Musk: Tesla Not Replacing Nvidia Despite Chip Development

Tesla’s forthcoming AI5 chip will be manufactured by Samsung in Texas and TSMC in Arizona, CEO Elon Musk announced on Wednesday’s Q3 earnings call. This move signifies Tesla’s continued push toward vertical integration, bringing chip manufacturing closer to its U.S.-based operations.

Musk detailed Tesla’s ambition to achieve “excess production” of the AI5 chip, with any surplus units being utilized within the company’s expansive data centers. This highlights Tesla’s dual strategy: supporting its growing autonomous driving capabilities while simultaneously expanding its in-house computing infrastructure.

“Our explicit goal is to have an oversupply of AI5 chips,” Musk stated, underscoring the strategic importance of these chips in Tesla’s future.

The transition to proprietary chips began in 2019 when Tesla opted to replace Nvidia Drive chips in its vehicles with its own processors. However, Musk clarified that Tesla will continue to leverage Nvidia’s powerful GPUs, particularly for training its complex AI models.

“We’re not about to replace Nvidia, to be clear, but we do use both in combination,” Musk explained. Tesla currently boasts computing capacity equivalent to 81,000 Nvidia H100 chips, demonstrating its significant investment in AI infrastructure.

This announcement comes after the departure of Pete Bannon, a former Apple engineer who spearheaded chip design and the development of Dojo, Tesla’s supercomputer dedicated to enhancing driverless technology earlier this year. Bannon’s departure has left some speculation about Tesla’s AI roadmap. 

First unveiled in 2024, the AI5 chip represents the latest iteration of Tesla’s Autopilot hardware, designed to process the vast amounts of data required for self-driving functionalities. Samsung’s $16.5 billion chip contract, confirmed by Musk as being with Tesla, underscores the magnitude of the project and the strategic partnership between the two tech giants.

While the initial plan involved TSMC manufacturing the AI5 chip and Samsung producing its successor, the AI6, Musk clarified that both TSMC and Samsung will manufacture the AI5 at their respective U.S. facilities. He further noted that the AI5 is designed as a “half reticle” chip, suggesting a smaller silicon footprint compared to the “full reticle” designs favored by Nvidia and Advanced Micro Devices (AMD). This potentially allows for greater manufacturing efficiency and lower costs.

Tesla’s foray into custom silicon aligns with a broader trend among hyperscalers like Google, Amazon, and Microsoft, all of whom are actively investing in developing their own AI chips as compelling alternatives to Nvidia’s dominant market position. The rationale behind this strategic shift lies in the potential for customized chips to offer superior cost-efficiency and enhanced performance in specific AI tasks.

Currently, Apple stands as the only other company besides Tesla that designs and utilizes its own silicon for both its consumer products and data centers supporting AI services. This level of vertical integration provides these companies with greater control over their hardware and software ecosystems, facilitating optimized performance and potentially unlocking new competitive advantages.

Musk emphasized that Tesla’s exclusive focus on in-house requirements for the AI5 chip has enabled its design team to streamline the architecture, removing legacy components that previously hindered performance.

“Tesla only has to satisfy requirements from one customer,” Musk stated. “That that makes the design job radically easier and means we can delete a lot of complexity from the chip.”

According to Musk, the removal of legacy GPUs, signal processors, and other outdated components has resulted in a chip with potentially superior performance-per-dollar for AI – potentially by a factor of 10. This bold claim positions the AI5 as a disruptive force in the AI chip market, challenging the established dominance of industry giants.

Musk acknowledged Nvidia’s accomplishments in navigating diverse and complex requirements but highlighted Tesla’s pursuit of “radical simplicity” in its chip design.

Notably, Musk’s AI startup, xAI, which maintains close business ties with Tesla, has emerged as a significant Nvidia customer. xAI is currently engaged in constructing a large-scale supercomputing facility in Memphis, Tennessee, its second such facility in the region, utilizing high-end Nvidia Grace Blackwell chips. This indicates a continued reliance on Nvidia for certain aspects of AI development, even as Tesla pursues its own custom silicon strategy.