Tesla launches own AI chip factory

Elon Musk says Tesla (TSLA) could start building a new Tesla AI chip factory as soon as next week.

If that happens, it would mark a major shift in who controls the technology that powers autonomous driving.

For years, Tesla has designed its own AI chips but depended on outside manufacturers to build them. Companies like TSMC and Samsung run the massive semiconductor plants that actually produce those processors. Now Musk is hinting Tesla may bring that step in house.

The project is called Terafab. The idea is to manufacture Tesla’s next generation AI processors, including the AI5 chip that will power future self driving systems.

These chips act as the brain of autonomous software. They process camera feeds, radar signals, and other sensor data fast enough for a vehicle to make driving decisions in real time.

If Tesla succeeds, the implications are huge. Automakers have always been buyers of advanced chips. Tesla would be trying to become a producer, giving it tighter control over supply at a time when AI processors are one of the most scarce resources in technology.

The disruption behind the news: Tesla is trying to control the most expensive layer of the autonomy stack.

Today the global AI chip market is controlled by a small group of companies.

Advanced semiconductor fabs cost roughly $20B to $40B to build and only a few operators in the world can run them efficiently.

If Tesla seriously attempts Terafab, it becomes the first car company in modern history trying to vertically integrate semiconductor manufacturing at scale. Vertical integration means a company controls more steps of the production process itself instead of relying on outside suppliers.

Why does that matter? Because autonomy isn’t limited by software, but by compute.

Compute refers to the processing power needed to run artificial intelligence systems. A fully autonomous vehicle fleet requires massive computing power. Each car runs AI inference locally, which means the vehicle’s onboard chip processes sensor data and makes driving decisions instantly.

If Tesla wants millions of robotaxis on the road, it needs tens of millions of AI chips. That’s where the math breaks existing supply chains.

TSMC and Samsung already have large backlogs from AI data center demand driven by companies like Nvidia. AI accelerator production is often booked years in advance. Tesla waiting in line doesn’t work if the company plans to deploy millions of self driving vehicles.

Building its own fab would give Tesla three advantages.

First cost control. Internal chips could reduce per vehicle compute costs by hundreds of dollars.

Second supply certainty. Tesla wouldn’t be competing with every AI startup and cloud provider for wafer capacity. Wafer capacity refers to how many silicon wafers, the circular base material chips are built on, a fab can process each month.

Third speed. Tesla could iterate AI silicon alongside its autonomy software rather than waiting for external production cycles.

That’s how the company could compress development loops from years to months.

If Tesla produces about 10M AI chips per year and saves even $300 per vehicle by internalizing silicon, that’s $3B in annual margin created purely from chip economics. At that scale, the fab stops looking like a risky manufacturing bet and starts looking like a margin engine that subsidizes the robotaxi rollout itself.

If Terafab works, Tesla stops behaving like an automaker and starts behaving like a vertically integrated AI infrastructure company that also sells cars.

What to watch next

The first signal will be location.

Chip fabs require massive electricity supply, large amounts of water for cooling and cleaning wafers, and government incentives such as tax breaks or subsidies.

Where Tesla builds Terafab reveals who is funding the AI arms race.

Watch for three numbers over the next two years.

First fab capacity. A meaningful facility would target at least 50,000 wafer starts per month. That means the factory begins processing 50,000 silicon wafers every month.

Second chip volume. A global robotaxi fleet could require 5M to 10M AI chips annually.

Third capex. Capex means capital expenditures, or the money spent to build long term assets like factories. A serious fab likely costs more than $25B.

Also watch the supplier reaction. If Tesla internalizes chip manufacturing, it threatens existing semiconductor alliances across the auto industry.

Every automaker today relies on outside chip suppliers. Tesla wants to break that model by building its own.

If Tesla controls both the autonomy software and the silicon running it, rivals could end up licensing Tesla’s tech instead of building their own.

That’s the strategic endgame. For more than a century, the auto industry outsourced many of its most important components. Tesla’s betting the future belongs to companies that build the critical pieces themselves.

There’s another implication too. If Tesla builds a chip fab, it wouldn’t have to use it only for cars. A Tesla AI chip factory producing 50,000 wafers a month could also make tens of thousands of AI accelerator chips for data centers. Those chips train and run large AI models.

That opens a second business selling AI compute. The same factory could power robotaxis while also competing in the cloud AI market. This is a strategy that fits Musk’s broader push to build interconnected technology platforms, much like his effort to turn X into an everything app with X Money.