Space Exploration Technologies Corp. (NASDAQ: SPCX) CEO Elon Musk stated that the company’s orbital data center satellites will cost far less than Wall Street analysts predict. Musk made this claim ahead of Starship’s maiden orbital launch, challenging recent cost models for gigawatt-scale orbital compute.

Cost projections diverge from Wall Street estimates

Investment firm Mach33 CEO Aaron Burnett and other analysts have placed the cost of 1 GW of SpaceX orbital compute between $160 billion and $180 billion. Bank of America estimates this figure at approximately $165 billion, while Wood Mackenzie cites $170 billion. BNP Paribas noted that launch costs alone exceed $100 billion.

In contrast, user Jo Bhakdi presented a model calculating the total cost at $68 billion, falling to $43 billion by 2035. Responding to this disparity, Musk asserted on social media platform X that the actual costs would be "way better" than Bhakdi’s figures by 2028.

Starlink positioned as global internet backbone

Musk also reiterated his view that Starlink satellite internet service could eventually carry a majority of Earth’s IP traffic. In response to forecasts regarding Starlink V3 constellation deployments by 2028, he stated it is "increasingly probable" that Starlink will dominate long-term traffic.

Musk argued that once Starlink carries the majority of IP traffic, it would effectively become the internet itself, with other networks connecting to it rather than the reverse. This vision aligns with previous statements where Musk described Starlink as potentially more reliable than cable infrastructure.

Launch timeline and market reaction

The comments come as SpaceX prepares for the first orbital launch of its Starship rocket. The launch, initially scheduled for September 22, was delayed and is now set for September 28. The mission is expected to carry Starlink V3 satellites.

Following these announcements, SpaceX shares traded up 0.28% to $149.09 during overnight trading sessions.

What the numbers show

The divergence between Wall Street’s $160-$180 billion estimate and Bhakdi’s $68 billion model highlights a significant uncertainty in pricing emerging space-based infrastructure. While BNP Paribas attributes over $100 billion to launch costs alone, Bhakdi’s total all-in figure is lower than that single component, suggesting a fundamental disagreement on launch economics or hardware efficiency assumptions.

Elon Musk stated that Space Exploration Technologies Corp (NASDAQ: SPCX) is on track to develop an artificial intelligence model comparable to GPT-6 within two to three months. The CEO expressed cautious optimism regarding the timeline, noting that rapid deployment of computing infrastructure remains a key competitive advantage.

Musk emphasized that if the company's growth rate, or second derivative, remains strong, SpaceX could reach a leading position in the AI sector within six months. He highlighted that while competitors like Anthropic and OpenAI have been operating for six and ten years respectively, SpaceX's AI efforts are only three years old. The executive argued that beyond a certain intelligence threshold, additional capability offers diminishing returns for specific tasks, using the analogy of a "Newton-level" intellect in a toaster.

Massive GPU expansion underway

To support these ambitions, SpaceX is aggressively scaling its hardware infrastructure through its subsidiary, now referred to as SpaceXAI. The company’s Colossus superclusters currently house approximately 670,000 GPUs. Musk detailed the specific composition of these clusters and outlined immediate expansion plans to add hundreds of thousands of new units over the next few months.

The expansion plan includes bringing another 220,000 GB300 GPUs online next week, followed by an additional 220,000 units in November. Musk noted that if successful, yet another 220,000 GB300 GPUs could be operational by late December, significantly boosting total compute capacity.

Strategic shifts and analyst views

SpaceX’s integration of AI capabilities has drawn attention from financial analysts. JPMorgan analyst Doug Anmuth noted that the company’s narrative is evolving beyond rockets and Starlink. The acquisition of the AI coding platform Cursor was identified as a strategic move to strengthen enterprise AI distribution. JPMorgan estimates Cursor generates about $4 billion in annual recurring revenue, with roughly 75% derived from business customers. This provides SpaceX with valuable coding data to enhance its Grok model.

Morgan Stanley analysts observed that SpaceX and Tesla Inc (NASDAQ: TSLA) are increasingly sharing technology, talent, and infrastructure to pursue opportunities in physical AI. The firm suggested that this growing overlap could eventually support a potential merger as investors look beyond traditional digital models.

What the numbers show

The disclosed GPU figures reveal a distinct strategic pivot toward next-generation hardware. Of the current 670,000 GPUs in the Colossus clusters, 550,000 are already GB200 or GB300 models, indicating that legacy H100 and H200 chips constitute less than half of the existing fleet. Furthermore, the planned addition of 660,000 GB300 GPUs by late December would nearly double the total cluster size, with the new additions being exclusively high-end GB300 units. This suggests a concentrated effort to maximize compute density and efficiency rather than simply increasing volume with older architecture.

Market data indicates that SPCX stock closed 0.22% lower at $148.03 on Thursday. On a year-to-date basis, the shares have declined 8.03%.