Elon Musk unveils Tesla AI5 chip with 40x performance boost, but production delayed until 2027

Elon Musk demonstrates first sample of Tesla AI5 processor, accidentally thanks TSC rather than TSMC -- claims 40X performance boost over the predecessor

Elon Musk on Wednesday showcased an image of one of the first samples of Tesla's AI5 hardware that will be used to drive AI applications in Tesla's cars, Optimus robots, and potentially xAI data centers. The AI5 processor is about half of the reticle size, uses industry-standard memory, and yet can be up to 40X faster than AI4 in certain scenarios, according to Elon Musk. "Congrats to the Tesla_AI chip design team on taping out AI5," Elon Musk wrote in an X post. "AI6, Dojo 3 and other exciting chips in [the works]. [...] And thank you to @TaiwanSemi_TSC and Samsung for your support in bringing this chip to production! It will be one of most produced AI chips ever." The Tesla AI5 processor module features a fairly small ASIC die (about half the reticle size, according to Musk's previous comments) surrounded by 12 memory packages from SK hynix (most likely GDDR6/7). The module uses an organic substrate, and the memory packages are marked like industry-standard DRAM products. While we do not know how wide AI5's memory interface is, 12 memory packages clearly indicate that we are dealing with a fairly wide memory I/O. If we are indeed dealing with 12 GDDR6/7 memory ICs, then the Tesla AI5 ASIC has a 384-bit memory interface. Depending on the memory type used, the Tesla AI5 can offer memory bandwidth between 768 GB/s and 1.536 GB/s. Exact performance of the AI5 has not been disclosed, though Musk claims significant -- up to 40X -- improvements over AI5 in select cases. "I think the Tesla chip team is really designing an incredible chip here: by some metrics the AI5 chip will be 40 times better than the AI4 chip," Elon Musk said during Tesla's Q3 2025 earnings call. "As a result of [outdated hardware] deletions, we can actually fit AI5 [on a] half [of a] reticle with good margin for the traces from the memory to the Tesla trip accelerators, the Arm CPU cores, and PCIe blocks." Although Musk claims that the AI5 has just been 'taped out' (which means that the final chip design has been sent to a photomask house), he actually shows an already fabricated processor with a 'KR 2613' marking on it, which suggests that the ASIC was packaged on the 13 week of 2026. Musk also mentions Taiwan Semiconductor (TSC) and Samsung for bringing the chip to production, though we are not sure that the producer of passive components has anything to do with bringing the AI5 processor to production. More likely, Musk meant Taiwan Semiconductor Manufacturing Co., better known as TSMC. Previously, the head of Tesla, SpaceX, and xAI said that the AI5 would be made by both TSMC and Samsung Foundry, though we do not know which contract chipmaker fabbed the current sample. Assuming that Tesla got the chip in March or early April and no re-spin is required, it is reasonable to expect the company to deploy the processor sometimes in 2027. Perhaps the most intriguing part of the announcement is that Tesla has apparently not given up on the Dojo system-on-wafer (SoW) processor for AI training, and the Dojo 3 processor is in the works. It was reported last August that the Dojo wafer-level processor initiative had been abandoned and the team behind it dismantled. Indeed, Peter Bannon, the head of the Dojo project at Tesla, retired last August, according to his LinkedIn page. Elon Musk said in July that the AI6 and Dojo 3 could feature a converged architecture (a converged ISA, we would speculate), which would enable the company to unify its software stack and could potentially allow the company to unify its hardware stack as well. "I think about Dojo 3 and the AI6 as the first [converged architecture designs]," Musk said in a July 23 earnings call (via Investing.com). "It seems like intuitively, we want to try to find convergence there where it is basically the same chip that is used where we use, say, two of them in a car or an Optimus and maybe a larger number on a on a [server] board, a kind of 5 - 12 twelve on a board or something like that. [...] That sort of seems like intuitively the sensible way to go." Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Tesla taped out AI5 chip, Musk says -- nearly 2 years behind schedule

