The state of the AI economy from bottom up
The article analyzes the current state of the AI economy from a bottom-up perspective, examining real-world adoption, investment trends, and the operational realities faced by companies building and deploying AI systems.
Background
- Exponential View is a widely-read newsletter by Azeem Azhar that analyzes the impact of exponential technologies (especially AI) on business, economics, and society.
- The article argues that today's AI economy is shaped by the huge compute demands of foundation models, creating a divide between a small number of big players (like OpenAI, Google, Microsoft, Meta) who can afford to train trillion-parameter models, and the rest of the industry.
- Key concepts: "inference" (running a trained model to generate output, which is now the dominant cost), "scaling laws" (the empirical finding that bigger models and more data produce better results, driving an arms race), and "commoditization" (open-weight models like Llama and Mistral are becoming cheap or free, squeezing profits for companies that just provide model access).
- This matters because the winners and losers of the AI boom are not who many expected — the big money may end up in infrastructure (Nvidia's chips, data-center landlords) and applications built on top of cheap models, not in foundation-model companies themselves.