All Issues
Issue 03 · July 2026 July 23, 2026

The chip that got expensive isn't the chip in your pocket.

Apple raised its prices and blamed AI. This time AI really did it — but not the way it feels, and not even with the memory in your laptop. Here's the part it's actually fighting you for, and who's cashing in.

The moral high ground always has another tee box. Your friend has a beautiful swing and terrible aim — every issue they drive the ball straight at AI, and every issue it lands two fairways over. First it was the water, until the almond latte in their hand turned out thirstier. Then the power bill, and they sliced that shot too, landing in the rough next to the real culprit: the whole internet humming at once.

This time, for once, the ball lands on the right fairway. In late June 2026, Apple raised prices on 14 products — every Mac, every iPad, the Vision Pro — by $30 to $1,300, its worst market day in over a year, and blamed the cost of memory chips driven by the AI boom. "See? Told you," they said — and they're basically right, AI demand really did reach into their wallet. But they're already lining up a putt at the wrong hole: blaming Apple, when the hand that moved the price is somewhere else entirely.


The one idea

AI didn't make everything more expensive. It made one specific thing expensive — and that thing is hidden inside almost everything.

What AI did was bid up the price of memory — the RAM chips in your devices — by buying it in quantities the world never built for. (Storage chips, the NAND in an SSD, are a related but separate squeeze — not what this issue is about.) Memory lives in your phone, your laptop, your car, the smart TV, the server hosting this newsletter. Squeeze that one part and the price ripples through everything it's buried in.

AI Context School · Issue 3
One part, hidden inside everything you own
EVERYDAY DEVICES · WHERE MEMORY LIVES Phone Laptop Console Car Smart TV This server MEMORY the one part in all six AI didn't touch the whole product — it pressed on this one part, which happens to live inside all of them.
Diagram: AI Context School. Illustrative — DRAM (system memory) is a standard component across consumer electronics categories.

Start with the real number.

Memory prices didn't drift up — they detonated. DRAM rose 98% in Q1 2026 alone, projected to climb another ~60% the next quarter — "RAMageddon." Memory used to be a tenth of what a device costs to build; now, in some 2026 reports, it's the single biggest line item. (CNBC, IDC) Apple held out for months. In late June, it stopped absorbing it.

AI Context School · Issue 3
Memory prices didn't drift up. They detonated.
DRAM PRICE INDEX · Q4 2025 = 100 100 Q4 2025 198 (+98%) Q1 2026 ~317 (proj.) Q2 2026 (est.)
Source: CNBC / IDC, Q1 2026 DRAM spot-price rise ~98% QoQ, ~58–63% further rise projected Q2 2026 ("RAMageddon"). Even legacy DDR4 rose ~50% QoQ in the same period, atypical for an aging part.

So why is memory suddenly scarce — and why can't they just make more?

Your RAM didn't get pricier because the world ran out of sand. Memory is silicon, etched in factories called fabs — dozens of them, spread across Korea, China, Japan, and the US. But almost all of them are owned by the same three companies: Samsung, SK Hynix, and Micron control about 90% of the world's DRAM supply between them. (SemiAnalysis) A real new competitor isn't a fast fix, either — a single new fab costs $15–20 billion and takes two to three years, and nobody's stood up a fourth major player from scratch in decades. So when AI shows up paying more, the same three companies do the fast thing available: re-point the fabs they already have.

AI Context School · Issue 3
Three companies. Every flavor. One shared bottleneck.
SHARE OF GLOBAL DRAM SUPPLY Samsung 38% SK Hynix 29% Micron 20% Other 13% THE SAME SHARED FABS three owners make every flavor DDR desktops LPDDR phones, thin laptops GDDR GPUs, consoles HBM AI accelerators A new fab costs $15–20 billion and takes 2–3 years to build. Dozens of fabs worldwide — but ~90% owned by three companies, and no new rival in decades. That concentration is the chokehold that turned a demand spike into "RAMageddon."
Sources: Q1 2026 DRAM share estimates (Samsung/SK Hynix/Micron); fab cost/timeline — SemiAnalysis, ChipHistory, multiple 2026 trade press.

