The Checkout-Line Tax
Where will consumers feel the cost of the AI buildout
Almost everything written about the AI buildout is about what it needs. More power, more memory, more advanced packaging, more chips. Far less is written about what it costs the people who are not building it and will never directly use it. That cost is real, it is already arriving, and it is landing in an order you can predict: phones first, then cars, then laptops and the long tail of everything else with a chip in it. And it is landing hardest on the people least able to absorb it, which is the part that makes it worth writing about.
Why it starts with phones
The device where the public notices first is the phone, and there's a clean reason. A smartphone is the most memory-dense consumer product relative to its price, it gets replaced every two to three years so a new cost shows up on the shelf fast, and its buyers are the most price-sensitive on earth. All three point the same way.
The mechanism is straightforward reallocation. AI data centers are consuming memory faster than the three makers can produce it, so those makers route their best capacity to high-bandwidth and server memory, where the margins are, and consumer devices compete for what remains. The numbers are not subtle. IDC projects worldwide smartphone shipments will fall 16.7 percent in 2026, a record drop, even as the average selling price rises 27.6 percent to $581. Mobile memory prices have risen on the order of 70 to 100 percent, with some entry-level memory components up several-fold, and the shortage is expected to run to at least 2028. Fewer phones, each one more expensive, is not a cyclical blip. It's a structurally higher cost base.
The tax is regressive
Here is the part that turns a price story into an equity story. The pain is not distributed evenly across the market. It is concentrated at the bottom.
On an entry-level phone, memory can now be around 43 percent of the entire bill of materials. On a mid-range device it's roughly 20 to 36 percent. On a flagship it's 10 to 15 percent, because the rest of the phone, the camera, the display, the brand, costs so much more. So when memory doubles in price, the cheap phone absorbs a body blow and the expensive phone barely flinches. The 173 million phones that shipped under $100 last year, overwhelmingly the phones of the developing world, are what IDC's analysts call an existential-crisis segment; brands are cutting the cheapest models, reverting entry devices to less RAM, and raising prices, sometimes all three at once. Meanwhile Apple held the iPhone 17e at $599 and doubled its base storage, delivering more for the same money.
Read those two facts together. The same shortage that forces a family in an emerging market to pay more for less phone lets the premium buyer get more for the same price. The tax is not just a cost. It's a regressive cost, heaviest on the buyers with the least room to pay it, and it shows up in quieter ways too: in emerging markets, a rising share of phones is now bought on financing, which is how a component shortage in a data-center supply chain turns into an interest payment for someone buying their first smartphone.
Cars are next, and slower
Automobiles are the second wave, and they behave differently in a way worth explaining rather than glossing.
A car has far more chips than a phone, hundreds to thousands, and the count keeps climbing with electrification and driver-assistance. Memory content per vehicle is projected to rise from roughly $111 today toward $275 by 2030, with per-vehicle memory demand tripling. But chips are still a smaller share of a car's total cost than of a phone's, and people buy a car once a decade, not every three years. So the same shock that's already visible on phone shelves reaches the car sticker slower and more diffusely, mixed in with a dozen other cost lines.
Slower does not mean absent. Automotive DRAM contract prices were revised in early 2026 to a record quarterly jump near 90 to 95 percent. Ford's CFO said publicly that while supply is adequate for now, the pricing pressure is already being built into forward plans, which is corporate language for "we will pass some of this on." A Chinese EV maker put the hit at three to five thousand yuan per vehicle on premium models, and an industry executive warned that automotive memory fulfillment could fall below 50 percent in 2026. The coping moves mirror the phone playbook: raise prices on technology-heavy trims, delay builds, or quietly simplify features. The buyer notices later, and blames the dealer or the economy rather than a memory shortage, but the money moves the same way.
Then everything else
After phones and cars comes the long tail, and it's the largest category by breadth even if each item is small. Memory is already around a third of the bill of materials on a laptop, and analysts expect the sub-$500 PC to become essentially unviable. More importantly, the price wave has already jumped the fence from memory into the rest of the chip world: microcontrollers, power semiconductors, image sensors, and analog parts have all seen increases, in some cases 15 to 50 percent, as the whole industry reprices around allocation. Those are the chips inside routers, appliances, thermostats, medical devices, industrial gear, the invisible silicon in nearly everything manufactured. None of it moves the way a phone price does, all at once and in public, but it seeps into the cost of ordinary goods across the board.
Say the economics precisely
This argument is easy to overstate, and the overstated version is wrong, so it's worth being exact, because the precise version is both more defensible and more interesting.
This is not "AI causes inflation" in the monetary sense. No one is printing money here. What's happening is a relative-price shock driven by allocation: a finite semiconductor manufacturing base is being auctioned to the highest-value bidder, AI infrastructure, and the goods that lose the auction get more expensive. It's cost-push, concentrated in specific categories, not a general price spiral. An economist will nod at that framing and dismiss the sloppy one.
Two honest bounds follow. First, magnitude: for most products chips are a real but modest share of the total cost, so outside the budget-phone tier this is upward pressure at the margin, not a doubling of sticker prices. Its power is that it is broad, persistent, and compounding, stacking on top of tariffs and rising power costs already in the system, rather than that it is individually enormous. Second, the budget-phone tier is the genuine exception, where memory is such a large share of cost that the effect really is large, which is exactly why the regressive point holds: the one place the magnitude is severe is the place the buyers can least afford it.
Who pays, and why it matters
Strip it down and the shape is the same ladder I keep coming back to, followed one step further than usual, out of the semiconductor industry and into the general economy. The cost of the buildout relocates downstream until it reaches someone who cannot pass it on. The hyperscaler pays more for memory and folds it into a cloud bill a business expenses. The device maker passes it to the retailer. The retailer passes it to the shelf. And at the end of the line stands a person buying a hundred-dollar phone or a used car, who pays the accumulated tax of the entire chain and has no one downstream to hand it to. That's the terminal rung, and it's a household, not a company.
There's a tension in that worth naming plainly, as economics rather than politics. The buildout is sold, sincerely, as a path to abundance, cheaper intelligence, better tools, eventually lower costs across the economy. That may well arrive. But the near-term mechanics run the other way: to build the infrastructure of future abundance, a finite manufacturing base is being redirected away from the ordinary goods people buy today, and the resulting cost falls most heavily on those with the least. Both things can be true. The long-run case for the buildout does not erase the near-term bill, and the bill has a distribution, and the distribution is regressive.
You do not remove a bottleneck. You relocate it. The industry spent this cycle watching the constraint climb the stack and, lately, descend below it. This is where it finally exits the industry altogether and arrives somewhere everyone can see it, if they know to look: the checkout line, in the price of a phone that costs more and does less, paid by someone who never signed up to fund a data center.