This is a YouTube video titled "Why Most AI Companies Are Burning Cash" containing a monologue or presentation discussing the computational demands and economic challenges faced by large language model companies, focusing on GPU utilization and software efficiency improvements. The speaker explains how boosting compute utilization can change the market dynamics and growth potential in the AI industry. The content is from July 2025.
We look at largely large language models companies today, and they're doing about an exaflop per pass. That's about a trillion trillion operations. That's a lot. I'm talking about a couple hundred GPUs for one model running at one time. That's why most companies like this are losing money. And the reason they're doing that is because they're operating at under 10% utilization of the available compute. Our software boosts them to 80% of the available compute. The number of chips automatically decreases. That changes the economics by a lot. Whether or not Nvidia is going to go away in a couple of years? Unlikely. Is the market going to grow 10x and allow for many more companies to emerge and own a lot of market share? Absolutely. But it requires a software breakthrough to get us there.
The AI Agent Getting More Supplier Data in 5 Weeks Than a Team Did in 2 Years
