Why are CPUs and GPUs so hard to make? Dr. Ian Cutress explains

Dr. Ian Cutress, also known as TechTechPotato on YouTube and the industry analyst behind More Than Moore, is a frequent and welcome guest on PCWorld’s videos and The Full Nerd. And with him in the studio in San Francisco, it would be a crime not to pick his brain on the topic of CPU manufacturing. So that’s exactly what Will did.

First up: what does it take to actually make a silicon chip in a factory? Or to be more accurate, a fabrication plant, or “fab” for short. I once heard it described as “A bag of goes in a machine on one side, the machine punches the bag, and you get a CPU on the other.” The reality is a little more complex. We’re talking thousands of individual steps with hundreds of different components and processes, some requiring scare and precious materials. And that’s just the physical stuff that goes into a chip. Actually getting the factory up and running is the work of years of planning and developing, and billions — sometimes hundreds of billions — of dollars from start to finish.

And of course, things are getting more and more complex as processors become more powerful, using smaller components and processes on the nanometer scale. While a fab can continue making chips for years or decades even after it’s no longer cutting edge, the new stuff is so difficult that it’s getting consolidated into giant competitors. Right now it’s Taiwan Semiconductor Manufacturing Company (TSMC), Intel, and Samsung. Between them they make chips for pretty much everyone, from AMD and Nvidia to Apple, Qualcomm, MediaTek, et cetera.

But there’s a new “focused effort” to make a new 2-nanometer fabrication plant from a company called Rapidus, which is the spearhead for a new push for chip production in Japan. And like the other (remaining) big players, Rapidus would court production from massive clients like those listed above, along with partner companies in the country including Sony, Toyota, and Softbank, among others with the blessing and tax support of the Japanese government. It should be ready to start making absolutely bleeding-edge chips by 2027.

So all this stuff is pretty high-end corporate stuff, and not much of it trickles down to us on the consumer side. But what about the chips that go into graphics cards? And if Nvidia, and to a lesser extent AMD, can make all the money in the world selling chips to the AI industry…why would they even care to sell chips to PC gamers?

As Ian explains, a lot of the same tech and development that goes into that AI moneyspinner also goes into standard, consumer-level graphics cards, to say nothing of other applications like game consoles. To say nothing of the extreme specialization — you can’t just get a low-binned AI chip and throw it into a $2,000 graphics card instead. It also helps to diversify and stay relevant to consumers…because investors are consumers, too. Even a billionaire occasionally shops for a laptop.

Thanks so much to Dr. Cutress for breaking down these complex topics for us. For more on the latest news and trends in the tech industry, subscribe to PCWorld on YouTube, and watch our weekly podcast The Full Nerd.