Elon Musk says Samsung will help manufacture Tesla’s next-generation self-driving chips

AI5 chip manufacturing and suppliers

Elon Musk says AI5 will use both TSMC and Samsung as foundry partners. He added that initial production will start at TSMC in Taiwan before moving to Arizona; Samsung’s Fab 2 in Taylor, Texas, is slated to come online in 2026.

The new Samsung fab is scheduled to be operational in 2026. This split is significant as TSMC reportedly plans to increase its advanced node prices by 5% to 10% starting in 2026.

AI5 chip performance and design

Musk claims AI5 could be 40× better than AI4 by some metrics. A 2,000–2,500 TOPS target has been reported by the media, but Tesla hasn’t published TOPS specs.

This increase in raw compute power is crucial for enabling more complex neural networks in vehicles. The AI5 chip has been designed as a “half-reticle” size, making it smaller than many other large AI chips. Musk has emphasized efficiency and cost per compute for AI5, but Tesla hasn’t provided audited $/performance comparisons.

Hardware 3 vs. hardware 4 specifications

The upgrade from HW3 to HW4 increased the number of CPU cores from 12 to 20, with the CPU frequency rising from 2.2 GHz to 2.35 GHz. Community teardowns describe HW3 as using custom accelerators around tens of TOPS per die, and HW4 as a modest step-up; exact system TOPS vary by method, so treat public ‘36 vs 50 TOPS’ as third-party estimates.

The camera system moved from 1.2-megapixel units in HW3 to 5-megapixel cameras in HW4. The new hardware can support up to 12 cameras total, compared to 9 in HW3. W4 appears to reintroduce a 76–77 GHz radar module (‘Phoenix’ in filings), but the supplier/model hasn’t been confirmed publicly.

FSD (supervised) safety data

Tesla vehicles using Autopilot or FSD (Supervised) were involved in one crash for every 6.69 million miles driven in the second quarter of 2025. This safety data comes from the company’s Vehicle Safety Report.

Tesla drivers not using Autopilot technology recorded one crash for every 963,000 miles driven during the same period. National 2023 data show that an automobile crash occurs approximately every 702,000 miles in the United States. FSD (Supervised) requires active driver supervision.

FSD (supervised) current pricing

As of late 2025, the one-time purchase price for Full Self-Driving (Supervised) is 8,000 dollars. The cost of the FSD package has varied, with a peak value of $15,000 in the past.

For customers who prefer a monthly option, the FSD (Supervised) subscription costs $99.00 per month. The subscription price remains the same regardless of whether the vehicle is equipped with Basic Autopilot or Enhanced Autopilot. Hardware upgrades are not included with the subscription price.

FSD expansion in Australia and New Zealand

Local reports confirm that FSD (Supervised) has launched in Australia & New Zealand with pricing of AU$149/NZ$159, but Tesla hasn’t published official early-usage kilometers. Australia is the first right-hand drive market to receive public access to this FSD version.

The rapid adoption saw drivers average about 80,000 kilometers per day with FSD (Supervised) active. The subscription service is priced at $149 per month in Australia and $159 per month in New Zealand. This expansion boosts global FSD utilization.

Cybercab design and production timeline

Tesla’s dedicated autonomous ride-hailing vehicle, the Cybercab, is scheduled to start volume production in 2026. It is engineered purely for autonomous use, featuring no steering wheel or pedals for human drivers.

The Cybercab is designed as a two-seater electric vehicle. It uses an “unboxed” manufacturing strategy for cost optimization. Enthusiast outlets reported late-2025 sightings of a Cybercab test program, including crash-test activity; Tesla has not issued an official test report.

Cybercab hardware and initial use

The Cybercab will be one of the first vehicles to use the powerful new AI5 chip. This chip offers a performance difference far exceeding that of previous chip generations, which is required for the highest safety and redundancy levels.

The vehicle has dihedral doors and a simple cabin design. Musk’s most recent guidance targets 8–10 U.S. metros by the end of 2025, with some routes in Austin expected to operate without safety monitors, pending regulatory approvals.

Robotaxi service launch and incidents

Tesla launched its initial Robotaxi service in a limited, invite-only program on June 22, 2025, in Austin, Texas. The debut used modified Tesla Model Y vehicles with human safety monitors to test the FSD software commercially.

