Samsung's Taylor fab begins Tesla AI5 chip production on 2nm, proving yield breakthrough

Samsung Confirms Tesla AI5 Chip Production on Advanced 2nm Node

Samsung has reached a critical milestone in its foundry business, completing the Tesla AI5 chip design and preparing to manufacture it at the Samsung Taylor fab using the Samsung 2nm process, according to industry officials 1

1

4

4

. James Kim, a principal engineer at Samsung Foundry, revealed on LinkedIn that "the Tesla-Samsung AI5 chip has reached tape-out" and is "scheduled to be manufactured at the Taylor fab using our latest 2-nanometer process and will soon be integrated into Tesla's newest products" 1

1

. Kim deleted the post after Korean news outlets picked it up, but the disclosure marks a significant shift in expectations for Tesla's AI5 self-driving chip development.

The tape-out stage represents the point where a chip's design is locked and handed over for manufacturing 4

4

. From there, the design undergoes prevalidation, photomask production, and wafer fabrication to create engineering samples, which must pass customer qualification before entering mass production. Elon Musk confirmed in April that Tesla had taped out the Tesla AI5 chip to both Samsung and TSMC, with each foundry producing "slightly different versions" because "they translate designs to physical form differently" 1

1

4

4

.

Unexpected 2nm Process Selection Signals Yield Breakthrough

The most surprising detail is Samsung's choice to manufacture the Tesla AI5 chip on its Samsung's advanced 2nm process technology rather than a more mature node 1

1

. Industry assumptions had placed Samsung's 2nm node with Tesla's follow-on AI6 chip, while AI5 would use established processes at TSMC. This shift matters because Samsung's 2nm yield has improved beyond the critical 60% threshold where the process becomes viable for high-volume customers like Tesla 1

1

4

4

. Earlier reports indicated that AI6 had slipped roughly six months due to Samsung 2nm yield issues, pushing mass production to Q4 2027 at the earliest 1

1

.

Elon Musk has already showcased a Samsung-made prototype, posting an image of the AI5 chip on X with the marking "KR 2613," indicating it was manufactured in Korea in the 13th week of 2026 1

1

3

3

. This evidence confirms Samsung's engineering samples were already coming from its Korean production lines before production at its Taylor, Texas, foundry took over. Samsung held an equipment installation ceremony at the Taylor fab in April and expects volume production of Tesla's AI chips in the second half of 2027 1

1

.

AI5 Specifications Target NVIDIA Blackwell Performance

The Tesla AI5 chip represents a substantial upgrade for Tesla's next-generation Full Self-Driving solutions, following the HW4 chip currently in vehicles 3

3

. Elon Musk stated the AI5 would deliver a monumental 40x improvement over the HW4 chip, with 8x raw compute capabilities and 9x memory 3

3

. The AI semiconductor is expected to offer close to 2500 TOPS of AI compute and 144 GB of memory per chip, designed with the latest transformer engine architecture 3

3

.

Elon Musk made ambitious claims in January, stating the AI5 would rival NVIDIA Blackwell and Hopper products while being significantly cheaper and using less power 2

2

. The chip will be offered in multiple configurations: a single-SOC design that rivals NVIDIA's Hopper, and a dual-SOC design that competes with Blackwell while costing much less to produce and consuming considerably less power 3

3

. The prototype showcased by Elon Musk featured a large primary die for compute and 12 DRAM modules from SK hynix offering high capacity and bandwidth 3

3

.

Samsung Lands Anthropic as Foundry Customer

Samsung's improving 2nm capabilities are attracting additional customers beyond Tesla. Recent reports suggest Samsung lands Anthropic as foundry customer, with the AI company expected to manufacture its own AI chips through Samsung Foundry 1

1

2

2

4

4

. Anthropic listed Samsung, SK hynix, and Micron as "strategic infrastructure partners" during its Series H funding round completed in May, stating they would help "scale our compute reliably at the pace our customers need" 2

2

.

Since Samsung is the only company among Anthropic's infrastructure partners with a foundry business capable of manufacturing non-memory chips, industry sources expect it to win orders for the products 2

2

. These orders could convert Samsung's foundry business deficit into a growth phase, according to insiders 2

2

. For Samsung, landing the Tesla AI5 chip on 2nm provides a proof point the company badly needs after trailing TSMC on advanced-node yield for years 1

1

.

Timeline Challenges and Path to Unsupervised Autonomy

Despite the tape-out milestone, significant hurdles remain before the AI5 reaches vehicles at scale. Elon Musk stated Tesla needs "several hundred thousand completed AI5 boards line side" before switching vehicle production over, with that volume not expected until mid-2027 1

1

. Engineering samples coming off the Texas production line in 2026 don't accelerate that timeline substantially. The AI5 is arriving nearly two years after Elon Musk first said it would be in vehicles, the Cybercab is launching on old HW4 chip hardware, and the AI4.5 stopgap computer quietly added to new Model Ys exists precisely because AI5 kept slipping 1

1

.

From Tesla's perspective, the company doesn't need AI5 to achieve unsupervised autonomy at scale, though similar claims were made about HW3 previously, and the company has yet to deliver on that promise 1

1

. Meanwhile, work is already underway on the next-generation Tesla AI6 chip and Dojo3 supercomputer 3

3

. Tesla resumed plans for the Dojo3 project in January 2026, and with TeraFab operational and developing DRAM, packaging, and chips under one roof, this goal should be within reach 3

3

. High-volume production from both TSMC and Samsung remains slated for late 2026 or early 2027 3

3

.