The AI Boom and Broadband

I’ve written several blogs recently about the impact of AI on broadband. For example, it’s becoming clear that a lot of AI applications will require better upload broadband speeds. ISPs that haven’t yet upgraded upload speeds will likely find themselves at a competitive disadvantage, similar to what happened to cable companies during the pandemic, when it became clear that upload speeds were inadequate to support multiple folks working and schooling from home. I also wrote a blog that discussed the big increase in web traffic that has sprung up from AI web crawlers that constantly search the web for new content.

There are other, more subtle, impacts from the AI boom. One of the unexpected issues is the sudden shortage of electricians in regions where AI data centers are being constructed. Electricians are needed during the massive effort to wire a new data center used for AI. What was unexpected is that data centers permanently employ multiple electricians to maintain the cooling systems and other ongoing needs at an AI data center. Data centers are offering attractive salaries and benefits and luring electricians who would otherwise have been busy building new houses or maintaining existing ones.

I’ve been reading stories about parts of the country where there is a sudden shortage of electricians. This can impact any ISP or other business that wants to build or upgrade a building. Folks might be surprised to know that ISPs use electricians regularly when they add new equipment or build new outdoor huts, cabinets, and repeater sites.

A more subtle impact of an electrician shortage is that it makes it harder to train new electricians. Most new electricians begin their career by being an apprentice to a licensed electrician, and if those electricians are taking full-time jobs at data centers, they are no longer working with new apprentices. Admittedly, a shortage of electricians is mostly a local issue, but it can be a real concern to ISPs who are trying to quickly expand networks.

Of more consequence is the impact that rapid data center expansion is having on the global market for memory chips. The memory market is under huge stress. While this market has historically seen boom-and-bust cycles, this time is different.

AI servers require far more memory per system than consumer devices, so the AI build-out is snagging a disproportionate share of global chip capacity and creating shortages, as suppliers prioritize orders for hyperscalers and vendors specializing in AI servers. This sudden market shift means fewer memory chips for consumer devices. For now, that is triggering higher prices for smartphones, computers, automobiles, and other consumer electronics.

But industry experts say this is not a cyclical shortage caused by a temporary mismatch in supply and demand. This is instead a permanent reallocation of the world’s silicon wafer capacity. For decades, the companies that made DRAM and NAND Flash for smartphones and PCs were the primary drivers for chip production. Today, the huge demand from hyperscalers like Microsoft, Google, Meta, and Amazon has flipped the market. In the last year, we’ve seen companies like Samsung Electronics, SK Hynix, and Micron Technology leave the consumer chip market. These manufacturers, and others, have pivoted to making higher-margin enterprise-grade components. Every wafer allocated to Nvidia is no longer available for a smartphone or consumer laptop.

Manufacturers of consumer electronics live on thin margins. This includes vendors like TCL, Transsion, Realme, Xiaomi, Lenovo, Oppo, Vivo, Honor, and Huawei. There are already rumors of these companies discontinuing lines of electronics that they don’t think are marketable at higher prices. Anything they continue to make will be sold at a much higher price since memory is 15% to 20% of the bill of materials for most electronics. Worst are the stories leaking out from these vendors that they are being asked to pre-pay for multiple years of chips to even be included in the supply chain.

Telecom companies need a lot of memory devices. The electronics used in networks are packed with chips. A more subtle impact over time will be the many broadband customers who find themselves unable to afford a new computer or tablet. This is also going to puta huge crimp on firms who have been refurbishing computers to support low-income homes. We’re just starting to see the impact of this change today, and chip experts say the full impact won’t be felt by the end of this year.