Ideas
Humanoids are a multi-decade trillion-dollar opportunity.
Andrew Kang believes robotics/physical AI is at a pre-ChatGPT inflection point and is underappreciated by investors. He points to early robot foundation models (Google DeepMind's RT-1/RT-2) showing generalizability, which can extend to simple factory tasks. He expects tens of billions of industry revenue within 2-4 years and sees a multi-decade, multi-trillion-dollar opportunity if humanoids reach human-like physical labor economics. He is focusing all his time and capital here.
Industrial automation robotics is a huge market.
Andrew argues the first killer application for humanoids/physical AI is simple industrial automation: pick-and-place, packaging, mixed-case palletizing, machine tending, and shelf stocking. These tasks are repetitive and poorly suited to deterministic automation because factory layouts, SKUs, and environments change. He estimates industrial automation alone can be tens of billions in revenue, with robots working 24/7 at low marginal cost to solve labor gaps.
Figure is top humanoid robotics investment.
Andrew's highest-conviction humanoid robotics company. He invested $19M before speaking to the founder and before the major markup. He views Figure as a multi-trillion-dollar future company, with no US humanoid competitor close except Tesla Optimus. He emphasizes Figure's vertically integrated approach—building intelligence, hardware, software, and manufacturing—which enables a real-world data flywheel from deployed robots, and its team recruited top talent from Boston Dynamics, Google DeepMind, and Tesla Optimus.
BOT offers public robotics venture exposure.
Andrew founded Robo Strategy (Nasdaq: BOT), a publicly traded closed-end vehicle, to give public-market investors access to high-quality private robotics and AI assets. He argues public capital markets can raise large sums faster than private venture and can value these assets at a premium to private NAV because the public buyer pool is much larger. He plans a barbell portfolio: 6-10 core companies around 70% and earlier-stage venture around 20-30%, using public-market scale to challenge SoftBank in robotics investing.
Path Robotics automates manual welding market.
Path Robotics builds application-specific robots for welding, a market that is about 80% manual and 20% automated despite tens of billions paid to welders annually and a welder shortage. Welding is hard to automate due to variations in metal, shapes, environment, and fixturing. Path has collected a large proprietary dataset of welding successes and failures, which is difficult and expensive to replicate because failures require scrapping metal and working inside customer processes. Andrew sees this as a large, defensible business in a specific application.
Favor vertically integrated robotics companies over model-only.
Andrew concentrates bets on vertically integrated robotics companies that build intelligence, hardware, software, and manufacturing. This gives more shots on goal: if the model layer monetizes, hardware/manufacturing can differentiate. He argues robot on-root data is necessary, third-party hardware creates reliability, support, supply-chain, and data-transparency problems, and the best results come from co-developing model and hardware. Model-only companies are now trying to build hardware but lack expertise and time.
Avoid robotics data-collection companies.
Andrew is skeptical of the many startups trying to become the 'Scale AI of robotics' by collecting egocentric video or robot data. He questions how much data is needed (millions vs billions of hours) and what happens when robots learn continuously/self-play, which could commoditize the data layer. He sees it as a possible short-term cash cow with a revenue/valuation spike followed by a decline over 2-5 years, not a durable decade-long investment.
US robotics companies are best investments.
Andrew believes America is the place to invest in robotics, even if China leads in hardware manufacturing. US companies have stronger capital markets and much higher margins (20-50%+), allowing them to generate larger profits on lower volume. Proposed bipartisan legislation to restrict Chinese robots in America is likely to pass, protecting the largest market for domestic robotics companies.
Avoid Chinese robotics companies' shareholder returns.
Andrew is less interested in Chinese robotics companies from a shareholder perspective. Although Chinese companies may build impressive hardware and benefit from manufacturing scale, they often operate at thin margins and lack the capital-market valuation upside of US peers. He passed on Chinese robotics investments because the returns were not compelling enough to offset the risk of a US public company investing in potentially dual-use technology. He still expects good outcomes but not the multi-trillion-dollar outcomes he seeks.
Avoid model-only robotics companies.
Andrew has become more bearish on model-only robotics companies. He believes open-source models will commoditize closed-source model layers, and it is unclear how model-only companies monetize without hardware. The end goal of robot foundation models is continual self-improvement, which could reduce reliance on external data vendors. Vertically integrated companies have a structural advantage.
Open-source robot models will commoditize closed models.
Andrew has become more bullish on open-source robot foundation models. He sees open-source models commoditizing closed-source models, similar to what is happening in LLMs (e.g., Kimi approaching frontier benchmarks), and believes this will happen in robotics. This favors hardware companies with manufacturing and supply-chain strength, especially if Chinese robots are restricted in the US.
This Delphi Digital video, published July 31, 2026,
features Andrew Kang
discussing Humanoid robotics, Physical AI, BOTZ, FIGR, BOT, Path Robotics, Vertically integrated robotics companies, Robotics data collection companies, US robotics companies, Chinese robotics companies, Model-only robotics companies, Open-source robot foundation models.
11 trade ideas extracted by AI with direction and confidence scoring.
Speakers:
Andrew Kang
· Tickers:
Humanoid robotics,
Physical AI,
BOTZ,
FIGR,
BOT,
Path Robotics,
Vertically integrated robotics companies,
Robotics data collection companies,
US robotics companies,
Chinese robotics companies,
Model-only robotics companies,
Open-source robot foundation models
