Nvidia GPUs: From Gaming Chips to AI Training Engines
Nvidia GPUs are the engines of the modern AI boom, and the story of how a video game chip became the most important piece of silicon in the world is genuinely useful for understanding where the next investment phase might go. It is also the story the presenter tells at the center of his “Elon Musk’s One Stock Retirement Plan” presentation, because his case for a new “Accelerated AI” winner only makes sense against the history of how Nvidia won the first phase.
What a GPU actually does
A GPU, or graphics processing unit, is a processor designed for parallel work: instead of doing one calculation very fast, it does thousands of calculations simultaneously. That is exactly what rendering a video game screen requires, which is why Nvidia built the architecture in the first place. But it is also what a neural network needs. Training an AI model is, at bottom, an enormous pile of matrix multiplications that can all run at once. When researchers realized in the early 2010s that gaming cards could train models far faster than conventional processors, the modern AI era began.
The 2016 moment Brown keeps coming back to
Brown, a former senior executive at Qualcomm and NXP Semiconductors, says he recommended Nvidia in 2016, the year the company launched its Pascal chip. “It was 45 times more powerful than traditional computer chips,” he says. “This is the chip that kicked off the AI revolution.” From that call to the peak, Nvidia delivered a 37,800% gain, and Brown’s model portfolio claims a 4,100% win on the stock, with the honest qualifier “if you had held on past their original recommendations and sold at their peak.”
The lesson he draws is not “buy Nvidia forever.” It is that the biggest gains came from owning the architecture that defined a new phase of computing, before the market priced it in. Nvidia in 2016 was the same size, he notes, as the small inference company he teases today.
Why training and inference are different jobs
Here is the technical heart of the pitch, and it is real computer science, not marketing. A GPU cluster is phenomenal at training: running a model’s construction phase for weeks on end, where raw throughput is everything. But once a model is deployed, the job changes. Now the model must answer requests in real time, thousands per second, where the cost is dominated by moving data between chips. That delay is latency, and Brown’s classroom analogy is the best plain-English version we have encountered: “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.” For a closer look at how the two architectures stack up, see our Cerebras vs Nvidia breakdown, and for where Wall Street sees the share price heading next, our NVDA stock price prediction review.
A GPU cluster is a long row of note-passers by design. So-called wafer-scale designs, like those from Cerebras (CBRS), the company our clue analysis matches to Brown’s teaser, collapse most of that distance onto a single enormous chip. Brown says the company’s technology “can produce intelligence up to 1,000 times FASTER than regular AI,” which is his characterization of the advantage rather than an independent benchmark, but the underlying latency problem is genuine and decades old. As he puts it: “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 dial-up-to-broadband analogy
Brown’s master comparison is the internet transition. “Dial up was ‘slow internet’… Broadband was ‘fast internet’… And the companies that made that transition possible made their shareholders incredibly wealthy.” In his telling, training is dial-up: it gets you there eventually. Inference at scale is broadband: it changes what products are even possible. Real-time coding agents, instant customer service, live cybersecurity analysis, the use cases he attributes to the teased company’s partners at OpenAI, Microsoft, IBM, AMD, Amazon, CrowdStrike, and GlaxoSmithKline all need answers in milliseconds, not seconds.
He anchors the timing to Elon Musk’s buildouts: Terafab, the $119 billion Texas chip plant, and Starmind, the FCC-filed plan for “up to one million AI satellites,” both of which he argues will multiply demand for fast inference. The near catalyst he cites is an AWS deployment, with details he expects on an analyst call, framed against a “before November 11” deadline.
Considerations
Nvidia is not losing its training fortress, and investors should be careful with any framing that says otherwise. Hyperscalers still order GPUs by the hundreds of thousands, and Nvidia’s annual cadence keeps raising the bar. The open question is the margin: how much of the fast-growing inference workload migrates to architectures purpose-built for it. Brown’s own hedge is the right posture for readers too: “I’m not recommending folks put their entire savings in this one stock.” His service’s 39.4% average gain since inception is the realistic anchor, not the 37,800% headline.
Where to learn more
Brown’s specific ticker and analysis are in “The One-Stock Retirement Plan,” included with a $179 trial of The Near Future Report, backed by a 30-day keep-everything refund guarantee. For background, read our review of Brown’s 10X Project pitch, the Near Future Report profile, and our the presenter presenter profile. For how another publisher frames the same AI memory and compute supply chain, our HBM memory explainer is 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.