Nvidia

After 9 Years, Jensen Huang Personally Delivers to Musk — The Long-Delayed AI Personal Supercomputer Finally Arrives

Honghao Wang

— 4 min read
After 9 Years, Jensen Huang Personally Delivers to Musk — The Long-Delayed AI Personal Supercomputer Finally Arrives

Starship’s 11th Flight — and an Unexpected Guest

Today marked Starship’s 11th flight, a complete success — with a surprise appearance by Jensen Huang himself.

The NVIDIA CEO personally flew to Starbase, Texas, standing beside the towering Starship to hand Elon Musk a computing “nuclear bomb” — the brand‑new NVIDIA DGX Spark personal AI supercomputer.

image

---

A Flashback to 2016

For veteran tech fans, this scene sparked nostalgia:

  • In 2016, Musk was still a co‑founder of OpenAI, working closely with Sam Altman.
  • Huang personally delivered the world’s first DGX‑1 supercomputer to OpenAI’s fledgling office.

image

Huang joked at the time:

> If this ends up being the only shipped product, then the project cost will have been a whopping $2 billion.

That $2‑billion beast played a key role in sparking the large model era.

By 2017, Google’s Transformer architecture inspired OpenAI’s first GPT model — powered by NVIDIA supercomputers.

---

Fast Forward Nine Years

  • Musk is now a regular atop the world’s richest list.
  • Huang has led NVIDIA to become (at times) the world’s most valuable company.

This time, the DGX isn’t a massive data‑center rack — it’s a performance monster for your desk. Its arrival loudly signals: AI supercomputing is now personal.

image

Spoiler: APPSO’s DGX Spark review unit is coming soon — hands‑on coverage ahead!

---

The Long Wait for DGX Spark

Delivering DGX Spark to Musk wasn’t easy.

  • Initially shown at CES in January as “Project Digits.”
  • Planned for May and then summer release — neither happened.
  • No units shipped until now.

Many developers feared cancellation. Industry rumors pointed to manufacturing delays in the Grace Blackwell GB10’s Grace CPU, co‑developed with MediaTek.

The GPU was ready; the CPU was not. A rare stall for NVIDIA.

---

Competition & Pricing

With rivals like Apple’s M3 Ultra Mac Studio boasting higher memory bandwidth, DGX Spark arrives:

  • Later than expected
  • $1,000 more expensive than early rumors

Is it worth the wait? NVIDIA thinks so — and here’s why: unique design philosophy, targeted for AI developers.

---

DGX Spark: Key Technical Highlights

image

1. Grace Blackwell GB10 Superchip

  • 20‑core ARM Grace CPU
  • Powerful Blackwell GPU
  • Up to 1 petaflop AI computing power

Data‑center performance from your desk.

2. Unified 128 GB Memory

  • CPU and GPU linked via NVLink™‑C2C
  • 5× bandwidth of PCIe Gen 5
  • 273 GB/s memory bandwidth (less than Mac Studio M3 Ultra, but NVIDIA prioritizes total capacity)

> Benefit: Smoothly run 200‑billion‑parameter models locally, avoiding complex model partitioning.

image

3. Advanced GPU Architecture

  • 5th‑gen Tensor Cores
  • Supports FP4 & FP8 ultra‑low precision
  • 5× FP8 performance vs previous generation

Turbo mode for AI workloads — faster inference, higher energy efficiency.

4. Scalable Clustering

  • Built‑in ConnectX®‑7 200 Gb/s NIC
  • Link two DGX Sparks → mini‑cluster with 256 GB shared memory
  • Handles models up to 400 billion parameters

---

Software Ecosystem Moat

DGX Spark ships with:

  • NVIDIA AI Software Stack (CUDA, TensorRT, NVIDIA NIM™ microservices)
  • Custom DGXOS (Ubuntu‑based)

Ready to develop, out of the box. No compatibility hassles — ultimate time‑saver for professionals.

---

Availability & Price

image
  • Launch date: October 15
  • Price: ~$4,000
  • Partners: Acer, ASUS, Dell, Lenovo
  • OS: DGXOS only — no Windows/macOS

Positioning: Local AI powerhouse for developers & power users, not mainstream creative pros.

---

Verdict: A Focused Powerhouse

For $4,000, you get:

  • Local handling of 200‑billion‑parameter models
  • Full CUDA ecosystem support
  • Tight integration of hardware & software

For sensitive workloads, maximum control, and top‑tier AI performance — DGX Spark is competitively priced and purpose‑built.

---

Complementary Tools for AI Creators

Platforms like AiToEarn let creators harness such computing power to:

  • Generate, publish, and monetize AI‑driven content
  • Post simultaneously to Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, and X (Twitter)
  • Access analytics and AI model rankings (AI模型排名)

When paired with DGX Spark, AiToEarn offers a complete high‑performance AI workflow — from computation to publication.

---

💬 What’s your take on DGX Spark’s delayed yet powerful debut?

Would you invest in one, or wait for a second‑gen model? Share in the comments.

Read more