AI Value Capture – The Shift to Model Labs (newsletter.semianalysis.com)

Recent developments in AI, particularly seen through the lens of "model labs," have led to a significant shift in how value is captured within the ecosystem. The performance of AI models like Anthropic's Claude has dramatically improved, propelling its annual revenue run rate from $9 billion to over $44 billion within a year. This growth is attributed to both the increased utility of tokens—which are now more valuable for end users—and major hardware improvements in chips like Blackwells, which can produce 30 times the tokens per second compared to previous generations. As a result, AI labs are now capturing a disproportionate share of value, shifting from minimal gains last year to substantial returns today. This transformation is not just about improved models; it's also about the economics of token production, which has drastically decreased, enhancing the margins for model creators and inference providers. Though companies like TSMC and Nvidia have yet to fully adapt to this rapid change in token value and demand, innovations in agentic AI have enabled businesses to complete complex tasks far more efficiently than before. As AI adoption continues to accelerate, the restructured landscape of the AI marketplace suggests that both the technology and economic dynamics are moving faster than traditional industry standards, setting the stage for a new era of AI-driven productivity.