A 2nm chip in a box you can hide behind a monitor, a DRAM market that won't stop hurting, and Intel losing ground it hasn't lost in three decades. That's Tuesday for you.
Apple dropped the M6 and M5 Ultra yesterday, and honestly, the M6 is the one I keep staring at. It's the first 2nm chip Apple has shipped — 12-core CPU, 12-core GPU, and a pair of 16-core Neural Engines riding along. The new Mac mini comes with the M6 or the M5 Pro from earlier this year, pre-orders are open, and boxes start showing up on doorsteps September 22. A 2nm desktop machine, small enough to disappear behind a monitor, for not-crazy money. We're basically living in the future now, and nobody blinked.
To be fair, the M5 Ultra is the actual showpiece. It's a quad-die monster — four dies stitched together with a 4.4TB/s interconnect, up to an 80-core GPU, and you can spec a Mac Studio with 512GB of unified memory. The catch? A maxed-out Mac Studio with the M5 Ultra will run you $18,299. That's not a computer anymore, that's a down payment. And it's partly the RAMageddon talking — Apple's memory pricing has been brutal this generation, which brings me to the part of the industry that's quietly on fire.
The DRAM shortage is the worst in years, and it's not subtle. Conventional DRAM contract prices jumped 90–95% in a single quarter. Memory now eats roughly 30% of hyperscaler capex, up from about 8% in 2023 and 2024. Meta is already running into trouble. Marvell showed up at FMS 2026 pushing a three-tier "AI memory infrastructure" — and in a move that says everything about the moment, they're pitching DDR4 recycling for CXL memory. Reusing old DIMMs because new ones cost too much. That's the kind of desperation you can't fake. On the consumer side, this is why a 32GB kit costs what it does now, and why the mid-2027 horizon for relief keeps getting quoted. PINE64 just halted its Linux hardware line over DRAM and eMMC shortages with nothing expected back before mid-2027. Open hardware fans are feeling it.
One bit of good news to balance the doom: AMD had its best quarter ever against Intel, per Mercury Research. Intel's x86 share slipped below 70% for the first time since 1995 — 69.3%, a 6.5-point year-over-year drop. AMD grabbed 30.7%, a company record. On the server side, AMD's adjusted share sits at 46.4% against Intel's 53.6%. But here's the wrinkle nobody's talking about loud enough: Intel's data center revenue was up 59% year-over-year in the same quarter. They're losing share and making more money at the same time. That's not a collapse, that's a rebalancing — messy for enthusiasts who pick sides, healthy for a market that had gotten too comfortable.
From my perspective, the most interesting hardware this week isn't the flagships. It's the mini PC lane. AceMagic's new F2A squeezes a 16-core Intel Panther Lake chip, 32GB of RAM, a terabyte of storage, and dual 2.5G Ethernet into something that sits on a desk and sips power. And there's a roundup of five mini PCs under $350 for students that makes a decent case for skipping laptops entirely — if your machine never leaves the desk, why are you paying laptop prices? A monitor, a keyboard, and $300 of mini PC gets you surprisingly far. I've said it before and I'll keep saying it: the back-to-school desk build is the sleeper value play of the season.
Quick add-on note on the AI side: Nvidia is now giving away open-weight Nemotron models and free inference APIs, the old printer-ink playbook — hand out the razors, sell the blades. It's smart, a little cynical, and it's going to work. The free tier hooks developers, the GPUs follow. There's also chatter about GPU cloud operator Lambda raising $3B ahead of a possible IPO. Take that with the usual grain of salt — unverified industry rumor, for reference only.
Look, if you're staring at your budget trying to decide between a new kit of RAM, a cheaper mini PC, or holding out for prices to drop — you're not alone, and the math genuinely matters this month. By the way, if you've got a number to run while you're deciding — NumberJoy does the heavy lifting on quick calculations without the sign-up friction. No email, just the answer.
That's it for today. The 2nm Mac mini is probably the right call if you're in the Apple ecosystem and can wait three weeks; the rest of us keep watching RAM prices and quietly hoping.