Here's the twist: AI doesn't even use the memory in your laptop. It wants a premium flavor called HBM, built for AI accelerators — not the DDR in a desktop or the LPDDR in your phone. It's not preference; the ordinary kind is architecturally too slow to feed an AI chip. (Scientific American) But all these flavors roll off the same fabs, and HBM eats 3–4× the factory space per chip — less room for everything else, so your ordinary RAM gets pricier even though AI never touched it. (TrendForce)

A quick map of the flavors. One family, different jobs: DDR is the stick you can hold (desktops, older laptops); LPDDR is soldered in for low power (phones, thin laptops, cars); GDDR drives graphics (GPUs, consoles); HBM is the stacked, top-shelf kind built for AI. Feasting on one flavor starves the rest, because they share factories.

The makers aren't rushing to overbuild, either — they've been burned before by flooding the market — and a June 25, 2026 lawsuit (Garciaguirre v. Samsung Electronics) accuses the three of using the HBM shift as cover to curtail older memory production, unproven but consistent with the shape of this issue. (Tom's Hardware) Worth widening the blame past the chipmakers, too: hyperscalers have locked in real, multiyear HBM contracts, so the demand isn't fake — but AI data-center spending is also running at a pace plenty call a bubble, which means some of this squeeze is chasing demand that may not last. Real shortage and speculative frenzy at the same time.

AI Context School · Issue 3
Same factories, different customer
SHARE OF DRAM WAFER CAPACITY 2025 Consumer memory (DDR/LPDDR/GDDR) · 81% HBM · 19% 2026 Consumer memory · 77% HBM · 23% BUT — HBM eats 3–4× the wafer space per chip. So a modest wafer-share shift (19%→23%) squeezes the memory in your devices far harder than the percentages alone suggest. AI didn't build new factories. It rented the same ones as a richer customer.
Source: TrendForce wafer-share data (HBM ~19% → 23% of DRAM wafers, 2025→2026); 3–4× figure is HBM vs. standard DRAM wafer-space multiplier.

So who actually absorbs it?

Not Apple, in the end — you. Same shape as last issue's grid: AI plugs into shared factories, and the cost splits across everyone who buys anything with a chip in it. The person who "doesn't use AI" still pays the tax at the register, buried in a part they can't see. (CBC) And for a growing share of devices, that cost is permanent: most phones and more laptops now solder memory to the board, so you can't add RAM later — whatever you pay on day one, you pay for the life of the machine. (Notebookcheck)

AI Context School · Issue 3
For soldered devices, the price you pay on day one is the price forever
DESKTOP — socketed RAM A module you can pull out and replace. Buy Upgrade later ✓ Pay 2026's price only for 2026's chip. Add more when prices ease. PHONE / THIN LAPTOP — soldered Fused to the board — not removable. Buy Locked at the day-one price — for the life of the device. The upgrade path that used to soften a rough pricing year is quietly disappearing from more and more devices.
Illustrative. Trend toward soldering memory (space/speed/battery) — Notebookcheck, xda-developers; EU Right to Repair Directive requires member-state implementation by mid-2026.

Which means the friend's anger isn't misplaced — it's just aimed at the wrong company. Call it corporate greed, because that's not wrong: a shortage is extremely good for margin, and the three companies controlling it are in no rush to end it. Here's the number that makes it concrete — while you're paying more for a phone, SK Hynix (one of the three) posted a 72% operating margin in the first quarter of 2026, an all-time record for the chip industry — higher than Nvidia's. (DigiTimes) That comparison is the tell. Nvidia is the poster child of the AI boom because it makes the AI chips themselves — it owns roughly 80–90% of the market and sells GPUs that cost a few thousand dollars to build for $30,000-plus, on ~75% gross margins. (Quartz) For a memory maker — supposedly a boring commodity business — to out-earn that is the whole story: the shortage emptying your wallet is the best year they've ever had, and the lawsuit above alleges it isn't an accident. Apple absorbed the hit as long as it could, then passed it down — but the ones actually profiting off the squeeze are the memory makers, and the AI buildout paying them a premium to keep it that way.


Concede the hard part.

None of this makes it hurt less. The pain is real and regressive — hardest on the student stretching for a laptop, the family replacing a dead phone. It's not a blip: this is a structural supercycle, and Gartner expects it to shrink PC and phone shipments this year. (Gartner) But there's a second thing, and it's the part that does something.


The re-aim

The cost is real. So is your leverage — aim it at the dial, not the price tag.

