Former Tesla Optimus Lead is Launching an Industrial Robot Startup

Ashish Kumar, a former AI lead for Tesla’s Optimus humanoid robots, has co-founded a startup developing non-humanoid robots that handle a variety of tasks that would normally be done by highly skilled workers in industries such as manufacturing and aerospace, according to two people with direct knowledge of the matter.

The San Francisco startup, Intelligent Machines, has hired several former members of Tesla’s Optimus team and is in the process of raising a seed round that could end up around $100 million, said one of the people.

Intelligent Machines is currently focused on developing software for robots, but its leaders have discussed designing the hardware as well.

The company uses reinforcement learning—a method that rewards AI for accomplishing certain goals and penalizes it for other actions—to teach robots to perform complex tasks, according to the two people. (Kumar’s doctoral research at the University of California, Berkeley, focused on RL.)

While RL has long been used to train robots, the methods have mostly been limited to teaching robots particular skills such as walking or playing ping pong. Recently, robotics developers have started to use RL techniques for more practical tasks—for instance, rewarding robots for successfully plugging in cables or folding napkins.

Kumar believes the latest AI models for robots can’t produce the precise control necessary to automate physical labor in industries like manufacturing, said one of the people.

It isn’t clear whether Intelligent Machines is relying on vision-language-action models, which are essentially derivatives of the large language models powering AI chatbots that have been trained to control robots. But we know that its approach differs from startups developing so-called world models, which are trained on videos to understand the laws of physics and simulate real-world environments in which a physical robot might operate. (See this prior column about world models versus VLAs.)

World model developers include General Intuition, which last month raised a $220 million funding round at a valuation of $6.2 billion, and World Labs, which chip maker AMD is in the process of acquiring for around $8.2 billion.

While investors continue to pour money into robotics startups, AI models for robots haven't been capable of triggering a “ChatGPT moment” in which a variety of robots can handle a wide range of physical tasks. In an interview last month, venture capitalist Vinod Khosla said that while he thinks this will happen in the next two years, he also predicted that more than half of robotics startups will see their valuations drop by 2030.

While Intelligent Machines’ robot prototypes are similar in some ways to Optimus robots, they are not humanoids and represent a “significant departure” from the way Optimus robots rely on AI, said one of the people with knowledge of the startup’s work.

Tesla has faced engineering challenges in developing the Optimus robot’s hands that have led to production delays and hampered its efforts to scale up production of the robots. Kumar spent more than two years at Tesla—including a year as AI lead for the Optimus team—before departing a year ago. (He then spent four months as a research scientist at Meta, according to his LinkedIn.) One of Kumar’s former colleagues on the Optimus team, Kamal Gupta, announced his departure from Tesla on LinkedIn over the summer.

Intelligent Machines is the latest of several startups founded by former Tesla employees. DensityAI, an AI chip startup founded last by former leaders of Tesla’s Dojo supercomputer program, is in advanced talks to raise hundreds of millions of dollars in a round that would value it at $10 billion, The Information reported last month.

Here’s what else is going on…

OpenAI Does the Math

A deluge of new math papers from OpenAI on Tuesday highlighted the intensifying struggle between AI innovators and leading scholars in the field of mathematics, which helped make AI possible.

The company published a whopping 722 new research papers detailing solutions on a range of problems across math subjects. That came less than a month after it reported solving the vaunted “Navier-Stokes existence and smoothness problem,” setting off a backlash and something of an identity crisis within mathematics.

This extends a pattern in which the AI labs—and OpenAI in particular—announce math breakthroughs that have little evident commercial value other than bragging rights but manage to upset many mathematicians.

The math disclosures have been partly influenced by OpenAI’s core AI researchers, who develop models using an array of math. Some of these researchers have sought to publish as much as possible about the math breakthroughs so that real-world mathematicians, including students considering whether to pursue doctoral degrees, can take stock of how AI will alter the field, according to an OpenAI employee.

Math scholars are certainly paying attention. After the Navier-Stokes breakthrough, 28 recipients of the Fields Medal—one of mathematics’ highest honors—published an open letter arguing that simply producing the answers to hard math problems with AI undermines the real value of the work solving such problems, which is to advance human understanding and insight. “The push by AI companies to solve mathematical problems as a benchmark is detrimental to the science of mathematics, and to the mathematical community,” they wrote.

