Tesla Tapes Out AI5 Chip With 40x Performance Claims Over Predecessor as Musk Teases AI6 and Dojo 3
Tesla's next-generation AI5 processor has completed tape-out, reaching the final design milestone before mass production, with volume deployment not expected until 2027.
Overview
Tesla CEO Elon Musk announced on April 15 that the company’s next-generation AI5 processor has completed tape-out, the final step in chip design before a design is sent to a foundry for fabrication. The milestone comes nearly two years after Musk first promised the chip would be shipping in vehicles, and volume production remains more than a year away. Musk also confirmed that work on successor chips AI6 and Dojo 3 is already underway.
What We Know
Musk shared the news via X in the early hours of April 15, writing: “Congrats to the @Tesla_AI chip design team on taping out AI5! AI6, Dojo3 & other exciting chips in work,” as reported by Electrek.
The AI5 processor represents a substantial leap over its AI4 predecessor. Musk has claimed the chip delivers up to 40 times better performance than AI4 by certain metrics, with approximately nine times more memory and eight times higher compute capacity, according to Tom’s Hardware. The AI5 die occupies roughly half a reticle and is surrounded by 12 SK Hynix memory packages, likely GDDR6 or GDDR7, providing up to 192 GB of LPDDR5X memory. Its 384-bit memory interface is estimated to deliver between 768 and 1,536 GB/s of bandwidth depending on memory type.
A sample chip bore the marking “KR 2613,” suggesting it was packaged during the 13th week of 2026 — meaning physical samples were already in hand before the public announcement, as Tom’s Hardware noted.
Manufacturing is split between two foundry partners. TSMC will fabricate AI5 at its Arizona facility, while Samsung will produce chips at its Taylor, Texas plant. Tesla signed a $16.5 billion contract with Samsung in July 2025 primarily targeting the next-generation AI6 chip, though Musk indicated actual spending could exceed that figure, as reported by TechCrunch. The dual-foundry strategy is intended to diversify Tesla’s supply chain as it scales chip production.
Musk clarified that AI5 will initially target the Optimus humanoid robot and Tesla’s supercomputer clusters rather than consumer vehicles. The Cybercab robotaxi, expected to begin production in Q2 2026, will ship with current-generation AI4 hardware. Musk has stated that AI4 is already sufficient to enable Full Self-Driving to surpass human safety standards, according to Electrek.
In a minor but widely noted gaffe, Musk tagged the wrong company in his announcement, thanking “@TaiwanSemi_TSC” — a passive component manufacturer — rather than TSMC’s official account, as Tom’s Hardware reported.
What We Don’t Know
Several important details remain unclear. Tesla has not disclosed the specific process node used for AI5 fabrication, making independent verification of performance claims difficult. The 40x improvement figure cited by Musk lacks a clear benchmark methodology, and it is not yet known whether this refers to raw compute throughput, inference performance, or a composite metric.
The timeline from tape-out to volume production also carries uncertainty. Automotive-grade AI accelerators typically require 12 to 18 months of testing and validation after tape-out, as Electrek noted. Tesla has said it needs several hundred thousand completed AI5 boards before switching production lines, with volume not expected until mid-2027. Whether the existing samples will require a design respin — common in complex chip development — remains to be seen.
The AI5 timeline has also slipped repeatedly. Musk initially promised the chip in vehicles by the second half of 2025. In July 2025, the design was declared “finished.” By January 2026, Musk described it as “almost done” — six months after the prior completion claim. The tape-out announced today places the project roughly two years behind its original schedule, according to Electrek.
Analysis
The AI5 tape-out is a genuine engineering milestone for Tesla’s in-house silicon effort, but its practical impact on Tesla’s near-term product lineup is limited. The Cybercab launches on AI4, and consumer vehicles are unlikely to receive AI5 until 2027 at the earliest. The chip’s initial deployment in Optimus robots and training clusters suggests Tesla views custom silicon as central to its broader ambitions in robotics and AI infrastructure rather than as an immediate upgrade to its automotive stack.
Musk’s comparison to Nvidia’s product line — claiming a single AI5 matches an H100 while a dual configuration approaches Blackwell-class performance — positions the chip as competitive with data center GPUs costing tens of thousands of dollars. If validated, this could reduce Tesla’s dependence on Nvidia hardware for internal AI workloads. The $16.5 billion Samsung foundry contract signed in July 2025, as reported by TechCrunch, underscores Tesla’s commitment to building a vertically integrated AI hardware pipeline that extends well beyond automotive applications.
The dual-foundry approach with Samsung and TSMC is notable in an industry where most fabless chip companies work with a single manufacturing partner per product generation. It signals Tesla’s intent to produce AI chips at a scale that warrants supply chain redundancy, though it also introduces complexity in maintaining consistent yields and performance across two different fabrication processes.