Ashish Kumar once steered the artificial intelligence efforts behind Tesla’s Optimus humanoid. Now he has left to build something different. The former lead has co-founded a San Francisco-based startup called Intelligent Machines. It focuses on specialized industrial robots. These machines target tasks performed by highly skilled workers in manufacturing and aerospace.

Sources told The Information that Kumar’s new company has already brought in several former Optimus engineers. The move marks yet another high-profile departure from Tesla’s ambitious but troubled robotics program. And it arrives at a moment when the electric vehicle maker is scaling Optimus production to hundreds of units a week while still wrestling with generalization issues in its AI.

Kumar earned a Ph.D. in artificial intelligence from UC Berkeley in 2023. He joined Tesla shortly after and quickly rose to lead the Optimus AI team. His tenure emphasized scalable learning methods. In a 2025 post on X he described swapping classical control stacks for reinforcement learning. The team scaled dexterity by learning directly from videos. “AI is the most significant bit to unlock humanoids,” he wrote at the time of his initial departure.

That exit went to Meta first. Kumar joined the social media giant as a researcher in 2025. Yet barely a year later he has circled back to robotics entrepreneurship. The shift from humanoid generalists to task-specific industrial machines reflects a calculated bet. Humanoids promise versatility but face steep hurdles in real-world deployment. Purpose-built robots for factories can generate revenue sooner.

Tesla itself continues to pour resources into Optimus. Recent reports show the company has increased output tenfold since the second quarter of 2026. Production now hits several hundred robots per week at its Fremont facility. Managers eye more than 1,000 units weekly by year-end. Yet challenges persist. Hands remain finicky. Suppliers create bottlenecks. And the robots still struggle to generalize tasks beyond narrow training environments. Electrek laid out these constraints in late September.

Such limitations help explain the talent outflow. Kumar is hardly alone. Chris Walti spent seven years at Tesla. He started in engineering, moved to mobile robotics, and ultimately led early Optimus development. In 2024 he launched Mytra. The warehouse automation startup has raised more than $78 million. Its systems move heavy payloads that off-the-shelf AMRs could never handle. Walti told TechCrunch that Tesla’s unique demands forced the team to invent solutions in-house. “Nothing in market was suitable,” he said. That pattern of building from scratch now drives his new venture.

Other alumni have followed similar paths. Jay Li led electronics for Optimus. He settled a trade secret lawsuit with Tesla and raised $11 million for Proception. The company ships dexterous robotic hands to researchers and humanoid developers. The Next Web reported the funding and settlement in June 2026. Li described the legal battle as a “resilience test.” He even suggested Tesla could one day become a customer.

These departures highlight a broader trend. Expertise honed inside Tesla’s pressure-cooker environment commands attention on the open market. Venture investors have taken notice. Funding for AI-powered robotics startups has climbed. Warehouse players like Mytra, hand specialists like Proception, and now task-specific industrial efforts like Intelligent Machines all draw from the same talent pool.

Kumar’s startup remains early stage. The Information reported it is seeking roughly $100 million in funding. Details on specific products stay sparse. Yet the focus on manufacturing and aerospace points to applications where precision and reliability trump general-purpose movement. Think aerospace assembly. Complex welding. Or repetitive yet skill-intensive machining. Robots that excel here can command premium pricing without the software complexity of full humanoids.

Tesla’s own Optimus strategy has shifted too. The company plans to lease units initially to factories that mirror its own operations. The goal is to gather data that refines the AI over time. This mirrors the Full Self-Driving approach. Ship hardware. Collect fleet learning. Iterate the model. CEO Elon Musk has called Optimus potentially the company’s most valuable asset. But delivery timelines have slipped repeatedly. No robots perform useful work at scale inside Tesla factories yet.

That gap creates openings. Startups can move faster on narrow problems. They avoid the scrutiny and internal politics that come with being part of a public company’s moonshot. Kumar’s decision to target industrial applications rather than another humanoid suggests he learned lessons from Optimus. General intelligence in physical form remains distant. Narrow intelligence applied to valuable tasks can succeed today.

Investors appear willing to back that thesis. The same firms that funded Mytra and Proception now watch Kumar’s progress. Y Combinator, First Round Capital, and others have placed bets across the ex-Tesla robotics diaspora. European efforts like UMA, founded by former Tesla Autopilot and Optimus AI engineer Rémi Cadène, show the trend extends beyond U.S. borders. Electrek covered UMA’s lightweight Northstar robot in July. The company targets European factories first and already talks to 50 potential customers.

Back in Silicon Valley, Kumar’s Intelligent Machines joins a growing cohort. The question isn’t whether Tesla alumni will build competing robots. It’s how their accumulated know-how will reshape factory floors. They understand the pain of ramping complex automation at volume. They know where current hardware and software fall short. And they carry hard-won insights from one of the most aggressive robotics programs on the planet.

Whether Intelligent Machines ultimately raises the targeted $100 million remains to be seen. What is clear is the continued flow of talent. Tesla built a formidable robotics brain trust. Now pieces of that trust are spinning out to attack adjacent problems with fresh capital and fewer constraints. The results could accelerate industrial automation even if Optimus itself takes longer than hoped to reach its full promise.

One thing seems certain. The next wave of factory robots may owe as much to lessons learned inside Tesla as to any university lab or traditional automation giant. Kumar and his peers left with more than just resumes. They left with battle-tested convictions about what works, what doesn’t, and where the real money lies in robotics right now. Factories, not homes. Specific tasks, not sci-fi versatility. Execution today, not generalized intelligence tomorrow.