Elon Musk: Tesla and SpaceX AI will keep buying Nvidia chips in bulk despite AI5

Elon Musk said that SpaceX AI and Tesla do not plan to give up Nvidia, even though they are actively developing their own AI chips. The comment appeared on March 19, 2026, amid discussion of Tesla's wish to produce new processors at Samsung's U.S. facilities.

Why Musk clarified this

The market began reading news about Tesla's in-house chips as a signal that the company was preparing to gradually move away from external accelerator suppliers. So Musk wrote separately that both Tesla and SpaceX AI will continue to buy Nvidia solutions in large volumes.

This is an important caveat: the Musk companies' in-house silicon is advancing quickly, but the need for compute is already so large that internal development alone will not cover it in the foreseeable future.

The reason for the clarification is easy to see. Investors and the industry are closely watching Tesla's plans to place production of its AI chips at Samsung plants in the U.S., and such news is easily read as a push for independence from Nvidia. But Musk's words point to a different logic: he sees in-house production not as an immediate replacement for current supplies, but as an additional layer of compute strategy.

It is also telling that he publicly used the name SpaceX AI for the enlarged structure formed after the merger of SpaceX and xAI for the first time.

What Tesla is betting on

Musk separately explained that the Tesla AI5 chip is designed not only as a data-center component. It can indeed be used to train large language models, but the main use case is different: edge AI computing directly in devices and vehicles, where latency, energy efficiency and cost are critical.

In other words, Tesla is developing the processor not for an abstract chip race but for specific products it needs to scale quickly, including:

• Optimus humanoid robots
• Tesla robotaxis
• FSD autonomous driving systems
• Some internal AI workloads in data centers

This shows that Tesla is building not just an alternative to Nvidia but a vertically integrated platform for its own needs. Where Nvidia dominates as a general-purpose accelerator supplier, Tesla is trying to design a chip for specific scenarios in which it controls the hardware, the software and the end product. That approach can yield savings in cost and efficiency, but it does not settle the question of production scale and capacity availability today.

Why Nvidia is still needed

Even with their own processor designs, Musk's companies still need huge volumes of external chips. The reason is simple: demand for model training, autopilot, robotics and related services is growing faster than new production can be brought online. So Nvidia remains not a temporary stopgap but a critically important partner for the near term.

In this model, in-house chips handle some specialized tasks, while off-the-shelf Nvidia accelerators keep current projects moving and let the companies expand compute clusters quickly.

Musk also reminded followers that he plans to show a major update to FSD, Tesla's autopilot software stack, in the coming weeks. He also promised to reveal more details soon about the project for a giant fab producing AI chips for Tesla and related companies. He had said earlier that his ecosystem's needs are so large that current contract component manufacturers will not be able to fully meet demand. In short, the strategy is twofold: buy from Nvidia in bulk while accelerating the build-out of an in-house manufacturing loop.

What this means

For the market, this is a signal that the era of "either our own chips or Nvidia" has not arrived yet. Large AI companies will combine both approaches: building their own silicon for key products while buying off-the-shelf accelerators in parallel, so growth does not stall.