Dell and Super Micro team up to build Musk's xAI supercomputer

Elon Musk is not taking chances when it comes to outfitting his AI supercomputer.

Musk confirmed on X that xAI is splitting the server rack contracts down the middle between Dell Technologies and Super Micro Computer. Dell is assembling half the racks for the upcoming system, while Super Micro takes the other half.

Dell and Super Micro team up with Nvidia

Dell CEO Michael Dell quickly backed up the news on X, posting a sneak peek of server racks being prepped and noting that Dell is building an "AI factory" alongside Nvidia to power the next version of Grok.

Super Micro also confirmed the partnership to Reuters. The San Francisco-based server builder is widely known for its tight-knit relationship with Nvidia and its expertise in liquid cooling, which has become a hard requirement for high-density AI clusters.

Musk's compute ambitions are expanding fast. He noted earlier this year that training Grok 2 required around 20,000 Nvidia H100 GPUs. For Grok 3 and future iterations, xAI expects to need 100,000 H100s.

Musk wants the massive system operational by the fall of 2025, according to earlier reporting from The Information. Sightings of Musk in Memphis, Tennessee have also sparked reports that the project could be anchored there.

Why it matters

For Grok users, this timeline sets expectations. Don't look for a massive model jump tomorrow; xAI's real heavyweight leap is aimed squarely at late 2025 once this compute engine comes online.

We think splitting the supply chain right down the center is a shrewd play. Commissioning a cluster of 100,000 GPUs creates brutal logistical bottlenecks. By pitting Dell and Super Micro against each other, Musk gets competitive delivery timelines, combines Dell's global manufacturing footprint with Super Micro's liquid-cooling edge, and avoids relying on a single vendor.

The broader takeaway is that the frontier AI table stakes keep rising. Building competitive frontier models now demands industrial-scale infrastructure planning on the order of 100,000 top-tier accelerators.

That said, assembling the racks is only part of the battle. Powering and cooling an installation of this size presents massive logistical headaches. Keeping 100,000 cards humming without cluster-wide networking hiccups or power constraints looks like the real test xAI must survive by next fall.