Inference demand seems to keep rising with each new generation of AI models. It feels like the ultimate Jevons Paradox in action: better AI capabilities lead to more usage, more token demand, and more workloads shifting from humans to machines. As models improve, companies don't just use less AI; they find more ways to deploy it. Now, it looks like OpenAI is facing this same trend of accelerating inference demand.
From where I stand, the winners might not only be the model creators but also the infrastructure providers powering the AI economy. I'm watching a few names in this space: $Microsoft(MSFT)$ for its AI ecosystem and cloud infrastructure, $Oracle(ORCL)$ for AI cloud capacity, $NVIDIA(NVDA)$ for accelerated computing, and $Broadcom(AVGO)$ for AI networking.
To me, the AI cycle appears to be moving from training scale to inference scale. Long-term demand for compute remains one of the biggest trends to watch.
Comments