Tesla’s ambition to dominate artificial intelligence and robotics has always rested on a simple premise: that its in-house silicon can outpace and outclass anything bought off the shelf. That premise is getting harder to sustain every time another senior engineer walks out the door. The latest departure is Shishuang Sun, a former Senior Director of AI Hardware Design at Tesla, who has joined DensityAI — the startup founded by Tesla’s own ex-Dojo leadership. Sun’s move, confirmed by his LinkedIn profile, marks another significant loss for a company that has spent the past year trying to convince investors its custom chip roadmap is intact.
Sun is not a peripheral figure in Tesla’s hardware story. He spent more than five years across two stints at the company, rising to one of the most senior hardware roles in the organization. His expertise spans the unglamorous but critical disciplines of chip packaging, power delivery, and system integration — the exact skills required to turn a theoretical chip design into a physical product that survives the heat, noise, and mechanical stress of a real vehicle or data center. His departure to DensityAI, where he now leads packaging and system hardware, is a signal that the talent drain from Tesla’s AI division is far from over.
A two-decade career arc through Tesla’s silicon ambitions
Sun’s professional trajectory mirrors the full rise and fall of Tesla’s in-house chip program. He first joined the company during the Jim Keller era, a period when Tesla was assembling the engineering muscle to design its first Full Self-Driving (FSD) computer. That effort was a landmark achievement: Tesla broke its dependence on Nvidia for inference workloads, producing a custom chip that could run the company’s neural networks efficiently inside every vehicle it sold.
Sun was part of the team that made that happen. His role in the Autopilot hardware group involved the physical design and validation work that turns a chip layout into a reliable, manufacturable component. When that chapter of Tesla’s history wound down, he left for Waymo, where he spent just over two years as a hardware engineering manager. At Waymo, Sun oversaw design verification, signal and power integrity, EMC, and PCB teams — the disciplines that ensure self-driving hardware functions correctly in the chaotic electromagnetic environment of a moving vehicle.
In January 2021, Sun made the decision to return to Tesla. The company was ramping up its Dojo supercomputer project, and Sun came back as a principal engineer on AI hardware covering both Dojo and Autopilot. His promotion to Senior Director of AI Hardware Design came in April 2025, a title that placed him among the top hardware executives in the company. His remit included IC packaging, vertical power modules, signal and power integrity, PCB design, and system thermal and mechanical engineering — a portfolio that reads like a checklist of every way a chip design can fail in the real world.
That promotion, however, came at a strange time. By April 2025, the Dojo project was already collapsing. Tesla had shifted its training compute strategy away from its in-house supercomputer, and the team that had built Dojo was leaving in droves. Sun’s elevation may have been an attempt to retain senior talent, but it clearly did not hold. He left Tesla in July, joining DensityAI after a brief transition period.
DensityAI: The startup that grew out of Dojo’s ashes
DensityAI is not a random competitor poaching talent. It is the direct successor to Tesla’s Dojo supercomputer program, founded by Ganesh Venkataramanan, the former Dojo chief. When Tesla formally shut down Dojo in August 2025, roughly 20 members of the supercomputer team followed Venkataramanan out the door to the new venture. Bill Chang, another Tesla veteran, serves as the company’s CTO.
The startup’s mission is to build chips, hardware, and software for AI data centers, with a focus on automotive, robotics, and industrial customers. That is, in essence, the exact mission Tesla abandoned when it killed Dojo and pivoted to Nvidia for all its training compute needs. DensityAI is effectively running the playbook that Tesla wrote, but without the distraction of also building cars, batteries, and humanoid robots.
Sun’s arrival at DensityAI gives the startup another layer of credibility in the packaging and system integration space. Venkataramanan is a chip architect by training, and Chang brings deep systems experience, but the physical design work that Sun leads is a different beast. Advanced packaging has become one of the most constrained and valuable skills in the semiconductor industry, and DensityAI now has one of the people who did it at Tesla’s scale for half a decade.
Why chip packaging expertise has become the industry’s bottleneck
To understand why Sun’s departure matters, it helps to understand what advanced packaging actually is and why it has become the critical constraint in modern AI hardware. The transistor design of a chip determines its theoretical compute capacity, but the packaging determines whether that compute can be fed with enough power and kept cool enough to operate. As AI workloads have grown, chips have become so power-hungry that the packaging and power delivery systems are often the limiting factor in what a chip can actually achieve.