Tesla has taped out its next-generation AI5 self-driving chip, CEO Elon Musk said overnight, a key milestone that sends the final design to the foundry for fabrication. The tape-out comes nearly two years after Tesla originally promised AI5 would be in vehicles, and it still leaves volume production more than a year away. In a post on X at 3:21 AM today, Musk wrote: "Congrats to the @Tesla_AI chip design team on taping out AI5! AI6, Dojo3 & other exciting chips in work." He shared this picture of the chip: A "tape-out" is the point at which a chip's design is locked and sent to a semiconductor foundry to begin fabrication. It is a meaningful milestone -- but it is not the finish line. After tape-out, the chip still has to be manufactured, tested in silicon, validated, and ramped to volume production. For an automotive-grade AI accelerator, that process typically takes 12 to 18 months. Tesla is reportedly using TSMC for AI5 production, while Samsung's 2nm line is tied to the follow-on AI6 chip. We reported last month that AI6 has already slipped roughly six months because of Samsung 2nm yield issues, pushing mass production to Q4 2027 at the earliest. Today's tape-out is worth putting next to Tesla's own earlier promises on this chip: Musk has previously described AI5 as up to 10x more powerful than AI4 and floated a 9-month design cycle for subsequent generations -- a cadence that, as we've noted, is unheard of for a major architectural overhaul in the semiconductor industry. Tesla's own "AI4.5" stopgap computer, quietly introduced in 2026 Model Y vehicles late last year, exists precisely because of this delay. The company needed more compute to keep feeding larger FSD neural networks while AI5 slipped. The practical implication of today's milestone is straightforward: AI5 is real and moving, but it will not be in a meaningful number of vehicles this year. Tesla has already confirmed the Cybercab -- scheduled for production in Q2 2026 -- will launch on AI4, the same hardware currently sold in Model Y, Model 3, Model S, Model X, and Cybertruck. Tesla has said it needs "several hundred thousand completed AI5 boards line side" before it can switch production lines, and that volume isn't expected until mid-2027. AI6, Dojo3, and whatever other chips Musk is alluding to remain on paper. Tesla's AI hardware progress is, honestly, bittersweet. On one hand, we want to see Tesla's self-driving compute get better. FSD is clearly compute-bound on bigger neural nets, and more powerful silicon is part of the path forward. A taped-out AI5 is real progress after years of slipping timelines, and it's good to see the chip finally heading to the fab. On the other hand, every new Tesla chip generation carries the same problem: it makes the last one obsolete before Tesla has delivered what it sold on that hardware. Tesla sold millions of cars with HW3 and HW4 on the promise of unsupervised self-driving. HW3, by Musk's own admission last year, can't do it. HW4 is running V14, and it still isn't unsupervised. AI4.5 is already in new Model Ys. Now AI5 is taped out, and AI6 and Dojo3 are coming. The pattern is hard to miss: Tesla keeps moving the goalpost to the next chip instead of delivering what was promised and sold on the current one. Every tape-out announcement is, in that sense, also a quiet confirmation that a lot of existing customers are not getting the product they paid for on the hardware they own. That's the part Musk never congratulates the team for.

Tesla AI5 AI chip taped out in partnership with Samsung and TSMC

TL;DR: Tesla has finalized the design of its AI5 chip for Full Self-Driving, matching NVIDIA Hopper's performance and potentially rivaling Blackwell with dual setups. Produced at Samsung and TSMC facilities, AI5 features up to 192GB LPDDR5X memory and aims for mass production by late 2026 or early 2027. Tesla CEO Elon Musk took to X (formerly Twitter) to announce that Tesla has completed the design of its AI5 chip for Full Self-Driving (FSD). Musk congratulated the Tesla AI team on the successful tape-out of AI5 and also shared an image of the full chip. We covered the rumors surrounding AI5 production earlier this year, and now it seems the chip is on track for mass production. The AI5 chip is an AI-focused processor that follows HW4, but Tesla has changed the nomenclature to adopt the "AI" naming scheme going forward. This chip will be employed in Tesla's next-gen Full Self-Driving (FSD) suite. Musk claims that AI5 will be a very capable solution, with performance equivalent to NVIDIA's Hopper architecture. He further adds that a dual AI5 SoC setup can rival NVIDIA Blackwell in performance. In the image of the chip Musk posted, we can see the DRAM chips sourced from SK Hynix. There are two rows of SK Hynix LPDDR5X memory on each side of the chip, with each row having three modules for a total of 12. If these are 16GB modules, that translates to 192GB of LPDDR5X memory for each AI5 SoC. Musk also thanked Samsung and TSMC for their support in production, which is apt, since the chips will be produced at Samsung's plant in Taylor, Texas, and at TSMC's plant in Arizona. This two-pronged production strategy is put in place to diversify the supply chain. Official information on the production timeline remains unknown, but industry experts are eyeing high-volume production in late 2026 or early 2027. As confirmed by Tesla's CEO, work is already underway on the next-generation AI6 chip and Dojo3. For Tesla, the AI5 tape-out was the main milestone, as Musk claimed it was almost "existential" for the company. Now, it can be sent to Samsung and TSMC for production. While full specifications of the chip are not yet known, Elon's previous discussions point towards a 40x improvement over the HW4 chip, along with new features.

Tesla A15 AI Chip Taped Out, Elon Musk Shares First Pictures & Confirms A16 & Dojo3 In Work

Elon Musk has shared the very first update on the Tesla A15 chip, confirming its successful tape out & plans for A16 and Dojo3. Elon Musk Shares First Pictures of Tesla A15 After Successful Tape Out, The Road Ahead Includes A16 & Dojo3 In the latest update on X, Elon Musk has congratulated the Tesla AI team on the successful tape out of its A15 AI chip. Elon has also shared pictures of the chip, featuring a large primary die in the middle that will handle all of the compute and 12 DRAM modules on the outskirts, offering high capacity and efficiency-optimized bandwidth. A15 is the follow-up to HW4 and will arrive as a next-gen FSD solution for Tesla. In a previous talk, Elon Musk stated that A15 will offer a monumental 40x improvement over the HW4 chip, with 8x raw compute capabilities, 9x memory, and will deliver brand new features. The chip itself is expected to offer close to 2500 TOPS of AI compute, 144 GB of memory per chip, and is designed with the latest transformer engine in mind. The chip is expected to be manufactured at TSMC and Samsung, with a high-volume production slated for late 2026 or early 2027. Elon has shared plans to move next-gen chip production to the upcoming TeraFab, but its announcement is still pending. Meanwhile, Elon also confirms that work is already underway on the next-generation Tesla A16 chip and Dojo3. Since Tesla announced a return to form in the chipmaking business, the plans for Dojo3 are back on track. Tesla resumed the plans for the Supercomputer project back in January 2026, and with Terafab operational and developing DRAM, Packaging, Chips, all under one roof, this goal should not be far away from now. Follow Wccftech on Google to get more of our news coverage in your feeds.