Key Highlights

  • Three deployment waves totaling 660,000 GB300 chips are planned for Colossus 2 by year end, with the first 220,000-chip wave going live within days and the final wave arriving in late December only if timelines hold, Musk said this week.
  • Colossus 2 already runs 110,000 GB200 chips and 440,000 GB300 chips, stacked on top of Colossus 1's 150,000 H100, 50,000 H200 and 30,000 GB200 units, forming one of the largest concentrated GPU fleets ever assembled by a single operator.
  • Nvidia is not merely a vendor in this relationship. The chipmaker holds an estimated USD 21 billion Equity stake in SpaceX and roughly USD 30 billion in Intel (NASDAQ:INTC) shares, tying its Balance Sheet directly to the customers buying its silicon.
  • SPCX shares slipped about 3 percent for the week and sit roughly 8 percent lower over three months even as chip orders accelerated, with retail sentiment turning bearish and online message Volume jumping 74 percent in a single week.

Three Waves, One Enormous Bet

Musk's roadmap for Colossus 2, the Data Center campus housing his AI computing ambitions, calls for three sequential waves of Nvidia's GB300 chips, each adding roughly 220,000 units. The first wave is expected to go operational within days, the second in November, and the third potentially by late December, a timeline Musk himself qualified as achievable only if we get lucky. The reason chips arrive in 220,000-unit increments rather than smoothly, Musk explained, comes down to a hardware constraint: the number of fiber optic cables that can physically be plugged into a central network switch. Layered on top of Colossus 1's existing 150,000 H100, 50,000 H200 and 30,000 GB200 chips, and Colossus 2's current 110,000 GB200 and 440,000 GB300 units, the completed buildout would push the combined GPU count past 1.2 million. Independent analyst estimates peg the hardware value of both sites at more than USD 100 billion, representing over two gigawatts of compute capacity, a scale that rivals the power draw of a mid-size city and underscores how aggressively Musk is racing to keep pace with rivals Training ever-larger AI models.

Why Nvidia Needs SpaceX to Win

The chip order alone would make Colossus 2 one of Nvidia's largest single deployments anywhere, but the relationship runs deeper than a purchase order. Nvidia's roughly USD 21 billion stake in SpaceX, disclosed alongside its approximately USD 30 billion position in Intel, means the chipmaker has effectively become an investor in its own demand. That structure has drawn scrutiny from analysts who worry about circular financing arrangements across the AI Supply chain, where chip suppliers fund the very buyers purchasing their products. For Nvidia, whose quarterly data center Revenue reached roughly USD 89 billion and grew 117 percent year over year, a customer capable of absorbing hundreds of thousands of GB300 units in a single year is strategically significant, even if SpaceX and its affiliated ventures never appear as a named customer in Nvidia's public disclosures. The arrangement also diversifies Nvidia's exposure beyond traditional hyperscalers such as Microsoft, Amazon and Google, giving it a foothold in Musk's parallel AI infrastructure buildout that spans xAI, Tesla and now orbital computing ambitions tied to SpaceX's launch capacity.

The Skeptics and the Stock

Not every investor is convinced the buildout translates into durable returns. SPCX shares fell roughly 3 percent over the past week and are down about 8 percent across the last three months, even as headlines about chip waves multiplied. Retail sentiment tracked across social platforms flipped from bullish to bearish, with message volume surging 74 percent as traders debated whether the Capital intensity of hosting over a million GPUs, including power procurement, cooling infrastructure and network switching limits, could be monetized fast enough to justify the spend. Skeptics point to execution risk: Musk's own admission that the third wave depends on getting lucky suggests supply chain bottlenecks, whether in chip packaging, power interconnects or fiber optic switching gear, remain real constraints. Bulls counter that even partial delivery of 660,000 additional chips would cement Colossus as among the largest AI training clusters in existence, and that Nvidia's willingness to keep prioritizing shipments to Musk's ventures signals confidence the capacity will find paying workloads, whether from xAI's Grok models or external compute customers.

Where Things Stand

Nvidia has never disclosed SpaceX or its affiliated Musk ventures as a top reported customer, so the mega-customer label remains inferential rather than confirmed. But the math is hard to ignore: 660,000 additional GB300 chips on top of an existing fleet approaching a million units, financed in part by Nvidia's own equity stake in SpaceX, points to one of the tightest supplier-customer-investor loops in the AI buildout. Whether that concentration becomes a strength or a Liability likely hinges on execution through year end, and on how quickly Musk's ventures can turn raw compute into revenue-generating AI workloads rather than idle capacity.