What Every MBA Should Know About Humanoid Robots

By: Shashank Mahajan

By the end of this article, I want you to be able to answer a serious question: Does it make sense for Kellogg to buy a Humanoid Robot to kick someone out of a study room (hot commodity!) if they’ve overstayed their reservation?  

Put your thinking hats on, it’s your time to shine! 

Caught Between Optimism and Obsolescence 

Created with AI

I’m still wrestling with where I fall in the AI debate. Do I want to be naive and say that AI is great for coding and knocking out my Fin2 assignments, but that’s it!? Or do I want to be a bearer of doom and declare that the only professions safe are plumbing, carpentry or anything that requires physical labor? Maybe neither. 

Instead of choosing a side, I decided to get closer to it. I’ve wanted to explore the humanoid space for a while, and I used my VC internship as a way to immerse myself in robotics. The role revolved around sourcing robotics and humanoid startups, hearing founders pitch, and dissecting their product, business model, and differentiators.

. 

So, what should every MBA really know about humanoid robots? 

Some of us will work in manufacturing, others will consult those who do, and a few may even acquire plants with the goal of automating the shop floor or digitally transforming operations. A smaller subset might end up building or running these systems on the product or ops side (like me). It’s tempting to imagine a humanoid simply replacing a factory worker, but instinctively we all run the same mental math: does the cost-benefit actually work? And as robots begin to operate alongside humans—in labs, warehouses, factories, or homes—questions of safety become unavoidable. How do we design the systems and guardrails that make humanoids reliable collaborators rather than liabilities (figuratively and literally)? 

Here are three "buckets” worth carrying with you (if they are not MECE, know that I don’t care): 

1. Humanoid robots are capital intensive 

At roughly $80K+ per humanoid in the U.S., the math rarely works today if you’re chasing near-term productivity gains of the kind Prof. Sébastien Martin talked about in Ops. That may change—Morgan Stanley projects as many as 1 billion humanoids by 2050, largely in manufacturing and logistics, which would push costs down as the category commoditizes. Until then, the more realistic option for operators and finance teams is RaaS (Robot-as-a-Service): hiring a robot rather than owning one, shifting upfront capital risk into operating expense while still experimenting with automation. 

2. A case for legs

When a founder pitches you a humanoid, ask: 'Why not a robotic arm or a wheeled robot?' This question is so trivial but, in the frenzy, it’s easier to get bogged down by cool tech. Humanoids only become compelling when they operate in human-designed environments—stairs, ladders, cluttered floors, and unstructured layouts. Amazon’s warehouses work with wheeled robots because Amazon re-architected the environment around them. Most factories and labs won’t have that luxury. If the environment can’t or won’t change, legs start to make sense.

3. Hardware is not a bottleneck; dexterity and perception are 

The acrobatics we see at CES (the Consumer Electronic Show) are, at the end of the day, just demos. What actually limits humanoids today is brain power—which might be surprising, given all the AI hype. Real intelligence looks unsexy: telling a glass cup from a ceramic one, applying just enough force to pick it up (if only my high-school self knew physics would show up like this), and planning a sequence of actions in real time. LLMs (Large Language Models) have given us a false sense of assurance that training on internet-scale text is comparatively easier (in relative terms only. I’m an MBAi and by no means do I want to signal that any of it is easy). Training VLAs (Visual Language Action Models) isn’t. Videos don’t encode force or touch, and real-world data collection—which might look like strapping cameras on people doing chores—is slow, expensive, and frankly absurd (will this spawn new jobs after OnlyFans?). In the near term, teleoperation and imitation learning, backed by simulation, will lead the way. 

So should Kellogg buy a humanoid robot? Probably not. Turns out a polite 'your time's up' and some Kellogg kindness accomplishes the same thing for $80K less and works on the Spanish Steps. 

If you want to hear me geek out more on humanoids, check out the Robotics Report I co-authored recently.

Read More Tech Talks: Data, Deals, and Government. What’s New with Nuclear? The Wall-less Empire: The Illusion of Open AI Markets The SpaceX Diaspora: How Rocket Science Is Transforming Startup Culture