AI Models: How They Run and Who Powers Them
AI models are the programs behind tools like ChatGPT, and if you have watched the presenter’s “Elon Musk’s One Stock Retirement Plan” presentation, you have heard him argue that the way these models run is about to change in a way that creates a new investment phase. This article explains how AI models actually work, what changes are underway, and who Brown says profits. No prior technical background needed; we will keep it plain.
What an AI model actually is
An AI model is a statistical engine with billions of tuned parameters, trained on enormous amounts of text, code, and images. Training is the expensive construction phase: thousands of processors run for weeks or months, adjusting those parameters until the model reliably predicts language and patterns. Once trained, the model is deployed, and every answer it produces from then on is called inference. When ChatGPT writes you an email, that is inference.
Brown splits these two phases into a memorable distinction: “Most of the investments in AI so far have focused on what I call ‘generic AI.’ And Nvidia was the big beneficiary of that first phase. But the future will be powered by what I call ‘Accelerated AI.’” Generic AI is the training-and-building era. Accelerated AI, in his usage, is the deployment era, where AI is “fully autonomous and making decisions on its own through a process known as inference.” His summary of inference is as good as any textbook’s: “It simply means we’re running intelligence at speed and scale.”
The problem: latency
Here is where hardware enters the story. During inference, a model’s work is split across many chips, and those chips talk to each other constantly. Every hop between chips adds a delay. Brown’s classroom analogy is the clearest explanation we have seen in any newsletter presentation: “It’s like trying to pass a note down a long row of people sitting in a classroom. Every single person who has to take the note and hand it to the next one adds a little delay… That’s called latency in a computing system.”
This is a real, decades-old engineering constraint, and Brown stakes his credibility on it: “As a former executive from two major chip makers… I can tell you this is a problem that has been plaguing the computer industry for 75 years. Nobody had been able to figure this out… Until now.”
The company he teases, which our clue analysis suggests is Cerebras Systems (CBRS), attacks latency by making one enormous wafer-scale chip instead of a cluster of small ones. Fewer handoffs means faster answers. Brown claims its technology “can support up to 120 trillion parameters” and “can make simultaneous computations at or above the level of the human brain.” The parameter figure is an evocative benchmark; the latency advantage is the genuine engineering core.
Who uses the technology
The presentation lists blue-chip partners to prove adoption, and each one illustrates a use case for fast inference. OpenAI uses it “to make ChatGPT respond much faster.” Microsoft partnered “to help software engineers write code faster.” IBM uses it “to increase the power of its flagship AI model called WatsonX.” CrowdStrike applies it “to enable real-time AI security against cyber threats.” Even GlaxoSmithKline made a deal “to speed up its drug research and discovery.” The common thread: real-time workloads, where every millisecond of latency costs money or degrades the product.
The biggest partner is Amazon. Brown says the company has “a deal with AWS, which is Amazon’s cloud business,” serving “more than 2 million businesses,” and that the full deployment details could surface on an upcoming analyst call. He ties the timing to a broader demand wave from Elon Musk’s projects: Terafab, a $119 billion Texas chip plant Brown says is documented in a Texas government filing, and Starmind, Musk’s plan, filed with the FCC, to launch “up to one million AI satellites.”
Why the dial-up analogy drives the investment case
Brown’s master comparison is the internet’s transition: “Dial up was ‘slow internet’… Broadband was ‘fast internet’… And the companies that made that transition possible made their shareholders incredibly wealthy.” His argument is that inference speed is the broadband moment of AI. The teased company, he says, is “the same size Nvidia was in 2016,” carries “a backlog of more than $24 billion,” is “sold out into 2027,” and has grown sales “about 500% a year since 2022.”
Considerations
Some healthy framing. The “1,000 times FASTER than regular AI” claim is Brown’s characterization, not an independent benchmark. Nvidia, AMD, and the hyperscalers themselves all build inference silicon, so the competitive race is live. And the shift from training to inference, while real, does not guarantee that any one company captures it. Brown’s own disclosure that The Near Future Report averages a 39.4% gain since inception is a useful anchor against the 37,800% Nvidia comparisons.
Where to learn more
Brown reveals the ticker and his full analysis in “The One-Stock Retirement Plan,” the lead report in a $179 trial of The Near Future Report with a 30-day keep-everything refund guarantee. For background, read our review of his prior Musk pitch, the 10X Project, our Near Future Report service profile, and our the presenter presenter profile. Brownstone has also covered the critical minerals behind this buildout, a good companion read.
NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.