This episode breaks down 'tokenomics'—the economics of AI tokens—into four pillars: token utility, demand, supply, and monetization. It emphasizes that token value depends on intelligence and interactivity, and that businesses should focus on cost per token rather than input metrics like GPU cost. The discussion highlights how NVIDIA's extreme co-design and software optimizations enable dramatic cost reductions, exemplified by Blackwell delivering 50x more tokens per watt than Hopper.
Summarized by Podsumo
Token value is defined by two factors: the intelligence embedded in the token (from model complexity and context) and how fast it arrives (interactivity), creating a spectrum for different use cases.
Cost per token is the key metric for evaluating AI infrastructure ROI, as it combines input costs with actual output, contrasting with misleading metrics like cost per GPU hour.
NVIDIA's extreme co-design across compute, memory, networking, and software enables Blackwell to deliver 50x more tokens per watt and 35x lower token cost compared to Hopper.
Agentic AI workloads are a major token multiplier, requiring multiple LLM calls per request, which further underscores the need for low latency and optimized cost per token.
Businesses can monetize tokens through four models: selling tokens directly (e.g., Fireworks), building AI-native products (e.g., Cursor), enhancing existing products (e.g., Adobe), or improving internal operations.
"Not all tokens are created equal and there is a way to look at token value. There are two key factors that impact open value. One is the intelligence embedded in the token or how much intelligence does the token carry and the other is how fast does it arrive. — Shruti Kopakar, NVIDIA"
"Jevons paradox is essentially, you would think that okay, the GPUs are way more productive. They're generating so many more tokens. Do you need less of them? And the answer is absolutely no. — Shruti Kopakar, NVIDIA"
"The best place to start is to first just think through what is the final outcome. Usually that starts with your customers. — Shruti Kopakar, NVIDIA"