Memory makers are responding with a huge build-out. Building a fab shell and cleanroom takes years, which is why the crunch has lasted so long, but many of these projects are now well underway. New capacity is what ended past memory shortages. New factories going in:
SK hynix
M15X, Cheongju, Korea: first wafers early 2026 (reported), ramping through 2027 Yongin Y1, Korea: first cleanroom Feb 2027 (company target) Yongin Y2, Korea: groundbreaking planned July 2027, first cleanroom June 2029 Micron
ID1, Boise: first wafer output mid-2027 Tongluo, Taiwan: an existing PSMC cleanroom it bought for $1.8B, with meaningful shipments mid-2027 ID2, Boise: late 2028 Hiroshima, Japan: broke ground July 2026, with initial output late 2028 Clay, New York (up to 4 fabs, $100B): broke ground Jan 2026, with first output in 2030 Samsung
P5, Pyeongtaek: mass production 2028 (reported) Yongin national complex: site work began Oct 2026, with the first fab targeting Oct 2029 (reported; Samsung hasn't given a date) Nanya
Fab 5A, New Taipei: 10nm-class EUV DRAM, with wafer starts in 2H 2027 (reported) Tesla / SpaceX (Terafab)
Research fab at Giga Texas, Austin: broke ground April 2026. There's no official output date. Grimes County, Texas (logic, memory, packaging; Phase 1 over $16.8B): land clearing since Aug 2026. Texas filings say commercial operations in 2028, but SemiAnalysis estimates late 2029.
The first real supply relief is 2027, a second wave comes in 2028–29, and a lot of that new capacity is aimed at HBM and server memory, so consumer DDR5 and GDDR will likely ease last.
Sources: Micron FQ4 2026 call, the SK hynix newsroom, TrendForce, Texas JETI filings via ENR, and Tom's Hardware/SemiAnalysis.
It's been like a year-ish...
And maybe 2027 is a truly optimistic relief on some level,
> a second wave comes in 2028–29, and a lot of that new capacity is aimed at HBM and server memory, so consumer DDR5 and GDDR will likely ease last.
'Consumer DDR5' has a pretty big set of knock-on effects though.
Unless you are working at a 'tech' company, the norm is that even developer machines will be constrained on memory usage for a given workload[0][1]. IOW companies trying to keep procurement in check wind up looking at procurement cost vs lost productivity.
[0] - Modern infosec and corporate software, at least where I'm at, seems to be vaguely more efficient than electron apps, IOW I can reboot, open teams and my IDE, after 4GBish of my IDE and 1GB of a fresh Teams instance I'm already around 16 of 32GB.
[1] - I had to fight hard to get 32GB as a developer baseline at my org. They wanted 16GB. Ironically the JVM based devs get 64GB but still have slowdown issues on biuld...
Honestly I think most of us are just spoiled kids crying when our expensive toys are being taken away. If memory cost 10x more to manufacture than it does now, we’d just have 10x less ram in our computers and our software would be written to deal with that. We’ve done it before. My first computer had 4mb of ram and it ran windows 3.1 just fine. We still had working compilers. If ram were 10x cheaper before the supply shock, I think we’d be crying that companies won’t buy us 320gb ram in our laptops. The java and rust compilers would somehow be even less ram efficient. Firefox and chrome would still happily use it all to run adware across the same 50 tabs I want to keep open.
Not everything. Theres good reason for LLMs to need that much ram. And scientific computing. And .. well I can’t think of much else.
The supply crisis is annoying. But I’m also a little glad there’s a reason to push back on fat, lazy software.
Good battery life is probably the biggest upside, but the Intel Core Ultra Series 3 Zephyrus G16 is just about as good (more on that alternative later).
The downsides are numerous:
- Middling GPU performance compared to discrete RTX mobile GPUs. You’re barely trading blows with an 5070Ti with the top end Spark chip. And yes, Nvidia markets RTX Spark for gaming, so we need to judge them in that context, too. Windows on ARM isn’t amazing for gaming for various reasons including lack of anti-cheat compatibility.
- Compatibility and/or performance issues on Windows ARM. You’re running a lot of stuff non-native
- As you mentioned, the CPU cores are nothing to write home about with very mid single core performance.
- Suffers greatly from “why am I not buying a MacBook Pro?” syndrome
- Poor/no Linux support, in a world where when Linux and Mac get the most attention for AI workflows.
I think there may be a case for being a good performance value for AI training, but for local inference I’m not seeing how that makes sense. Heck, for local inference I think it’s better to run a workstation as a server and then use the laptop as a client. You benefit from a better hardware value as well as putting less of the strain on the laptop, saving battery life. The fact that it’s plugged in means you can run long-running jobs and scheduled tasks, which is often the strength of local AI. The cost of one of these RTX Spark laptops will get you a dual R9700 inference workstation if you’re smart about the build.
You’ve also got projects like Strata that are going to define trends for how people run local AI. Why bother with $6000 integrated graphics machines or trying to squeeze quality models into one or two expensive 24-32GB GPUs when you can get 60 tok/s on Qwen 3.8 Flash Next on a 16GB RX 9060XT ($500) with 64GB of DDR4 RAM?
Want to do local inference on a laptop? Grab a Zephyrus G16, get the 64GB/RTX 5080 spec, run Strata, then as a bonus you can run AAA games at full blast. This costs less than $5000.
Waiting ~1 year for the M7 Pro chip on MacBook Pro models is another option that is almost certainly a better idea if the rumors about that chip are true.
I’m struggling to find a way to find a worse value buying something, anything that isn’t an RTX Spark. For that money there are just so many alternatives.
10 months ago when I got my Spark it was competitive in price with a similarly spec'd out Mac Studio. As RAM prices have gone even more parabolic, so has its price. And so has the Mac's. And the Strix Halo competitors.
That said the compute power of the GB10 is awesome, my DGX Spark-like has been better than I expected.
Not everything with AI is LLMs running on llama.cpp.
Tailscale is the key ingredient.
Aside from this, I’m led to believe that the AMD stack when it comes to local models has more overhead related to its need to translate more NVIDIA native instructions and therefore can be less performant than Nvidia. Not sure how true this is anymore, but ive yet to hear otherwise.
A Ryzen AI Max+ 395 powered tablet with 70Wh [0] gets the same battery life given the same task as a Nvidia N1X powered laptop with 99.9 Wh [1].
Power Consumption in their "Cyberpunk / Stress Test" is also lower at Ø109.3 (108.8-110.8) Watt for the x86_64 vs Ø134.5 (132.1-137.5) Watt for the arm64 system using the same settings, resolution and display brightness. They also get the same performance when normalising for resolutions and settings in Cyberpunk, so this is not an instance of one providing significantly more performance.
[0] https://www.notebookcheck.net/Asus-ROG-Flow-Z13-GZ302EA-Conv...
[1] https://www.notebookcheck.net/Asus-ProArt-P16-RTX-Spark-Lapt...
[1] https://www.lttlabs.com/articles/2026/10/16/nvidia-rtx-spark...