Modern AI accelerators draw hundreds of watts of power, sometimes approaching a kilowatt per chip. Getting that much electricity into a package the size of a postage stamp, while extracting the equivalent of a space heater’s worth of heat, requires engineering that is closer to plumbing than to traditional semiconductor design. Voltage regulator modules, interposers, thermal interfaces, and mechanical mounting systems all have to work in concert. A chip that works perfectly in simulation can fail catastrophically in the field if the package can’t handle the current density or thermal load.
This is precisely the expertise Sun cultivated over his career. His work on vertical power modules and thermal management at Tesla was about making sure the company’s custom silicon could actually function in production vehicles and in the Dojo data center. The skills he developed are transferable to any AI hardware project, and DensityAI clearly recognized that value.
Tesla’s chip roadmap is now more dependent than ever on scarce talent
The timing of Sun’s exit could not be worse for Tesla. The company taped out its long-delayed AI5 chip in April, nearly two years behind Elon Musk’s original timeline. Volume production for AI5 is not expected until mid-2027, assuming no further slips. Beyond that, AI6 is slated for Samsung’s 2nm process, a node that has been plagued by yield problems. Musk has also floated the idea of a “Dojo 3” supercomputer, a resurrection of the project that was supposedly dead.
On the manufacturing side, Tesla and SpaceX are reportedly pitching a $20 billion-plus chip factory in Austin called “Terafab,” which would represent an unprecedented leap into semiconductor manufacturing for both companies. That project, if it moves forward, would require an enormous hiring spree of exactly the kind of engineers that Tesla is now bleeding.
Every one of these initiatives depends on people who can take a tape-out and turn it into a shipping product. The chip design itself is only the first step. The packaging, power delivery, and thermal work that Sun led at Tesla for five years is what determines whether a chip is a paperweight or a product. Losing a senior director with that specific profile, a full year after the Dojo shutdown, suggests that Tesla’s internal retention efforts are not working as well as the company’s public statements suggest.
What this departure says about Tesla’s AI strategy
The steady exodus of hardware talent from Tesla raises a fundamental question about the company’s AI strategy. Tesla has staked enormous value on the claim that it can build better AI hardware in-house than it can buy from Nvidia or other suppliers. That claim has driven the company’s valuation for years, with Musk repeatedly promising that Dojo would make Tesla the leader in AI training compute.
But custom silicon is a people business before it is a fab business. The knowledge required to design, package, and deploy a world-class AI chip resides in the heads of a relatively small number of engineers. When those engineers leave — not to retire or to go to a giant like Apple or Google, but to build a direct competitor founded by their former boss — it is the strongest possible signal that they no longer believe in Tesla’s vision.
Venkataramanan’s departure from Dojo was framed by some as a personality conflict or a strategic disagreement. Chang’s exit was similarly explained. But now, a year later, a sitting senior director of hardware design has walked out the same door. The cumulative effect is harder to dismiss. These are not individual cases of malcontent; they are a pattern of people voting with their feet.
The irony is difficult to overstate. Tesla killed Dojo, citing a shift in strategy toward Nvidia’s more mature ecosystem. The team that built Dojo did not disappear. They spun up a competitor that is now executing on the same vision, with the same technology approach, and an increasingly deep bench of the exact people who made Tesla’s custom silicon work. The company that walked away from its supercomputer project is now watching the successor company build the future it abandoned.
What is at stake for Tesla’s valuation and its robotics ambitions
The stakes are not abstract. Tesla’s valuation includes a significant premium for its perceived leadership in AI and robotics. The Optimus humanoid robot, the robotaxi network, and the full self-driving software suite all depend on having compute hardware that is fast, efficient, and cheap enough to deploy at scale. If Tesla falls behind on custom silicon, it will have to buy compute from Nvidia or others, which erodes the cost advantage that is central to the robotaxi business model.
Musk has acknowledged this dependency, which is why he keeps promising new chips and new supercomputers. AI5 is supposed to be a major leap over the current HW4 platform, and AI6 is meant to be even more significant. But those promises are only as good as the team that can deliver them. The intellectual property of a chip company lives in its engineers, not in its patents or its fab contracts. Each senior departure represents a piece of institutional knowledge walking out the door, often directly to a competitor.
