Is Robotics The Next Megabubble?

Andrew Kang of Mechanism Capital argues that robotics is a potential ten trillion dollar industry and that the key historical barrier was never hardware design but intelligence. He traces the inflection point to ChatGPT's emergence in 2022, which signaled that digital AGI was achievable within a reasonable timeframe and implied physical AGI would follow. When Kang entered the space in late 2023 or early 2024, the total market cap of all private robotics companies was approximately twenty billion dollars. That figure has since risen to roughly one hundred to two hundred billion dollars, and Kang believes the industry still has one hundred to potentially one thousand times upside remaining.

The economic case for humanoid robots rests on cost comparisons with human labor. An economy model humanoid robot is estimated at twenty thousand dollars and a premium model at fifty thousand dollars, while a personal assistant in America costs forty to fifty thousand dollars per year and an executive assistant can run two hundred thousand to four hundred thousand dollars all-in annually. The humanoid form factor is described as the most versatile because robots can go anywhere a human can go and beyond, with a prediction that by approximately 2035 humanoid robots could serve as personal assistants performing real-world tasks and become a status symbol comparable to the iPhone.

The current state of robotics intelligence is described as roughly equivalent to GPT-3 level capability in large language models, with the qualification that GPT-3 level performance in a physical robot is not visually impressive because a robot picking up a cup only half the time does not generate the same public reaction as a fluent chatbot. A prediction is made that a GPT-5 equivalent breakthrough in robotics will arrive in approximately one year, faster than the two years it took to go from GPT-3 to GPT-5 in language models. The reasoning is that key research learnings from scaling LLMs, including reinforcement learning from human feedback, mid-training structure, and data annotation infrastructure, can be applied directly to compress the robotics development timeline. The speakers are careful to note that connecting an existing LLM such as Claude or ChatGPT to a robot does not work because those models lack knowledge of physical properties such as pressure, grip force, physics, spatial semantics, and affordances.

Portfolio companies discussed include Figure AI, described as comparable to Apple for its design sensibility and user experience focus. Apptronik is described as roughly one to two years behind Figure AI and holds partnerships with a large-scale US hardware manufacturer and Google DeepMind. Standard Bots is described as the only US-based industrial arm manufacturer operating at scale and is noted as not reliant on China for key components, which is framed as strategically important. Dyna Robotics has achieved a 99.9 percent task success rate in post-training research, which is framed as critical because even a one to five percent failure rate is unacceptable for field-deployed robots. Daximate is identified as the first US company to commercially sell humanoid robots, having sold thousands last year primarily to major research labs, and that commercial availability enabled a developer ecosystem for collecting data and building robot skills.

Venture capital interest in robotics is described as having inflected meaningfully in the last three to four months at the time of recording, partly because a reassessment of software valuations caused investors to question the durability of software businesses and look toward physical AI. Tier one and tier two VCs are now actively looking at the space, though the level of attention is described as nowhere near current interest in AI companies. Valuations at which robotics companies are raising have already moved to multiples higher than just a few months prior. In public markets, meaningful robotics exposure is described as limited essentially to Tesla at approximately two trillion dollars in market cap, while the bulk of value creation is expected to occur in private markets.

Kang launched a public company called RoboStrategy that trades on the NASDAQ, inspired by MicroStrategy but applied to robotics venture investing rather than Bitcoin. The core mechanic is multiple arbitrage: private companies in sectors like aerospace and software trade at three to eight times multiples while comparable public companies trade at fifteen to forty times, so acquiring a private company at the lower multiple causes it to immediately rerate upward on public books, making each acquisition accretive to NAV per share. Public capital markets are approximately one hundred times larger than private capital markets, which defines the addressable demand for such a vehicle. Kang is targeting a scale comparable to SoftBank, reasoning that a ten trillion dollar robotics market would be approximately seven times the current Bitcoin market cap. He acknowledges that a compression in the premium to NAV, for example from four times to two times, would result in a fifty percent loss in market price for investors, and the fund will not issue shares at a discount to NAV.