Elon Musk: Sending AI Computing Power to Space Next Year
Musk stated in response to a question from a netizen that he is "highly confident" SpaceX will launch the NVIDIA Vera Rubin NVL72 AI computer next year. This is his latest statement following the announcement on August 4 during the SpaceX earnings call that the company will "fully transition to NVIDIA architecture". He referred to the "Vera Rubin architecture as the best architecture" and specifically praised its wireless rack design over traditional rack solutions.
In terms of hardware details, the Vera Rubin NVL72 single rack configuration consists of 36 Vera CPUs and 72 Rubin GPUs, with the entire system code-named "NVL72 VR200", paired with ConnectX-9 super network cards and BlueField-4 DPUs, composed of 18 computing trays and 9 NVLink switching trays. A single rack weighs about 1,800 kilograms and has a power consumption of 120 kilowatts.
The satellite platform's power supply relies on a 210-kilowatt solar array, and cooling is achieved through a liquid cooling system that directly dissipates heat into the vacuum of space via liquid heat exchangers. For the launch vehicle, the Falcon 9 can carry a payload of 22,800 kilograms to low Earth orbit, with a fairing height of 13.1 meters and a diameter of 5.2 meters, allowing for multiple racks to be launched in a single mission.
This project is externally code-named "Starmind AI1", a satellite computing power project jointly designed by SpaceX and NVIDIA. Netizen Sawyer Merritt previously disclosed that each Starmind satellite will be equipped with NVIDIA Rubin GPUs and Vera CPUs for "data center-level" space computing power. In terms of the timeline, the first system is planned to be in orbit by Q4 2027, with "large-scale deployment" expected in 2028.
The capital ties are also close: NVIDIA invested $10 billion in xAI's Series E financing in January this year; in February, SpaceX acquired xAI in an all-stock transaction valued at $1.25 trillion, with NVIDIA's stake in xAI subsequently converted into Class A shares of SpaceX. After SpaceX completed its IPO in June, NVIDIA held approximately 122.8 million shares of SpaceX, which, based on the closing price of $170.86 on June 30, had a market value of about $21 billion, making it NVIDIA's second-largest holding after Intel (with a $30 billion stake), accounting for 33.1% of its disclosed stock portfolio; however, as of August 15, SpaceX's stock price fell back to $140, reducing this holding to about $17.2 billion.
In terms of market mechanisms, NVIDIA's stock price rose after Musk announced the "full transition to NVIDIA", as the market viewed SpaceX's exclusive procurement commitment as a certainty for incremental orders in NVIDIA's data center business. The buying logic is event-driven rather than fundamentals-driven. The beneficiaries are NVIDIA—locking in long-term hardware orders while holding equity in SpaceX; the pressured parties are other GPU manufacturers excluded from SpaceX's supply chain and short sellers skeptical of the timeline: according to external statistics, only 19% of Musk's previous 602 public commitments were fulfilled on time, while 35% were severely delayed or unfulfilled, and this record is being used by the market as a pricing reference for whether the Starmind AI1 launch timeline can be met.
Additional information: xAI previously trained the Grok series models exclusively using NVIDIA Hopper and Blackwell chips, with no credible evidence found for procuring alternative acceleration cards like AMD Instinct, consistent with this "full transition to NVIDIA" statement historically.
Source: Public Information
ABAB AI Insight
The binding between Musk's enterprises and NVIDIA has been well established: xAI has exclusively used NVIDIA Hopper and Blackwell chips to train the Grok series models, never procuring alternatives like AMD Instinct; in January 2026, NVIDIA invested $10 billion in xAI's Series E financing, and just a month later, SpaceX acquired xAI in an all-stock transaction valued at $1.25 trillion, objectively upgrading NVIDIA's equity exposure in xAI to equity exposure in SpaceX—a much larger hardware buyer.
The flow of funds can be fully restored: NVIDIA invested to acquire xAI equity, and after xAI was absorbed by SpaceX, this equity automatically converted into SpaceX stock; after SpaceX's IPO in June, NVIDIA's balance sheet added approximately 122.8 million shares of SpaceX stock, which once valued at $21 billion, becoming its second-largest holding after Intel. This means NVIDIA's identity with SpaceX has shifted from merely being a chip supplier to also being a shareholder, bundling hardware sales and equity investment into the same transaction, motivated by the capital tie to secure an exclusive procurement commitment of "full transition to NVIDIA architecture" to lock in future orders for data center and satellite computing power.
This "investment for orders" strategy is highly similar to NVIDIA's previous operations with AI infrastructure companies like OpenAI and CoreWeave—using equity investment to encourage downstream customers to prioritize capital expenditures on its chips. The "satellite computing power" battlefield is still in the very early stages of industry validation: NVIDIA's previous Space-1 module only deployed dozens of Rubin accelerator cards, and if Starmind AI1 truly scales to 72 GPUs per rack and multiple racks, it will be the most ambitious attempt of its kind, representing the initial stage of "ground data center capacity spilling into orbit".
Essentially, this is a combination of capital concentration and supply chain locking: NVIDIA, by holding shares in the largest downstream buyer, transforms originally dispersed and comparable procurement decisions into exclusive supply agreements solidified by equity relationships, effectively creating a "chip manufacturer holding shares in customers" mechanism, aimed at replacing pure price competition with capital ties during a period of tight GPU supply and increased buyer bargaining power. However, this mechanism also deeply ties NVIDIA's stock price to SpaceX's stock price and performance capability—should the Starmind AI1 launch timeline experience a Musk-style delay, NVIDIA's holding value and hardware order expectations would be pressured simultaneously, creating risk resonance rather than risk hedging.
ABAB News · Cognitive Laws
Investment for orders, suppliers becoming shareholders is the real lock-in.
Fulfillment rate determines valuation, the commitment itself never does.
Those who bundle hardware and equity lock in the next round of demand.