The Middle Class of Tech Is Quietly Vanishing
A global memory chip shortage triggered by the AI boom is quietly squeezing budget phones and mid-range laptops out of the market.
Go try to find a smartphone for under $100 on Samsung’s website. You won’t. Head over to Best Buy, and you’ll find only two or three options in that price range. A few years ago, that shelf had a lot more on it. The cheapest tier of the market is the one shrinking fastest, and it’s not hard to see why. For these low-end smartphones, the bill of materials cost has increased 20% to 30% since the beginning of the year.
Laptops tell a similar story. You can still find a few for around $300, but good luck getting real work done on one. Manufacturers are putting so little RAM into budget machines that basic multitasking becomes a chore. Even Apple’s MacBook Neo, a laptop built specifically to hold a low price point, is having to get creative just to stay under $600. What’s left is a thin, underpowered bottom and a healthy top end for anyone who can pay for it.
What’s happening in these markets to cause such change?
Your first guess might be inflation, but that’s probably not it. Smartphones and laptops were exempted from last year’s “Liberation Day” tariffs. Your second guess might be a lingering supply chain crisis, the kind with backed-up ports and container ships stuck in canals from a few years back. But that’s not it either. There’s a simpler explanation sitting in plain sight.
Blame the Inputs, Not the Economy
While our focus has been on the output side of the market, every output starts as a stack of inputs. We actually teased one of those inputs at the start: the bill of materials, the actual cost of the parts inside the device. When that number changes, the sticker price eventually follows.
The input driving everything at the moment is memory, DRAM for active processing and NAND flash for storage. Only three companies make the overwhelming majority of the world’s supply: Samsung, SK Hynix, and Micron. Before you jump to collusion or price fixing to explain the results, it turns out a very large buyer has been in the room with them over the past year.
The same DRAM and NAND that goes into a laptop is what data centers need to run AI models. Last October, OpenAI signed agreements with two of the big three manufacturers to secure memory for a $500 billion infrastructure buildout. That single deal set off an industry-wide scramble to lock down memory contracts, and suddenly the buyer is much more visible.
AI data centers are on track to consume something like 70% of the world’s memory output, up from roughly 20–30% just a few years ago. Every wafer that becomes part of an AI accelerator is a wafer that doesn’t become RAM in your next laptop.
The Economics of a Scarce Wafer
This new supply crunch is actually a resource allocation problem in its purest form. Chipmakers can only produce so many wafers each quarter. Factories are running close to capacity, and building a new one takes years, not months. As a result, a fixed, limited supply has to be divided among competing uses, and something has to decide who gets it. In a market economy, that something is price. Whoever’s willing to pay more gets the wafer.
Of course, phone and laptop makers haven’t stopped needing chips. They still have contracts to fill and products to ship. But the industry is competing for those scarce resources against an industry with some very deep pockets. As a result, they end up buying less of the higher-priced wafers, and we’re seeing the results on the shelves in the form of fewer products at higher prices.
That’s the resource allocation problem resolving itself in real time: a fixed supply flowing toward whoever values it most, as measured by willingness to pay. The bill-of-materials increase from the start of this piece is what that reallocation looks like from the outside.
Allocatively Efficient, Not Exactly Fair
Go back to the chipmaker for a second. They are more or less obligated to their products to the highest bidder. A publicly traded company has a duty to its shareholders to make the most of its output, and right now, the most profitable use of a limited number of wafers is to sell them to AI data centers.
But “whoever values them most” is doing a lot of work in that idea. It doesn’t mean whoever needs them most. Rather, it means whoever has the biggest budget. Big tech companies have already poured over $1 trillion into the global AI infrastructure, even as people increasingly fight to keep them out of their backyards.
And yet the demand for AI services and the data centers built to serve them keeps growing, perhaps more quickly than a lot of people would like. While excitement has morphed into concern for many people, a growing share are interacting with AI several times each day. Our desire to get quick information from a chatbot, whether that’s a new recipe using ingredients in our pantry or a question about a suspicious mole on our leg, has resulted in a surge in demand for memory chips that are no longer being allocated toward affordable cell phones and laptops.
And this is the part that can feel uncomfortable for some people. None of this requires bad intentions anywhere in the chain of reasoning. Allocative efficiency is proof that the system is working. It just optimizes for one thing, willingness to pay, and treats every other kind of value as invisible.
Final Thoughts
For those of you relying on Moore’s Law as a source of optimism, you may need to brace yourself. For decades, the industry ran on an assumption that next year’s electronics would reliably be faster, smaller, and cheaper. That’s no longer a safe bet. Even Tim Cook has publicly admitted Apple can no longer fully shield customers from rising component costs, something the company has spent years managing to avoid saying out loud.
Unfortunately for those in the market for a new phone or laptop, it may be several years before a fix arrives. Micron’s planned New York facility won’t be fully operational until the 2040s, which means relief isn’t arriving quickly even if AI demand eventually cools.
To be clear, none of this is an argument against markets. They’re still remarkably good at moving resources toward their highest-value use, but it’s instead a reminder that “highest value” and “most needed” aren’t always the same thing.
Know someone who’d get something out of this? Forward this newsletter their way, or send them the link to subscribe.
Microsoft raised the price of its Surface Laptop models by $500 [The Verge]
The DRAM market is highly concentrated, with three companies accounting for 90% of the total revenue [The Motley Fool]
On June 25, seventeen plaintiffs sued Samsung, SK Hynix, and Micron in federal court in California, accusing the three under the Sherman Act of coordinating to restrict supply and inflate prices [PCMag]
Global estimates of spending on data centers could reach $7 trillion by 2030 [McKinsey & Company]
83% of Americans agree with a statement from the pope’s encyclical about AI on AI’s lack of ability to know what love, work, friendship, or responsibility mean [YouGov]