OpenAI responded to the uproar last month by saying it would work with an independent body called the Advisory Group on Mathematics and AI, whose recommendations the company said helped inform how it presented its new research Tuesday.

The advisory group, in a statement on its website, said it appreciated the company’s engagement, but it also sounded a warning. “The future of mathematical research cannot consist only of understanding results produced by AI labs,” the post said. “Mathematicians must be able to formulate their own questions, develop their own approaches, and explore directions that have not been selected as examples of an AI system’s capabilities.”—Jason Dean

Big Number

SpaceX, the AI and rocket company led by Elon Musk, is seeking to raise $40 billion to buy Nvidia chips in a financing round led by Apollo Global Management, the Financial Times reported on Tuesday.

Overheard

Anthropic said it is expanding and restructuring its cybersecurity access programs to enable more cyberdefenders to use its most advanced Claude models to secure their systems against potential AI attacks.

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

Waymo increased its first-ever private debt offering to $5 billion, up from a previously planned $3 billion-plus, as the robotaxi operator faces rising AI and fleet-procurement costs.

Google entered a 20-year power purchase agreement for 3,590 megawatts from Constellation Energy enabling more than $4.3 billion of new investment by Constellation to upgrade 11 nuclear units across Illinois, Pennsylvania and New Jersey.

Lambda, a cloud provider for AI computing, is raising up to $4 billion at a $14.5 billion pre-money valuation in what could be its last private funding round before a planned 2027 IPO, led by Coatue Management and Blackstone, The Wall Street Journal reported.

Vinci, which uses AI to simulate chip and hardware physics, raised $250 million in a Series B funding round at a $1.5 billion valuation led by Advent, Temasek and Xora Innovation.

Multiply Labs, which builds robotic automation systems for biomanufacturing, raised $75 million in a Series B funding round led by NantWorks.

Turba Labs, whose software squeezes more computing capacity out of existing AI infrastructure, raised $52 million across seed and Series A funding led by Creandum and Cusp Capital.

Hadrian, which builds an agentic AI offensive-security platform, raised $40 million in a funding round led by Forgepoint Capital International and Smartfin.

Vitalize, which makes an AI operating system for hospital staffing and scheduling, raised $31 million in a Series A funding round led by Oak HC/FT.

Flai, which makes AI agents for car dealerships, raised $27 million in a Series A funding round led by Base10 Partners.

WhiteLab Genomics, which uses AI to design genomic medicines, raised $26 million in a Series B funding round led by AVP.

Melius, which makes AI agents that handle ad-campaign and video production, raised $25 million in total funding, including a $20 million Series A led by CRV and a $5 million seed round led by General Catalyst.

Procuros, which builds an AI-native connectivity layer for B2B supply chains, raised €20 million (about $22 million) in a Series A funding round led by XAnge.

Siena, which makes an AI-driven customer experience platform for consumer brands, raised $17 million in a Series A funding round led by York IE.

Avarra, which makes an AI sales-automation platform, raised $17 million in a Series A funding round led by Duration Ventures.

Kuaishou's AI video unit Kling AI has selected CICC, Goldman Sachs and UBS to lead a Hong Kong IPO targeting at least $1 billion, as early as 2027, Bloomberg reported.

SAP agreed to acquire TechWolf, a Belgian company whose AI maps employees’ skills and tasks for workforce planning, to add the technology to SAP SuccessFactors.

Serval, an AI-native enterprise-service-management company, acquired Ensignia, a software supply-chain security startup, to add verification controls around the autonomous agents Serval deploys across IT, HR, finance and security workflows.

Mistral released Mistral Large 4, a new large multimodal model. The model is currently only accessible through a public guardrail endpoint, but the company said it plans to make the model’s weights public in 3 three weeks after testing.

Google released an updated image model Nano Banana 2.1 built on Gemini 3.6 Flash, saying it improves visual quality and cuts prices by around half from Google’s previous image generation model.

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.