Scrutiny documented safety incidents in June 2025, including the steering wheel jerking violently, causing the car to drift over double yellow lines. Other issues included erratic braking near police cars and violating an accessible parking space.

AI4 sensor suite changes

Hardware 4 (AI4) included an overhaul of the sensor suite. The new system is speculated to include a high-definition radar, possibly the 77 GHz Arbe Phoenix SAR unit. This HD radar has a projected range of up to 300 meters.

The camera modules feature a characteristic red tint believed to improve low-light performance. The new HW4 repeater cameras have narrower indicators than HW3. A new heating element was also added for the front-facing windscreen cameras.

Hardware 4 computer power and memory

The FSD Computer 2 (AI4) increased its CPU core count from 12 to 20 and its NN Processor frequency from 2.0 GHz to 2.2 GHz. It now features three Neural Network (NN) Processors, compared to two in HW3.

HW4 is estimated to be manufactured using TSMC’s 7nm or N4 (4nm class) process node, a shrink from the 14nm process node of HW3. HW4 consumes about twice as much power at idle. Retrofitting from HW3 to HW4 is not planned due to extensive changes.

AI5 production timeline and goals

The AI5 chip design was finalized in 2025, with mass production for consumer vehicles now planned for late 2026 or early 2027. Initial production will begin in Taiwan before moving to TSMC’s new fab in Arizona.

Tesla’s goal for the AI5 chip is an oversupply. Any chips not used in new vehicles will be deployed in the company’s data centers. This move is strategic, ensuring a stable supply for their long-term AI needs.

The AI6 chip contract

Tesla’s next-generation chip, after the AI5, is named the AI6 and is part of a significant contract with Samsung Foundry. In July 2025, Samsung revealed a contract valued at 16.5 billion dollars (approximately 23.69 trillion Korean won).

Mass production of the AI6 chip is not expected to occur until 2028 at the earliest. Samsung’s new facility in Taylor, Texas, is scheduled to begin operation in 2026 and will be used to produce these future AI chips.

Samsung Taylor, Texas fab investment

The new semiconductor facility in Taylor, Texas, represents Samsung’s largest-ever investment in the U.S. The initial minimum investment in the Taylor fab is set at $ 17 billion, covering buildings, equipment, and property improvements.

The Taylor facility is scheduled to begin operations in 2026. This location will focus on producing advanced logic chips for areas like 5G and artificial intelligence. The Taylor, TX fab is a multi-billion-dollar project with thousands of projected jobs; U.S. announcements confirm major federal support, and operations are expected to begin in 2026.

AI5 expected performance leap

The AI5 chip is analyzed to promise a massive 40x performance boost over the current HW4. This leap is attributed to having an estimated 8 times more raw compute power and 9 times more memory than HW4, which has 16 GB of RAM.

The AI5 is also expected to feature a five times improvement in memory bandwidth. This architectural change aims to resolve existing hardware bottlenecks. The added power will support larger AI models, including those with 10 times more parameters.

Looking for a more affordable Tesla option? Check out Tesla introduces lower-cost Model Y Standard variant in Europe.

Tesla’s AI compute

Tesla continues to rely on Nvidia GPUs for its AI model training efforts. The company confirmed it uses a combination of Nvidia hardware and its own in-house designed AI chips for different tasks.

As of October 2025, Tesla confirmed it possesses a total computing power equivalent to 81,000 Nvidia H100 chips. This in-house chip strategy positions Tesla alongside tech giants like Google and Amazon, who are creating their own AI hardware.

Want to know what’s behind the latest Tesla safety probe? Read more in NHTSA launches investigation into nearly 2.9 million Tesla vehicles.

What do you think about Samsung powering Tesla’s next big AI move? Drop your thoughts below.

Read More From This Brand:

• Elon Musk Slams ‘Corruption’ Over Missing Superchargers
• Musk hits half-trillion valuation as Tesla shares surge
• Tesla and Elon Musk face shareholder lawsuit over Robotaxi statements

Don’t forget to follow us for more exclusive content right here on MSN

If you liked this story, you’ll LOVE our FREE emails. Join today and be the first to get stories like this one

This slideshow was made with AI assistance and human editing.