The Terafab proposal adds another layer of risk. Building a chip factory is one of the hardest engineering and operational challenges in the world. It requires a workforce that understands semiconductor manufacturing at a level that no car company has ever approached. Recruiting for such a facility will require attracting talent from established chipmakers like TSMC, Samsung, and Intel. The prospect of that recruiting process succeeding while Tesla is simultaneously losing its most senior chip hardware leaders to a startup is not encouraging.
DensityAI’s growing advantage in the AI hardware talent market
For DensityAI, Sun’s arrival is a validation of its model. The startup does not need to build cars or run a global charging network. It can focus entirely on AI hardware, offering engineers the chance to work on cutting-edge problems without the chaos of Tesla’s corporate environment. The fact that it is pulling senior directors — not just junior engineers — suggests it is now competing at the highest level of the talent market.
The company’s positioning in the automotive, robotics, and industrial AI markets is also strategic. These are the markets where Tesla claims to be the leader, and where the need for efficient, reliable AI compute is most acute. If DensityAI can build a compelling alternative to both Nvidia and Tesla’s in-house efforts, it could become a significant supplier to the very industries Tesla is trying to dominate.
There are also financial implications. Custom silicon is expensive to develop, and DensityAI will need substantial funding to bring its chips to market. But the team it has assembled — including Venkataramanan, Chang, and now Sun — is precisely the kind of pedigree that attracts venture capital. Investors have seen the Dojo story up close, and they know what these engineers are capable of building.
The practical answer to the industry’s most pressing question
What is the actual impact of losing a packaging engineer to a startup, and why should anyone outside the semiconductor industry care? The answer is that advanced packaging is the difference between a chip that performs on paper and a chip that performs in the field. A cutting-edge AI chip that cannot be powered or cooled effectively is just an expensive slab of silicon. The engineers who solve those problems are as valuable as the architects who design the compute cores, and they are substantially harder to replace.
Sun’s specific skills in vertical power modules and thermal management are among the most sought-after in the industry. As AI chips consume more power, the packaging problem becomes proportionally harder. Tesla now needs to find someone with comparable experience to lead AI hardware design through the AI5 ramp and the AI6 development cycle. That search will take time, and time is not a luxury Tesla has, given that it has already fallen nearly two years behind its original AI5 timeline.
A pattern of attrition that demands attention
The bigger picture is that this is not an isolated incident. Tesla has lost its Dojo chief, its CTO of the supercomputer project, a significant portion of the Dojo engineering team, and now a senior director of AI hardware design. This is not a trickle; it is a systemic outflow that started well before the August shutdown announcement and has continued uninterrupted since.
When a company pivots away from a major internal project, some attrition is expected. But the exodus from Tesla’s AI hardware group has a specific character. The people leaving are not scattering to a dozen different companies. They are concentrating in one rival organization that was founded to continue the work Tesla abandoned. That is a strategic threat, not just a human resources issue.
The most telling detail is the timeline. Musk announced Dojo’s shutdown in August 2025, but the exodus began earlier and has now extended more than a year past the announcement. The idea that the situation has stabilized is directly contradicted by the fact that a senior director was still leaving in July 2026. The month of July in 2026 is when Sun lists his start at DensityAI. That is a full year after Tesla supposedly regrouped around AI5 and AI6, and the bleeding is still happening.
Custom silicon at this level is a long game. A senior director of hardware design does not leave on a whim. The decision to uproot a career that had been rebuilt around Tesla’s AI ambitions, rejecting a promotion that had been awarded only three months earlier, is the kind of choice that comes after deep reflection. Sun looked at the roadmap, looked at the leadership, and decided that DensityAI was a better bet for the next stage of his career. That assessment carries weight.
The company that Tesla created in its Dojo project has become its most significant competitor in AI hardware talent, and that competition is only intensifying. As DensityAI scales up, it will need more packaging engineers, more power delivery specialists, more systems integrators. The talent pool is finite, and Tesla is no longer the most attractive destination in that pool.
For now, Tesla’s public story is one of confidence in its AI5 and AI6 roadmap. Behind the scenes, the evidence suggests a different story: a hardware team that is losing its most experienced people to the very startup that grew out of its own failed project. The chips are being designed, but the question of who will be left to package them, power them, and ship them remains open. And in the world of custom silicon, that question is not a detail. It is the whole game.
