It turns out that there’s good reason for NVIDIA having become the world’s most valuable company in the midst of the AI revolution.
Dylan Patel, founder of the prominent semiconductor industry analysis company SemiAnalysis, recently offered a strikingly candid comparison that perfectly captures the current state of the GPU market. His analogy, while provocative, illuminates just how dramatically the artificial intelligence boom has transformed what was once a straightforward hardware purchase into something resembling an underground economy.
“So, just to be clear, my opinion on how you buy GPUs is that it’s like buying cocaine,” Patel explained. “This is described to me, not me. I don’t buy cocaine. Someone tells me this, someone tells me this. I’m like, holy shit. It’s right. You call up a couple people, you text a couple people, you ask ‘yo how much you got? What’s the price?’ Like this exactly.”
The comparison becomes even more vivid when Patel describes the typical GPU procurement process: “We have Slack connects with like 30 neo clouds and like there you go, as well as like some of the major ones. And we just send them a message like, ‘Hey, customer wants this much, you know, this is what they’re looking for.’ And then they send quotes and then I know this guy.”
This underground network approach to GPU acquisition reflects a market reality that would have seemed absurd just a few years ago. What Patel describes is essentially a gray market economy where high-end graphics processing units—particularly NVIDIA’s H100 and A100 chips—are traded through informal networks of brokers, cloud providers, and intermediaries. The process involves hushed conversations, whispered recommendations, and the kind of relationship-dependent transactions typically associated with illicit goods.
The implications of this comparison extend far beyond mere market dynamics. We’re witnessing the emergence of a genuine arms race in artificial intelligence infrastructure, where access to computational power has become as strategically important as access to oil or rare earth minerals. Major tech companies like Microsoft, Google, Meta, and Amazon are reportedly stockpiling GPUs, with some estimates suggesting that demand outstrips supply by factors of ten or more. The shortage has become so acute that companies are willing to pay premiums of 50-100% above list price, assuming they can even secure allocation. GPUs are also being smuggled in countries like China where the sale of some kinds of GPUs is restricted.
This scarcity has created a multi-tiered market where established relationships matter more than purchase orders, and where the ability to source GPUs has become a competitive advantage in itself. Startups flush with venture capital find themselves competing not just on technology and talent, but on their founders’ ability to navigate these informal procurement networks. The result is a market that operates more like a commodity trading floor than a traditional technology supply chain, complete with spot pricing, futures contracts, and the kind of relationship-dependent deal-making that Patel’s colorful analogy so aptly captures.