Coding while flying is often the use case people bring up, but the power brick these laptops come with sometimes trip the plane's power plug in coach and you can't run local AI on battery for long. Also, over time planes will get better connectivity (Starlink etc.) and that particular full-offline scenario will go away.
For gaming, there are better options with dedicated GPUs like the Asus Zephyrus and similar.
What am I missing?
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...
It seems like this list also doesn't account for China and state sponsored memory industry (which will be more immune to a downturn)
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.
The author’s discussion of their personal use case starts on the seventh sentence in TFA. And frankly, it sounds like a pretty solid use case:
“Being able to run things locally and mess around with RL, quantisation and fine-tuning on a laptop would be sick. I already publish my own GGUF quants on Hugging Face, with over 160,000 downloads in the past month.”
For example, network is not always available and stable; File transferring is limited by connection speed, hardware passthrough is basically nonexistence; Dealing with power management (Unexpected power loss) and so on.
I bet they all can be solved(to some bearable degree), but at that point, just bite a bullet and get a powerful laptop.
I'm sure you can come up with some scenario where you are programming in the wilderness with no internet connection but the market for that has to be in the tens of people.
That said the compute power of the GB10 is awesome, my DGX Spark-like has been better than I expected.
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...
Uh, well you'd just download lemonade server and use models designed explicitly for the hardware from hugging face.
It's supported by AMD.
Tailscale is the key ingredient.
but separately everything about Microsoft’s login experience is pure pain.
Multiple broken redirects, broken links, buttons that loop, and more just to hit a CAPTCHA AI can solve anyway.
The only thing worse is for the poor soul developers who have had to login to their dev portal to setup SSO Microsoft for an app. It’s the worst I have ever seen in my life (even using Supabase that hand holds you through it).
(Expired but there was coupon https://slickdeals.net/f/20164014-hp-omnibook-x-14-nvidia-rt...)
I've bought some low-priced things on amazon, just to have them cancelled. Spooky since I figure someone got my name, address, phone number from amazon.
The OS makes a big difference for most developers, and the DGX Spark stack and experience is mature.
Am I being too stubborn for not wanting a cloud account tied to my desktop operating system?
In fact if getting stalled by a single screen in the installer is a showstopper, then you're in for a world of hurt with Linux.
I wonder how its performance is on dense models.
Why would you ever run anything else on equipment like that, which doesn't make any sense to me when enterprise grade hardware that is stationary would be more powerful anyway...because you aren't taking that on BART (I don't steal but seeing that thing in your lap, might bend that rule for myself since no real chance Bart PD would care anyway) or probably ever moving it from the place you use it now in ways that a mobile rack in your garage would be any less convenient for you.
Or if you felt strongly about consumer facing hardware, meaning really you pay through the nose to live in an efficiency in "SoMA" aka the Tenderloin, buy the mini-pc variant is just as easy to move but much less easy to break.
What challenge has been overcome?
I can see that hardware was bought, and that's cool hardware and all, but consumption is rather uninspiring.
I suppose it's the arm64 part? Or the review of some new hardware other people might be interested in?
Because that sounds interesting! But I didn't see anything like that when I've (admittedly) skimmed through the review.
That was like three years ago with the release of M3, do you not count MacBooks as laptops or something? Not that you could run Linux on it back then, but you don't even mention the existence of Apple silicon at all?
For example, I cannot currentl use the ms suite, like word or excel, because it asks me to log into an institutional account that I dont have, despite my account being the owner of the licenses for the suite running on my laptop. Make that make sense. All because i opened an institutional word document on it, i've tried fully reinstalling, no dice. i've tried nuking caches and registry entries, no dice. Window's basic functionality is indistinguishable from maleware.
Windows Pro > Work or School Account > Join a Domain instead > Set password & recovery questions
The one component here that is not present on most machines straight from the manufacturer is the presence of the Pro version. The home version of windows requires the account with increasingly limited alternatives.
This isn't a new trick - just the regular approach to editing windows install media.
Shouldn’t have to do that of course, but it can be done.
Also that means you're forced into a manditory arbitration clase and a class action waver.
Well, reading the reviews it's clear these are filling the windows gap MacBook Pro gap in windows laptop lineups.
And the answer to "why am I not buying a MacBook Pro?" for many people is because it's a MacBook pro. I've never owned Apple hardware in my life and have no desire to do so now but I do wish there was equivalent quality non Apple hardware (for me that I can run Linux on, but for many people, Windows would be fine). If the price was lower, I'd be in the market for one of these.
So far the only options has been gaming/creator laptops with dedicated GPUs. I currently have an MSI Ge66 and it's a great laptop in many ways but it runs so hot, and the fan is so loud. I'd pay a premium (but not $6000!) to reduce that. Also, the unified memory is a big draw for both Blender and local AI.
Not everything with AI is LLMs running on llama.cpp.
The massively lopsided compute makes speculative execution techniques (MTP/DFlash etc) way more effective, which makes modern MoE models quite usable
I’m running Qwen 3.8 Flash Next with Nota AI’s conservative NVFP4 quant at 30tk/s lowest decode, usually more around 45-55 for coding, and with prompt processing at a solid 2000tk/s
And that’s without very optimised kernels or anything, mostly standard vLLM
Honestly, this whole line of commentary comes across very gatekeeper-ish. Isn't half of what makes it to the front page people showing off their cool shit?
Not defending MS btw.
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.
Theres good reason for LLMs to need that much ram. But very little else we do on our computers needs anywhere near as much ram as we use. Have faster computers made video games more fun in the last two decades? I don’t think so.
The supply crisis is annoying. But I’m also a little glad there’s a reason to push back on fat, lazy software.
You can buy a version from gog that runs on Windows 10 requiring 2 GB of RAM and an unspecified 1.8GHz CPU [2]