Tesla Loses Another Dojo Chip Architect to DensityAI

Tesla has lost another senior chip engineer to DensityAI, the AI-hardware startup assembled almost entirely from its own former Dojo team.

Shishuang Sun spent more than five years at Tesla as an AI hardware architect working on Dojo and the Autopilot computers, and was promoted to senior director of AI hardware design in April 2025. Electrek reports he now lists his role as packaging and system hardware at DensityAI, having joined in July 2026.

Who DensityAI is

DensityAI was founded by Ganesh Venkataramanan, who ran Dojo at Tesla until he left in 2024. It builds chips, hardware and software for AI data centres aimed at automotive, robotics and industrial customers — which is, more or less, a description of what Dojo was supposed to become.

Tesla shut Dojo down in August 2025 and redirected the effort into its AI5 and AI6 inference chips. Roughly twenty members of the supercomputer team have since gone to DensityAI, including Bill Chang as chief technology officer. Sun is the latest, not the first.

Why packaging is the awkward part

Losing a chip architect sounds like a design problem. Sun's specialism says otherwise: IC packaging, power modules, signal integrity, PCB design, thermal and mechanical engineering. That is the discipline that turns a working design into hardware that can be manufactured at volume and survive a decade in a car.

Tesla taped out AI5 in April 2026. A tape-out is the design going to the foundry; everything between that and a chip sitting behind a windscreen is packaging, power delivery, thermal design and validation. It is the phase Tesla is in right now, and it is the phase Sun was hired to run.

That is a narrower concern than "Tesla is losing its chip talent", and a more specific one. Tesla has said AI5 will be built by both TSMC and Samsung, on different processes, which multiplies the packaging and qualification work rather than dividing it.

What it means for a Tesla owner

Very little today, and potentially a lot in 2028.

AI5 is the computer intended to run the FSD versions Tesla has been promising for cars it has already sold. The JPMorgan note published last week put AI5 volume production in 2027 and the AI4-to-AI5 transition running through 2028, and Tesla has separately claimed wafer-efficiency records for AI6. Those timelines assume the ramp goes smoothly.

Owners with a stake in this are the ones holding a paid FSD licence on AI4 hardware, plus anyone who bought on the promise of a retrofit. Tesla has already demonstrated that hardware upgrades are physically possible — a builder fitted AI4 into a 2022 Model 3 — while maintaining that doing it at fleet scale is not. Every month AI5 slips is a month that argument stays unresolved.

What to hold lightly

This is one engineer, sourced from a LinkedIn profile change and reported by a single outlet. Tesla has not commented, no AI5 schedule has moved on the record, and senior engineers leave large companies constantly.

The reason it is worth noting is the pattern rather than the person. A rival staffed by twenty of your former specialists, run by your former programme lead, recruiting from the exact discipline your current programme depends on, is a different situation from ordinary attrition. Watch whether AI5's 2027 production date survives Tesla's next quarterly call.