Not now, Armageddon

You’d think 2026 already had enough on, but no — someone has gone out and helpfully coined an entirely new genre of Armageddon: not nuclear, not biblical, but supply chain:

• Apple’s Tim Cook is warning memory chip prices will increase “significantly”  

• Micron’s ops guru Manish Bhatia says the shortage is now “unprecedented”, and

• Tesla’s Elon Musk is even floating the idea of his own in-house chips plant (‘fab’).

So what’s driving this impending RAMageddon?

Intrigue’s hard-core nerds will forgive us when we casually split chips into three families:

First there are the compute chips that usually dominate our attention given their AI role, particularly to train models so they can answer your urgent query about the best cocktail for when you’ve only got a splash of brandy: the answer is a sidecar btw. We’ve long tracked these frontier chips, their NatSec implications, and the debate on if the US thwarts China by hooking it on these chips, or cutting China off them (Trump is camp hook).

Second, there are the storage chips you use every day to save that selfie you grabbed at Chili’s, or your work spreadsheet named “Final FINAL Version - Seriously FINAL.xl”.

And third, there are the memory chips you also use every day, but perhaps without realising: if the above storage chips are like your fridge, then these memory chips are like your kitchen bench where you whip ingredients into a crisp sidecar run programs.

We’ll get angry emails about us separating families two and three, but the bottom line is there’s now a supply crunch hitting family three: memory chips. Why? AI.

That might seem counterintuitive given we already flagged it’s family one (compute) that’s really driving AI, but AI also needs vast amounts of family three (ie, memory, or RAM). And that AI demand also means suppliers are pivoting to higher-margin types of RAM (like HBM), squeezing supply for everyday RAM even further.

So with hyperscalers announcing AI data centres more quickly than Jacob Elordi can pivot to his next mexi-mullet, it’s no wonder RAM demand has now outstripped supply.

In fact, there are only three main RAM suppliers, and Hynix (🇰🇷) already pre-sold its entire 2026 output by last October, soon followed by Samsung (🇰🇷) and Micron (🇺🇸).

So, what does that all mean in practice? 

The thing is, AI isn’t the only use-case for this RAM. To the contrary, RAM sits in your desktop, laptop, smartphone, tablet, gaming console, smart TV, and beyond.

And with some chip prices spiking 75% in a single month, anything with a screen is now facing price hikes (Nintendo), delays (PlayStation), and/or weaker margins (Apple), let alone the gloomers warning entire gadget categories could get wiped by high prices.

Sure, suppliers say they’re ramping-up production, but the reality is this crunch means those three firms are now printing cash, so will all be watching one another to see who actually blinks first via a costly new fab that’d seize more market share but crash prices. Even if they announce new fabs, you’re looking at a 3-5 year wait. And the next one already due (Hynix’s project in Cheongju) won’t come online until maybe the end of 2027.

Plus realistically, while China has suppliers nibbling at the edges, the three main players above still control almost all key memory chip output, protected by big moats of complexity, yield, ‘packaging’, and customer qualification.

So that’s why RAMageddon is now upon us.