Nvidia lays out RTX Spark roadmap for laptops and desktop PCs at Computex 2026 — three generations outlined, Rubin with LPDDR6 memory, followed by Rosa Feynman

tom's Harware ^ | June 1, 2026 | Jeffrey Kampman

[fireman15]

Nvidia is fully committed to transforming Windows on Arm into an agentic AI platform.

Along with its first-generation RTX Spark platform for desktop and laptop PCs, Nvidia CEO Jensen Huang revealed the company's commitment to future generations of those platforms on its future roadmaps. The company is committed to producing at least two additional generations of Spark platforms for its partners.

Beyond the Grace Blackwell RTX Spark chips (the top-end RTX Spark Superchip and an as-yet-undetailed smaller chip), Huang promised that every future generation of the company's platforms will include a Spark chip.

That means there will be a Vera Rubin pair of Sparks powered by LPDDR6 memory, and a future Rosa Feynman Spark with a presumably even faster (but as-yet unannounced) memory generation. That multi-generational promise is an important point of trust in Nvidia's commitment to transforming Windows PCs for the agentic AI era.

Building a full product and partner ecosystem is a much larger challenge than simply building and shipping a chip. It's clear that Nvidia has a small army of OEM partners ready to take those chips to market and a deep partnership with Microsoft and ISVs to unlock the capabilities of its platforms for Windows and the applications that run on it.

---

TOPICS: Business/Economy; Computers/Internet; Hobbies
KEYWORDS: ai; linux; nvidia; windows

[TexasGator]

“32GB of RAM used to be about $90, now it is $390 +300”

Please provide a link to the $99 price. Nice cherry picking.

“4070 GPU used to be about $250, no w they are $850 +600”

That is a SUPER! Nice Cherry picking.

“Sandisk 1TB NVME used to be about $90, now they are $300 +210”

Please provide a link to the $90 price. Nice cherry picking.

[fireman15]

The Datacenter bubble will pop, since token-based billing isn't going to work. Nobody wants to pay full-freight for AI.
Jensen needs to move product, so a return to the desktop with local models makes sense.

Yes, this is a survival move. A bunch of manufacturers dumped producing consumer products in favor of the more lucrative Server Farm market. It is interesting that Nvidia seems to be adjusting their priorities. Of course, the Spark products are high end with 128GB of high bandwidth Unified Memory.

[CatOwner]

I'll let you too go back and forth. I have a link to the PC Part Picker website that I had saved and captured an image of in late September 2025. I went back to that link, made one small change on something no longer available, and checked the price.

The PC was spec'ed for 64GB of RAM (photo and video editing). Two SSDs, one 2TB and one 1TB. That's where my price increase went. If you don't need that much in a PC, fine. But that's what I wanted and expected to use 7-8 years.

[TexasGator]

“I purchased this 64GB DDR5 5600mhz kit for $210 in November,”

I would sure like to see that receipt!

[TexasGator]

“The PC was spec’ed for 64GB of RAM (photo and video editing). Two SSDs, one 2TB and one 1TB. That’s where my price increase went. “

Hmm. You would need a decent CPU and GPU for that.

$1400? No way.

[TexasGator]

“The PC was spec’ed for 64GB of RAM (photo and video editing). Two SSDs, one 2TB and one 1TB. That’s where my price increase went. If you don’t need that much in a PC, fine. But that’s what I wanted and expected to use 7-8 years.”

Mine purchased last year before price increases.

I9, 5080, 64GB, 2TB drive - $3300.

[bobcat62]

There’s a guy on YouTube, using the username Kiraa.
He points out in one video that a loaded Mac Studio could be depreciated for $280 per month.
That’s a big improvement over full-freight token-based recruiting.

[TexasGator]

My bad. $2600 after $400 Costco price reduction.

I almost bought it from the vendor for $3300.

[dennisw]

A woman was paying $2000 monthly to use Claude AI. She switched to Chinese/CCP Deep Seek AI. Price $5 monthly and 80% as good as Claude.

[dennisw]

Thanks to AI, the desktop PC I had spec’ed out back in September 2025 for around $1,400 would now cost around $2,600 for essentially the same components. Most of the cost increase is tied to RAM and solid state drive (SSD) costs. And guess what’s driving those costs through the stratosphere: AI data center.
____________

What video card? This must be a gaming PC. In Best Buy I saw no “regular” desktops. Only 10 diff gaming desktops. Plenty o laptops and tablets and Apple gear.

This will do great on internet $101 @ eBay >>> Dell Latitude 5520, i3-1125G4 @ 2.00GHz, 8GB RAM, 256GB NVMe-No OS/AC
Add a bit more RAM + external monitor.

[dennisw]

This guy will help you buy the right gamers PC. One expert said 16gb RAM is enough. Its the video card that counts.

https://www.youtube.com/@Weenietechbuilds/videos

[fireman15]

I would sure like to see that receipt!

I am glad that you are impressed; but the price had already gone up about $40 from the lows so I could have done better. We all know what a legendary super shopper that you are.

[bigbob]

This will drive the stock price to create even more”NVDIA millionaires”

[Mr. Blond]

Become an Nvidia millionaire so you can afford to buy a pc.

[fireman15]

A woman was paying $2000 monthly to use Claude AI. She switched to Chinese/CCP Deep Seek AI. Price $5 monthly and 80% as good as Claude.

Obviously, you are joking, but there are many ways to get more for your money with commercial AI models while still getting high quality results.

I like the technique used in this video using Claude Code with Ollama and locally run LLMs. Here is a discussion with Gemini 3.5 with the link to the YouTube video:
https://gemini.google.com/share/c5e608d85268

I also like to use multiple LLMs on a Hostinger VPS using Ollama with Open WebUI while only subscribing to a couple services and using APIs for the others. This is discussed in this YouTube video with a summary by Gemini 3.5:
https://gemini.google.com/share/0b38f7d1cf6d

[FrankRizzo890]

There will come a time in the not too distant future where we’ll look back and LAUGH about how we used to have to go online to use AI. I’m not tech-savant, and even I can see where this goes.

[FrankRizzo890]

Not yet. It’ll be the usual early adopters who will buy them and put them to use doing cool stuff at work and at home. And as technology marches forward the price/size will decrease, and in 10 years you’ll have a full blown AI chip in your phone.

[FrankRizzo890]

I built an OLTP data center out of Alpha based machines running Windows NT. They were LIGHTNING fast for the time.

[FrankRizzo890]

Which became popular because 3DFX decided to stop selling chips to OEMs, and to only make the boards in-house. A STUPID decision that killed the company. Which allowed the guy who made that decision to come be the new CEO of the company I was working for, and to kill us as well.

I’m kinda bitter.

[fireman15]

This will drive the stock price to create even more ”NVDIA millionaires”.

My guess is that Nvidia’s stock prices likely will decline as the demand from newly constructed AI Data Centers declines. What I am excited about is the potential of these machines and the competition they will provide to AMD and Apple for running up to 70B and even 90B Models along with providing iGPU performance that in many cases will be superior to GPUs. The competition should eventually drive the prices down.

The point of the thread was that for the last 10 or 15 years we have had only incremental computer improvements from year to year. This is a revolutionary change to PC architecture that will eventually dramatically increase the performance vs cost especially when running LLMs but also for many types of gaming and other demanding applications.