Back Research Notes Oracle, Inference, and the Persistence of Disbelief Published on September 17, 2025 By Jordi Visser Aside from Moonshots with Peter Diamandis, there are very few podcasts I listen to regularly. I’m much more of a thematic or guest-specific listener. In a world of too much information, podcasts are no different, so I need to be selective. Back in January, one theme dominated my listening: the DeepSeek moment. When something impacts investor decisions and mindsets to the degree DeepSeek did, I want as much information as I can get on it. This particular theme ranged from Silicon Valley experts, traders, tech analysts, and pure quants. I wanted all their voices to understand if I was missing something. The problem wasn’t finding opinions; it was filtering through the endless supply, as every single podcast was talking about it for weeks. Beyond the podcasts, the calls I received for my own AI opinion on the situation were greater than for any other event this year. It reached the point where I would guess that everyone in the investment industry remembers where they were when DeepSeek happened, even though there were no official numbers or factual information released. It was purely an embedded fear of the dreaded AI bubble. Now fast forward and contrast that with last week, when Oracle dropped its latest earnings report. I’m not sure we’ve ever seen an earnings report like this. The numbers were staggering: Remaining Performance Obligations (RPO) surged to $455 billion, up 359% year-over-year. Yes, I know, few of us knew what RPO was before last week, but now we do. Sometimes we big numbers don’t shock us any more so to conver this into how large it is, if the New York Yankees had 50,000 seats and charged 500 dollars a seat and the 80 games were sold out every year since 1900 with those metrics, they would have $252 billion in sales. For the last quarter alone, the Oracle sales increase was over $300 billion. But the most important story for markets and GDP, in my view, was the breakdown of what is driving the growth. Buried in that figure was the important macro point: inference demand has exploded. Training large language models may have dominated the AI narrative since the launch of ChatGPT, but it is inference, the actual deployment of AI into real workflows and the real economy, that is officially blowing apart expectations. I highlighted that we had reached this point in my May 15th report, The Inference Inflection: Where Real-Time AI Meets Real-World Opportunity . In that report, I listed 22 companies that should benefit from the move to inference from training, as enterprises begin using the thinking part of artificial intelligence to help make decisions. As of today since May 15th, an equal-weight basket of those names is up 26% vs. 13% for the SPX. The company earnings reports for Q1 showed that the inference demand was growing, but Oracle confirmed it had grown significantly. According to the earnings commentary for the company, one Oracle customer even requested all of the company’s available inference capacity. That is not hype. That is not theory. That is real. As Larry Ellison put it on the call, it is “insatiable demand.” Inference demand should matter to every investor, strategist, analyst, and economist because it marks the seismic shift in AI from building models to using them, the point where AI moves from potential to real-world value creation. Remember this comes just weeks after an MIT report said AI had a “95% failure rate” that was intended to get clicks and scare people. Unlike training, which is episodic, inference is perpetual and scales with every interaction, making it the true driver of sustained adoption, productivity, and economic impact. The sheer size of Oracle’s order book puts this AI acceleration in perspective: the increase in backlog is nearly equivalent to the quarterly increase in U.S. nominal GDP. For a single company’s contracted pipeline to rival macroeconomic output growth at the national level is extraordinary. And yet, despite this, the Oracle calls for my AI opinion never came like they did repeatedly for DeepSeek. In addition, I have yet to come across one podcast that dedicates more than a news segment to its importance. The reality is the conversation about Oracle’s numbers died much faster than when a Chinese company few people had heard of released a model to the public. This is not surprising to someone who only speaks to investors about AI and where we usually spend most of the time with pushback. In this case, what makes this moment remarkable is not just Oracle’s numbers, but the reaction to it. I thought for sure this would get everyone not on board to find a seat on this train, but even in the face of insanely large evidence, skepticism reigns. If you don’t believe in AI, I understand why general economic, market, and even consumer sentiment won’t improve. It is the economy and the stock market. Last week, we got two examples of this negative sentiment. The AAII Investor Sentiment survey made the investor point clearly: last week’s reading showed Bears at 49.5 compared to just 28 for Bulls. Across all of 2023 and 2024, there was only one week when Bears were higher than 49.5, November 2023, and that was after an 11% drop in the S&P 500. Also during last week, we got the University of Michigan Consumer Sentiment, and it came in at 55.4. In the last 45 years back to 1980, or a total of 540 releases, only 5 times have there been readings lower than the 55.4 reading. In those cases, the SPX had just fallen over 20%. This time it sits at all-time highs. If I had to guess why so many people remain bearish, it is simply that most people do not believe in AI. Last week, the cover of The Economist had an AI title, and everyone acted like this was it, the curse of The Economist cover article, but the article according to ChatGPT was a 4 on a 1-10 Bearish to Bullish scale. Regardless of the numbers from Oracle or Nvidia or the Mag7’s 26% YoY earnings growth, there is still a feeling that AI is just hype that will eventually fail. If you take away AI, there is no economy, so I actually agree that if you leave everything else the same, the economy and market would be in trouble. But that is not the case and not what is happening. As I have my conversations around AI each week, I can say they almost never start with anything positive. The ones making money on it are still focused on what can go wrong. The ones not making money are argumentative about AI itself and what it is actually doing. I consider myself a curious person who loves to learn about new things and go deep into rabbit holes about the future. This is why I have always been a macro person. I want to understand in my mind what five years will look like and work backwards. I believe in five years we will be curing all diseases and exponentially expanding lifespan, have flying cars from NYC to JFK so we never have to be on the Van Wyck again, or that we will get on an elevator with a humanoid next to us, and so many other mind-bending things. The economy we know today will not exist. But for many, they don’t see it that way. This is my attempt to now take the Oracle numbers and be understanding as to why there is a blind spot. Macro investors have struggled to capture the real impact of AI on GDP, and I agree there is no real way to measure it the traditional way GDP was created. Simon Kuznets, the father of the GDP measure, did not know AI would be happening or that we would still be using it, so forgive him for not focusing on the intangibles like software, the cloud, and now intelligence embedded in machines. Joseph Schumpeter saw it with Creative Destruction. Part of the problem is that AI’s effects are multi-layered and don’t fit cleanly into traditional macroeconomic models. It especially breaks the relationship between labor and GDP. Here is my view on what has happened for its impact on the way GDP is measured. First