Nvidia’s 16 October laptop launch has a hidden price ladder, because a year of memory shortage pushed the average desktop memory price up 88% [1]. At $18.12 a gigabyte, 128GB of Double Data Rate 5 (DDR5) memory now costs $2,319 in parts. Nvidia’s cheapest 16 October laptop with that same capacity — 128GB of unified memory shared by the processor and graphics chip — sells for $5,109 [2][3]. The two numbers are the same purchase read from opposite ends.
The price list hides two ladders
Six makers put the same superchip into slim Windows machines — Asus, Dell, Lenovo and Microsoft among them — and the top configuration is advertised by its memory rather than its speed [2][3]. The core counts and prices split like this, with a graphics processing unit (GPU) count on every rung [3]:
| Rung | Price | Processor cores | GPU cores | Memory |
|---|---|---|---|---|
| Entry 24GB | $2,599 | 18 | 5,120 | 24GB |
| Full-chip floor 48GB | $3,999 | 20 | 6,144 | 48GB |
| Top 128GB | $6,999.99 | 20 | 6,144 | 128GB |
The 24GB and 32GB rungs both carry the cut chip. The full part shows up only at 48GB and above [3].
So the entry model is not a smaller version of the machine beside it. It is a different machine wearing the same name.
My read starts with the label. Nvidia’s release names one model that runs locally on day one: Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter model sized to a 24GB footprint [2]. That is exactly the capacity of the cheapest configuration, on a device where the operating system, the browser and an agent runtime (a background helper for AI tasks) also live. The label describes the model’s own footprint, not the room left on the machine.
What the gigabyte is worth
The index behind that $18.12 quote reads 211 desktop DDR5 kits daily, reports the median and the mean, and shows the 30-day move: 13.3% higher on 31 October than on 1 October [1]. Nine of the last eleven monthly readings were rises. The two declines, in May and June, were each followed by four straight increases [1].
What strikes me here: the ladder sells two things at once. The chip rung is fixed at purchase — no firmware revision or add-in card changes it later. The memory rung has a daily quote. A buyer who pays for the top step is buying the chip they needed plus memory at whatever the maker set before this month’s move.
Renting has its own clock. An Nvidia H100 accelerator rents for a median $3.84 an hour across 39 providers, up 14% in four weeks and 29% in a year, while a three-year reservation runs about 48% below on-demand [4]. An Nvidia H200, with 141GB, lists at $5.40 an hour on the dedicated clouds [4]. For a workload that peaks for a few weeks a year, the peak is the part to rent.
By the numbers
- $294 — what 128GB gained in thirty days, as the index mean moved from $17.36 to $19.66 a gigabyte [1].
- $1,160 — 64GB of desktop DDR5 at the 31 October median, as two retail kits [1].
The trade the ladder hides
The entry machine is a real computer and a fair price for what it holds. It is also the rung with the least headroom, because the memory it has is the memory the reference model needs before anything else runs. The step above buys 24 extra gigabytes and 1,024 extra GPU cores. Memory can be deferred in a way silicon cannot, but not inside these machines: the rung fixes capacity for the life of the laptop.
What could be deferred is the purchase itself. A full-chip machine today, with rented hours for a peak that arrives twice a year, is a smaller commitment than the largest rung bought to cover a peak it would otherwise sit idle through.
What I keep circling is how little of the launch price the memory explains, and how much of the decision it drives. Retail parts set the floor under the top rung at less than half of what the machine sells for, and that floor moves monthly while the price on the sticker does not. A buyer comparing rungs by capacity is comparing a fixed number against a moving one.
The signal to watch is the rung, not the gigabyte. If the next launch keeps 48GB as the full-chip threshold, the cutoff is the product line, and the memory on either side of it is a market price with a daily quote. That is why the hourly price of memory is the rent-or-buy math worth running.
At a glance
- The shift: Nvidia’s 16 October laptop launch prices local capacity on a memory ladder, with the full chip starting at 48GB.
- The number: $2,319 is what 128GB of desktop DDR5 costs at the 31 October median, $294 more than a month earlier.
- The risk: The cheapest rung carries a cut chip, not just less memory.
Related reading
- Price of a GPU-Hour Is Now Set by Memory, Not Compute — more on Major Purchases & Assets
- Two Sparks Cost $3,048 More — more on Major Purchases & Assets
- Your Render Bill Multiplies by 22.97 — more on Major Purchases & Assets
Sources
[1] RamRadar, “RAM Price Index” (data to 2026-10-31, read 2026-11-03): https://ramradar.app/ram-price-index [2] NVIDIA Newsroom, “NVIDIA RTX Spark PCs Available Today — A New Beginning for Windows PCs” (2026-10-16): https://nvidianews.nvidia.com/news/nvidia-rtx-spark-pcs-available-today-a-new-beginning-for-windows-pcs [3] Tom’s Hardware, “Nvidia RTX Spark devices are now officially on sale … starting at $2,599” (2026-10-16): https://www.tomshardware.com/laptops/nvidia-rtx-spark-devices-are-now-officially-on-sale-secure-the-surface-laptop-ultra-or-other-rtx-spark-laptops-here-starting-at-usd2-599-where-to-buy-pricing-and-availability-revealed-on-several-models [4] getdeploying, “GPU Rental Price Trends (2026)” (week of 2026-10-26, read 2026-11-03): https://getdeploying.com/gpu-price-trends