NVIDIA’s CEO Jensen Huang has once again blurred the line between marketing theater and technological history. This time, he personally hand-delivered what he calls the “world’s smallest supercomputer,” the DGX Spark, to none other than Elon Musk – just as it officially hit the retail market. 
The symbolic moment wasn’t just a gesture of camaraderie between two tech titans; it was a statement about where AI computing is heading: smaller, faster, and more personal than ever before.
The DGX Spark, unveiled earlier this year at CES 2025, is the latest evolution in NVIDIA’s AI hardware ecosystem. It’s an astonishingly compact device that houses the new NVIDIA GB10 Grace Blackwell Superchip – capable of delivering up to one petaflop of AI performance at FP4 precision. To put that into perspective, that’s supercomputer-level processing power that can literally fit on your desk. The unit carries 128GB of unified CPU-GPU memory, designed so data scientists and AI developers can train, prototype, and deploy inference models locally without ever touching the cloud. Combined with NVIDIA’s ConnectX networking and NVLink-C2C technology, the Spark boasts five times the PCIe bandwidth of traditional systems, turning what looks like a small workstation into an AI powerhouse.
During the now-viral delivery event at SpaceX, Huang greeted engineers and staff before joining Musk in the cafeteria – where, in a scene oddly human for two of the most powerful people in tech, they shared pizza while reminiscing about the first DGX-1 delivery to OpenAI years ago. Huang reportedly joked that Spark is the spiritual successor to that machine – only this time, it fits under your arm instead of inside a server rack. Musk, meanwhile, was seen grinning as he unboxed the system, calling it “a cute beast.”
The DGX Spark will be available at major retailers starting October 15th, with configurations offered through partners such as Acer, ASUS, Dell, GIGABYTE, HP, Lenovo, and MSI. The device had been slated for a July release, but the complex manufacturing process involving the custom MediaTek-NVIDIA GB10 SoC pushed it back by a few months. Despite the delay, early benchmarks show that Spark’s performance-to-size ratio is unmatched in its class – though not without controversy. Priced at around $3,999, it’s a premium piece of hardware, and reactions have been polarized.
Supporters hail it as a marvel of miniaturization and a meaningful step toward democratizing AI power. Critics, however, are less charitable. Online debates compare its performance unfavorably to AMD’s 395 series miniPCs and suggest the MediaTek-based architecture isn’t quite living up to the hype. Some claim it’s slower in inference tasks, while others argue the closed NVIDIA OS limits flexibility. Yet, even detractors admit that as a design object, the DGX Spark is iconic – a small, angular block that could sit as easily in a lab as on a CEO’s desk.
Whether you see it as the future of edge AI or just another luxury tech toy for the elite, one thing’s certain: the Spark represents NVIDIA’s next big gamble. By putting a supercomputer in the hands of individuals like Elon Musk, Huang is signaling that the AI revolution won’t stay confined to data centers. It’s moving into homes, startups, and personal labs – one glowing cube at a time.
3 comments
bro it’s basically Strix Halo but on ARM and locked down to Nvidia OS. useless
he literally hand delivered it cuz no one else was buying that slow box 🤣
lol imagine paying 4 grand for a mini toaster with a Mediatek chip 😂