ASUS Opens ProArt RTX Spark Pre-Orders From $2,599.99

Fremont, California-based ASUS announced on 7 October 2026 that global pre-orders have opened for its ProArt P16 (H7607BA) and ProArt P14 (H7407BA) laptops, the first ProArt machines built on the NVIDIA RTX Spark Superchip, starting at $2,599.99. Both systems pair a Blackwell RTX GPU with an NVIDIA Grace CPU and up to 128GB of unified memory, and the company claims up to 1 petaflop of AI performance. Pre-orders run through the ASUS Store and Best Buy, with a compact ProArt GR1X mini PC following later this year.

Context — why the AI PC category is moving to unified memory

The launch lands as PC makers compete to keep larger language and diffusion models running on the machine rather than in a rented cloud. ASUS frames the pitch around three claimed advantages: predictable computing cost instead of per-token billing, privacy for prompts and source files, and control over which models a creator runs.

The company cites an illustrative total-cost-of-ownership case in which daily agent workloads on cloud services could exceed the price of a ProArt RTX Spark machine within five to ten months. That comparison assumes a $6,000 configuration, 36-month amortisation, roughly $2 per month of electricity at 80 watts, and published list prices per million output tokens from two named AI providers checked in September 2026.

ASUS also ties the hardware to an agent layer. It has partnered with ComfyUI, the open-source visual workflow platform, and with Nous Research for its Hermes Agent, so that plain-language instructions can drive local image and video pipelines. Both partners supplied quotes in the announcement, with Nous Research chief executive Dillon Rolnick saying the goal is intelligence users can own rather than rent, and Comfy co-founder Yoland Yan pointing to memory capacity as the constraint that local workflows previously hit.

The timing matters because the Windows installed base has been the slowest of the major platforms to host capable on-device generative models. If ASUS can sell memory capacity as the differentiator, the argument shifts from raw clock speed to how many parameters a buyer can hold resident.

Data — what the configurations and prices show

ASUS published five P16 SKUs and four P14 SKUs. The P16 spans 24GB of RAM with a 512GB SSD and a 4K OLED panel at the entry point, through 32GB and 64GB tiers, up to a 128GB / 2TB / 4K configuration. The P14 spans 24GB and 32GB with 512GB of storage and a 3K panel, then 48GB and 128GB with 1TB.

The price ladder runs from $2,599.99 for the P14 to a $2,799.99 starting figure for the P16, with the 128GB P16 carrying the $6,000 figure used in the company's cost comparison. Both machines use LPDDR5X memory soldered on board, which means buyers cannot add capacity after purchase — the memory tier chosen at checkout is the ceiling for model size.

Model	Memory	Storage	Display	Starting MSRP
ProArt P14	24GB–128GB	512GB–1TB	3K OLED	$2,599.99
ProArt P16	24GB–128GB	512GB–2TB	3K / 4K OLED	$2,799.99

The P16 weighs 3.9lbs and measures 0.63 inches thick; the P14 weighs 3.26lbs at the same thickness. Batteries are rated at 99.9Wh and 90Wh respectively. Both ship with Windows 11 Home or Pro depending on SKU, and both use a 140W USB-C adapter.

Analysis — where the margin and the risk sit

The strategic bet is that memory, not compute, becomes the binding constraint on local AI. Soldered LPDDR5X at 128GB is an expensive bill of materials, which is why the top P16 SKU sits far above the entry price. That spread lets ASUS capture margin from buyers who need headroom rather than from the volume tier.

For NVIDIA, the launch extends the CUDA software ecosystem into thin creator laptops, a segment where the company has less presence than in gaming. If local agent workflows become a standard expectation, the addressable base for high-memory mobile parts widens beyond workstation buyers.

The counter-argument is that cloud pricing has fallen repeatedly, and the report's own cost case depends on list prices that promotional rates can undercut. The comparison also assumes heavy daily utilisation; a creator running occasional generations will not amortise a $6,000 machine in ten months. ASUS acknowledges that cloud services remain useful for frontier-scale models and burst capacity, and positions the hardware as one option in a local, cloud or hybrid mix.

Positioning is split. Buyers with steady, privacy-sensitive workloads are the natural long side of this trade, while cost-sensitive hobbyists stay on metered cloud access. The open question is whether ComfyUI and Hermes adoption is broad enough to pull the mid-tier SKUs rather than only the 128GB halo configuration.

Outlook — what to watch next

Three things determine whether this becomes a category or a niche. First, retail availability: the P16 64GB and 128GB SKUs are listed as coming soon at Best Buy, and Micro Center shows coming soon across both lineups, so the memory-heavy configurations are not yet broadly stocked.

Second, the ProArt GR1X mini PC, which ASUS says arrives later this year in 64GB and 128GB configurations through select premium retailers. That always-on form factor targets studios and small teams rather than individuals, and its pricing has not been disclosed.

Third, software depth. ComfyUI's community pipeline additions and Hermes' skill set will decide whether buyers keep the machines busy. ASUS did not disclose the commercial terms of either partnership, nor unit volume targets, and did not state whether the Google AI Pro bundle applies to every SKU.

Frequently Asked Questions

What does the ASUS ProArt RTX Spark launch mean for retail investors?

It signals that PC makers are pricing memory capacity as a premium feature rather than a commodity spec. The $2,599.99 floor and the far higher 128GB configuration create a wide margin band for ASUS, while NVIDIA gains another channel for its CUDA platform. Neither company disclosed unit volume guidance, so the revenue effect is not quantifiable from the announcement alone.

Can the ProArt P16 and P14 run large AI models without an internet connection?

ASUS says the systems are engineered to run larger AI models locally without cloud dependency, using up to 128GB of unified memory. Whether a specific model fits depends on architecture, quantisation, context length and available system memory, which the company lists as the governing variables. Workflows can remain available offline, and cloud services can still be used for frontier-scale models or burst capacity.

Why does the announcement compare laptop cost to cloud token pricing?

ASUS argues that owning local capacity converts variable pay-per-use spending into predictable hardware cost. Its illustrative case assumes heavy daily agent workloads, a $6,000 configuration, 36-month amortisation, roughly $2 per month of electricity and published list token prices checked in September 2026. The company notes actual results vary by model, utilisation and caching, and promotional cloud rates can change the outcome.

Bottom Line

ASUS is betting that soldered 128GB memory, not raw compute, is what convinces creators to buy local AI capacity instead of renting it.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. CFD trading carries high risk of capital loss.