NVIDIA RTX Spark
Earlier today, Jenson Huang (NVIDIA CEO) officially announced the NVIDIA RTX Spark during a keynote address at GTC Taipei, which coincided with the start of Computex 2026.
This is a long-rumoured chip that can deliver 1 petaflop of AI performance with graphics performance comparable to an NVIDIA GeForce RTX 5070 laptop GPU. The key specification is below.
- NVIDIA RTX Spark
- 20 Core Grace CPU
- 6,144 Core Blackwell RTX GPU
- 128GB LPDDR5X Unified Memory
There are already hundreds of articles online covering the details of the announcement; therefore, this article will cover my initial thoughts.
In my opinion, the NVIDIA RTX Spark announcement is both exciting and underwhelming.
It is exciting, as more competition can only be good for the consumer. Until now, the PC laptop market has been dominated by Intel and AMD. In 2024, Qualcomm joined the party, providing the first “real” alternative based on the ARM architecture.
Therefore, having a player the size of NVIDIA focused on chips that can be used by consumers should help drive innovation, hopefully resulting in more interesting product options.
In addition, the ARM architecture has continued to mature. Compatibility is still a challenge, especially with modern games. However, for most users, Windows on ARM is now perfectly usable and even offers benefits over x86.
It is safe to assume that NVIDIA pushing ARM will only further improve the compatibility situation, and Jenson has already committed to ensuring modern games, including kernel-level anti-cheat software, works as designed.
Unfortunately, the announcement was also underwhelming. The NVIDIA RTX Spark is not new, as the leaks indicate it has been ready from a hardware perspective for some time. In addition, the performance (especially the CPU) will likely be underwhelming, performing similarly or slightly below the already released NVIDIA DGX Spark, recognising that it is the same underlying chip with less effective thermals (due to it being in a laptop).
The CPU was developed in partnership with MediaTek and is likely “good enough” for general use, but certainly not going to rival a modern Qualcomm or Apple silicon. In most cases, this might be acceptable, as most software is not inherently CPU-limited. For example, it scores approximately 3,100 in Geekbench 6 single-core tests, compared against an approximate score of 4,200 for the Apple M5.
Finally, the cost of any laptop using the NVIDIA RTX Spark will likely be very high, especially when equipped with larger memory options (up to 128GB). I would not be surprised to see the Microsoft Surface Ultra exceed £6000 for a higher configuration.
Although competition is good, I do not expect the NVIDIA RTX Spark to impact the market broadly with this first release, recognising that the target audience will be very small, specifically those looking for a laptop that can develop and run local AI models.
Overall, I will be watching this space closely. I don’t expect NVIDIA to sell many RTX Spark laptops. However, if there is one company that can afford to start small and scale over time, it is NVIDIA.