Not the Tokens
Here’s a set of numbers that look like ordinary business-page noise — until you line them up against the gear you put in front of clients every day.
Start with Apple. Tim Cook told the Wall Street Journal that price increases on Apple’s devices are now unavoidable — the cost of the memory inside them has gotten too big to keep absorbing. Not likely, not under review. Unavoidable. And the market believed him instantly. The moment he said it, the memory-chip makers rallied — Micron, Western Digital, SanDisk, South Korea’s SK Hynix, Samsung — their stocks jumped, because investors heard one thing in that warning: the price of memory is going up, and it’s going to stay up long enough to be worth betting on. When the companies that make a component start trading like they’re sitting on something scarce, that’s the market telling you the cheap era is ending. And the size of the move is the part that should land — the memory going into these devices has roughly quadrupled in price in a year. Not crept up. Quadrupled.
And it isn’t staying inside Apple’s supply chain. The reporting widened to the obvious next step — that Apple’s devices themselves are expected to get more expensive, with the rising cost of the chips named as the reason. So this isn’t a commodities-desk story that lives and dies on a trading screen. It’s a finished-product story. The thing getting pricier is the laptop, the phone, the tablet — the actual machine that shows up on a desk.
Now hold those two facts next to each other, because that’s the whole signal for now. The CEO of one of the most powerful hardware companies on earth says a core component’s price increase is locked in. The market agrees, with money. And that increase is already being passed down toward the devices people buy. No projection, no maybe — prices on the tin are moving up, the people closest to the supply are treating it as durable, and it’s heading for the endpoint.
That’s where we are before we’ve said a single word about why it’s happening — or what it does to a business whose whole promise is a predictable, flat monthly number.
An Auction for the Parts
The reason memory is repricing — and repricing in a way the market treats as permanent — isn’t really about Apple, or phones, or any one company’s costs. It’s about who’s standing in line for the same parts.
Look at what’s happening to servers, because that’s where the contest is visible. IDC counted the market for the quarter and found something that’s never been true before: non-x86 servers — the AI-focused machines, built largely around Nvidia’s Arm-based chips — generated fifty-eight point seven billion dollars, nearly forty-eight percent of all server revenue, up a hundred and seven percent year over year. More than doubled. That’s the AI build-out, showing up as a single number. And right beside it, in the same report, the ordinary x86 server — the workhorse that’s run business computing for thirty years — actually shrank, down almost three percent. IDC names the reason out loud: those conventional servers are being held back by shortages of DRAM and NAND. Shortages of memory.
Sit with that pairing, because it’s the whole mechanism. The AI machines aren’t just growing faster than everything else — they’re growing while the normal machines starve, and they’re starving for the exact same thing. Silicon and memory aren’t infinite. There’s one pool. And right now an enormous share of it is being bid away to feed AI infrastructure, which means the chips left for the laptop, the workstation, the standard server cost more — because something with deeper pockets got there first. That’s not a price hike. That’s an auction, and your client’s hardware refresh is the low bidder.
And the fix everyone points to only confirms how deep this runs. When Washington announced a deal for Intel to build chips for Apple on American soil, Intel’s stock leapt double digits in a single morning. But notice what that actually is: standing up chip capacity is a project measured in years and fabs, not quarters. The market cheered a solution that won’t relieve the squeeze for a long time. So this isn’t a spike that clears by next quarter — it’s the new floor, and it’ll still be the floor across your next several hardware refresh cycles.
Short Silicon
So what does this do to the MSP — the one whose whole model is a clean, predictable monthly number?
Start with where the channel already is, because the leading edge of this is visible. Industry reporting on the U.S. channel describes partners actively pivoting away from pure hardware resale toward services — with thinning hardware margins named among the headwinds pushing them off the boxes. Reselling tin, in other words, was already getting harder to make money on before our story even starts. Read that carefully. Before the squeeze we’ve been describing even fully arrives, the channel was already discovering that reselling hardware had quietly become a worse business — thin enough that the smart money is walking away from it. The component crunch isn’t the first hit to hardware margin. It’s the next one, landing on a channel that’s already bleeding there.
Now run that through your own agreements, because here’s the uncomfortable part. If you’ve bundled hardware into a flat all-in monthly fee — or you’re running Hardware-as-a-Service, or you’ve signed a client to a three-year refresh at a price you quoted on last year’s component costs — then you have, without ever deciding to, taken a position in the memory market. You are short silicon. Every machine you’ve promised to deliver at a fixed price, while the parts underneath it climb, is a bet you lose a little more on every quarter. You didn’t set out to become a commodities trader. The contract structure made you one.
So the choice this forces is clean, and it’s about where the risk sits. You can get out of the trade — pull hardware out of the flat number, make it a transparent line the client sees and you reprice at each refresh, indexed to what the parts actually cost the day you buy them — so the inflation is the client’s variable, not your margin leak. Or you can keep selling the all-in price, keep the bet on your own books, and hope the memory auction cools before your next refresh cycle does the choosing for you.
Why Do We Care?
The fair objection here is that none of this is new — hardware has always been a thin-margin pass-through, and handing a price increase to a client is the most ordinary thing an MSP does. That’s true, and it misses the one thing that changed: the objection assumes you can still reprice, and the entire point of the bundle, the Hardware-as-a-Service contract, the locked three-year refresh is that you agreed not to. The cost increase isn’t the new risk — the new risk is that you signed away the right to pass it on, right as the parts underneath started climbing.
What to Consider
Pull your exposure list before you do anything else. Go through every agreement where hardware is baked into a flat monthly fee, sold as Hardware-as-a-Service, or locked to a multi-year refresh at a fixed price — and flag the ones you quoted on last year’s component costs. That list is the set of contracts where you’re currently short silicon, carrying a parts-inflation bet you never decided to make. You can’t reprice what you haven’t found.
Put a component-index clause into every renewal and new SOW. The fix isn’t refusing to bundle — it’s keeping the right to reprice. Pull hardware out as a transparent line the client sees, tied to what the parts actually cost the day you buy them, or write a price-adjustment clause that lets the refresh number move with component cost. Do it as standard contract language now, so the next increase is the client’s variable instead of your margin leak.
Time the refreshes you control against the squeeze, not the calendar. Where a client’s refresh timing is yours to set, pull forward the ones you can fund at today’s prices rather than waiting for a floor that’s still rising. And stop quoting new multi-year hardware numbers into a flat fee while memory is mid-auction — locking a three-year price to a part that’s appreciating is the exact trade this episode is warning you off.
If this trend continues, within the next twelve to eighteen months the MSPs who pulled hardware out of the flat fee will be quoting refreshes at cost-plus while the shops still selling an all-in monthly number spend their renewals explaining a margin hole the memory market dug for them — and a component-index clause will be a standard renewal ask, not a clever one. And the clients who can’t absorb the increase will simply stretch their refresh another year, which quietly hands you an older, riskier fleet to defend — so this lands on your security promise, not just your margin.