•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Leaked excerpts from SpaceX’s confidential S-1 filing, related to a planned raise of up to $1.75 trillion, indicate the company is considering internal development and production of certain processors to meet its own needs. The chips referenced are described as GPUs, rather than the ASICs typically used to accelerate AI workloads.
SpaceX’s S-1 is the document required to be filed with the US Securities and Exchange Commission to disclose risks and financial condition ahead of an initial public offering. Reuters, which accessed the document, reports that self-producing GPUs is listed as one of the reasons investment costs are expected to rise in the near term.
SpaceX has not issued an official comment, and the scale of the investment tied to the GPU plan remains unclear.
The GPU discussion appears shortly after Elon Musk’s statement that SpaceX will use Intel’s 14A process in the TeraFab chip manufacturing joint venture, with SpaceX operating the plant. Musk previously confirmed that SpaceX would build and operate large-scale semiconductor fabrication facilities to produce chips designed by Tesla. However, the move toward self-developing GPUs appears to be a new direction.
In the AI hardware race, major players are pursuing different chip paths. Nvidia dominates the GPU market with versatile chips designed to handle large data volumes in parallel. Alphabet’s Google is developing TPU, specialized processors optimized for tasks such as AI model training or operating chatbots.
Reuters also notes that naming conventions for AI accelerators vary by company. AMD and Nvidia use the term GPUs; Google uses TPU; Microsoft refers to its Maia accelerator; and SambaNova Systems calls its AI chip an RDU. Many large cloud providers and independent hardware firms still classify AI accelerators as ASICs, meaning chips designed for specific applications.
Reuters says SpaceX’s omission of any Tesla AI ASICs suggests the company may be designing and producing a processor that does not align with Tesla’s current AI chips. The report also highlights uncertainty around what SpaceX—or its xAI unit—can build in terms of highly customized chips intended to compete with GPU and AI accelerators from companies such as AMD, Nvidia, and SambaNova.
Elon Musk has previously described Tesla’s AI processor, AI5, as a GPU even though it lacks traditional hardware blocks for graphics processing. The inconsistency, Reuters reports, reflects how these chips are built for highly specific workloads, making it difficult to determine whether SpaceX intends to produce Tesla’s AI5/AI6 or instead develop a wholly new GPU line.
GPU production is described as one of the most complex tasks in modern industry. Even Nvidia, the leading GPU designer, does not manufacture chips directly and instead outsources production to TSMC. TSMC has spent decades and billions of dollars refining advanced fabrication processes that require exotic materials and thousands of steps with atomic-level precision.
The semiconductor value chain is segmented across multiple stages, including design, fabrication, packaging, and testing, with different firms involved. Musk’s stated goal for Terafab is vertical integration, controlling the process from design through production.
Premium gym chains are entering a “golden era” that is ending or already in decline, as rising operating costs collide with shifting consumer preferences toward more flexible, community-based ways to exercise. Long-term memberships are shrinking, margins are pressured by higher rents and facility expenses, and competition from smaller, more personalized…
