Back Research Notes Factories That Manufacture Intelligence’: The New Semiconductor Inflection Published on September 18, 2025 By Jordi Visser “Now, we’re entering the era of physical AI, AI that can perceive, reason, plan and act.” Jensen Huang Based on conversations the last two days I wanted to follow up the Oracle paper with one connecting the Nvidia/Intel deal today. As I highlighted, Oracle’s historic earnings and commentary were the first real proof that AI demand is outstripping supply. I have again been surprised that most of the questions and comments start with the negatives of RPOs, margins, the debt they may have to take to meet the supply and etc. The important message from their results is 1) AI demand is massive, growing and there is not enough capacity for it and 2) Inference is the driver, not training. Forget what you think about Larry Ellison or whether they ever get the capacity to meet those obligations. This wasn’t about future hype, it was about businesses already committing billions of dollars to use AI every day. The problem is that the industry can’t build capacity fast enough. Oracle showed the world that the limiting factor is no longer interest in AI, but the ability to run it at scale. And for that, everyone should be focused on compute and power. “These aren’t data centres. These are factories that manufacture intelligence.” Jensen Huang This set the stage for NVIDIA’s $5B investment in Intel. Oracle’s backlog made clear that the real challenge is not just creating AI models, but running them constantly in real time, what’s known as inference. Graphics chips alone can’t meet that need. What’s required are complete systems where different processors share the load: some handling logic and control, others accelerating the math, all tied together more efficiently. By partnering with Intel, NVIDIA gains access to mainstream processor technology and new ways to connect chips, allowing it to build systems better suited for inference at scale. “The next big thing is Physical AI, AI with a body. Robots, autonomous machines, industrial systems… It’s all coming.” Jensen Huang The partnership also marks a turning point in how AI will be used. The first phase of AI was all about training massive models in giant data centers. The next phase is about running those models billions of times a day and eventually embedding them into machines from cars and robots to personal computers; embodied AI. Embodied AI was the first deep dive paper I wrote for 22V and whether it was the inference one after or the Micron one or the Tesla one, the next AI investment theme would be the shift to the physical world. This is the PMI trade. This is the tide that lifts many more boats. That shift requires blending different kinds of chips into unified systems. By aligning with Intel, NVIDIA is helping ensure the computing power needed for this next stage is built, secured, and ready to scale. AI is now infrastructure, and this infrastructure, just like the internet, just like electricity, needs factories.” Jensen Huang For the broader semiconductor industry, this could be the important inflection point. Based on market reactions today, they think it is so far. If you look at the best performing semiconductor names, it is the old economy CPU related names that are still trading below all-time highs and look a lot like the ISM PMI chart back to 2021 which makes sense since they are for cars, airplanes, washing machines, industrial controls, medical devices and not for LLM training. Oracle revealed the scale of demand; NVIDIA/Intel is a sign of the capacity response. The implications ripple through the entire supply chain, from memory makers to the companies that supply the specialized tools to build chips. Investors see this as confirmation that AI is not just another technology cycle, but a structural shift that will drive growth across the industry. Together, Oracle’s backlog and NVIDIA’s partnership with Intel may mark the start of a new era where building capacity for inference running AI everywhere, all the time becomes the central challenge of technology. And looking ahead, 2026 won’t just be about GPUs or data centers, cooling systems or gas turbine shortages. The story will be about compute and power expanding across many more players as we enter the embodiment stage, where AI becomes the “brain” inside machines and where demand is racing faster than supply can keep up.