They've done this before, and it barely cost them. Between 1998 and 2002, these same companies rigged the price of memory sold to Dell, HP, IBM, and Apple. In 2005 they pleaded guilty to criminal charges — Samsung fined $300 million, Hynix $185 million — and more than a dozen executives were sent to federal prison. (DOJ, Tom's Hardware) They settled a follow-on suit for $345 million in 2006. This June's case is the third time. The reason you've never heard about any of it is the whole point: memory is an invisible part, and the fines were pocket change — roughly $730 million in penalties against a market now handing SK Hynix alone a 72% margin. Getting caught was cheaper than stopping. This isn't a scandal that ended. It's a business model.

So be honest about what actually moves a company like this. Not smarter shopping — enforcement. The only forces that have ever disciplined this cartel are antitrust regulators and real competition, and both are weak right now: the 2018 version of this lawsuit was thrown out for lack of proof, and the memory market is more concentrated today than it was in 2005. Follow the current case. Push the regulators whose actual job is policing three companies with a 90% grip and a rap sheet. Whatever ends this, it will not be the three of them deciding they've squeezed enough.

You can also refuse the lock-in that makes the squeeze pay — buy repairable, choose socketed RAM where it exists, back the right-to-repair laws moving through the EU and US states. But be clear-eyed: that's harm reduction, not a fix. An oligopoly this entrenched doesn't get beaten at the checkout. And don't feed the panic while you're at it — a stampede to buy ahead of the next hike is the next hike.


Context is the skill.

AI bid up one hidden part, memory — not even the flavor in your laptop, a cousin built in the same factories — and that part lives inside everything you buy. Apple isn't the villain, but it isn't innocent either: it passed the cost down instead of eating the margin hit. The real profiteers are three companies posting record-shattering margins off the shortage — companies that already pleaded guilty to rigging this exact market once, and are being sued for doing it again. "The market did it" is the story they'd like you to accept. The market is three convicted price-fixers with a 90% grip and a lawsuit pending.

That's what context buys you here: not a shrug at the register, but a target. Watch the case. Buy what's repairable. Push for more competition, because that — not their goodwill — is the only thing that's ever brought memory prices down.

Context is the skill — not just for AI, but for any price hike where the headline names the wrong culprit and lets the right one hide behind "the market."

Know someone who just rage-quit their cart over a price tag? Forward this — then go read who's been convicted of rigging this market before.

Sources & Notes
The price hike & the numbers

Apple raised 14 products June 25, 2026 ($30–$1,300; Macs/iPads/Vision Pro/HomePod/Apple TV, not iPhone/Watch/AirPods), blaming AI-driven memory costs; worst market day in over a year — CNBC, Bloomberg, MacRumors. Microsoft also hiking — CBC.

DRAM +~98% Q1 2026, +~60% projected Q2 ("RAMageddon"); DDR4 legacy +~50% QoQ; memory now up to ~35% of PC bill-of-materials — IDC, Tom's Hardware, Gartner.

How memory is made & who makes it

Three makers (Samsung ~38%, SK Hynix ~29%, Micron ~20%) make ~90% of DRAM from dozens of fabs across Korea/China/Japan/US; a new fab costs ~$15–20B and takes 2–3 years — SemiAnalysis, ChipHistory.

HBM (for AI) uses 3–4× the wafer space of standard DRAM and grew from ~19%→23% of wafers 2025→2026; ordinary RAM is too slow to feed AI chips (the "memory wall") — TrendForce, Scientific American.

The profits & the rap sheet

Record margins: SK Hynix ~72% operating margin Q1 2026 — an all-time semiconductor record, above Nvidia's ~65% operating margin (Nvidia's ~75% is its gross margin; it makes AI GPUs costing ~$3K to build, sold for $30K+) — DigiTimes, Quartz.

The current suit: Garciaguirre v. Samsung Electronics, filed June 25, 2026 (N.D. Cal.), alleges the three coordinated to restrict DRAM supply — prices up ~700% over four years; ~90% combined share; unproven — Tom's Hardware, Quartz.

The history: 1998–2002 DRAM price-fixing conspiracy (victims incl. Dell, HP, IBM, Apple); 2005 criminal guilty pleas — Samsung $300M, Hynix $185M, Infineon $160M, Elpida $84M, and 12+ executives jailed; Micron escaped via DOJ leniency; a 2006 civil case settled ~$345M; a 2018 suit was dismissed. This is the third major action — DOJ (2005), DOJ (exec plea), Tom's Hardware (history).