement in cost efficiency far exceeds previous key technologies like electricity and computer memory.
This sharp decline in inference costs and increased accessibility of models have reduced the cost of AI experimentation and accelerated iteration speeds. The Internet Queen's AI trend report suggests that this makes productization feasible for nearly anyone with an idea, leading to a surge in developer activity. For example, Meta’s Llama model saw its downloads increase 3.4 times in just eight months. Additionally, the report notes that despite differences in model costs, AI model performance is quickly converging, narrowing the gap between top-tier frontier models and smaller, more efficient models.
The report concludes that this evolution in AI economics is triggering a "creative explosion." Developers can choose the most suitable model based on technical or financial needs rather than being limited to a single supplier, reshaping the business models of AI model providers and forcing them to rethink how they can achieve profitability.