Full-Stack Robotics

TL;DR

These startups sell an end product or service, making money based on outcomes or output

Companies

Intuitive Surgical, SpaceX, AutoStore, Ocado, Zipline, Anduril, Field AI, Earth AI, Cuby, Freeform

Compound portfolio companies Wayve and AIFleet

Overview

Pros

• Often faster to first product because operating the robotics internally means it’s far easier to have technical crutches behind the scenes like a human-in-the-loop are gradually removed
• Enables successive vertical integration of value chain and expansion to multiple products
• Multiple durable moats

Cons

• While a first product may be faster, the overall R&D effort is longer and more capital intensive
• Requires several core competencies (e.g. efficient production/manufacturing, R&D, product, R&D translation into product, GTM/sales) which all must be executed effectively
• Core limiter is thus fundraising rather than sales cycles

Value Capture: 4/5

Moat: 5/5

“Suppose you develop a new technology that is valuable to some industry. The old approach was to sell or license your technology to the existing companies in that industry. The new approach is to build a complete, end-to-end product or service that bypasses existing companies.”

Full-stack robotics startups can either deliver a product (e.g. satellite-based internet, manufactured parts) or service (e.g. robotic restaurant, surgery). The extent to which one is full-stack is fluid. AutoStore currently stops before owning demand because it takes far too long to build scaled demand in grocery, logistics, etc., while Field AI doesn’t make its own robots as that’s not a necessary core competency. Even SpaceX started out as a contractor for NASA and still runs errands for other satellite companies.

The last example speaks to the long-term opportunity for full-stack robotics companies to redefine who the end customer is (integrating up the value chain) and expand into multiple adjacent products. Product expansion is generally more straight forward than with RaaS because full-stack startups sell an end product/service. Whereas RaaS startups sell a robot serving an intermediary good/point solution which usually would mean selling a new kind of robot to the same customer. Somebody needs an end product, not everybody needs multiple robots.

As Mike said in On Inflection Points:

The important part to understand when building a company that takes advantage of domain progression is the scalability and transferability of your technology over time. While SpaceX can innovate over time, improving the economics of their first domain (launch services), eventually reaching scale to invade new domains, there isn’t material wasted R&D that doesn’t capture value along the way, as each innovation improves their core domain use-case.

Or, as the CEO of Compound portfolio company Runway puts it:

SpaceX's advantage is not just knowing how systems connect. It's having the institutional memory of 10,000 failed experiments that no competitor can replicate. Every custom component represents hundreds of iterations that taught them something essential. When Blue Origin buys off-the-shelf components, they're buying someone else's assumptions. SpaceX's value lies in knowing why each bolt matters, how each system interlocks, and having the capability to go arbitrarily deep when needed. They make their own bolts not because they must, but because understanding requires building. Knowledge compounds through integration. You should only buy or integrate tools when they don't interfere with your organizational learning rate or disrupt your long-term assumptions. The value in AI isn't headcount or compute alone. It's organizational knowledge and custom infrastructure that lets you test the right ideas in minutes, not months, and iterate furiously. Organizational knowledge is the product. The speed of iteration is the moat.

The production process, capacity and iteration infrastructure that full-stack startups craft after a few years are the largest long-term moats they have. The operational efficiency and iteration speed of the startups’ internal processes are what enables them to win out over incumbent’s scaled distribution and earn higher operational leverage at maturity.

However, all this requires the rare team that can juggle multiple core competencies simultaneously and execute on all of them well. More parts of the business must be built before the startup achieve PMF and growth inflection points.

This forces full-stack startups to raise more capital with less traction than peers. Past demonstrations of capital efficiency, product/technical progress, and demand help alleviate this most critical challenge.

The other side of the same coin means they get greater value capture and multiple, highly durable moats at maturity.

Taken to their extremes, these companies are often the era defining ones, Schelling Point companies.

Not coincidentally, SpaceX, Intuitive Surgical, DJI, and AutoStore are basically the only robotics companies that have achieved >20% EBITDA margin. Or, as Peter Thiel argued, there have only been two types of businesses that have captured massive value over the last 250 years: vertically integrated complex monopolies (Ford, AT&T, Tesla, SpaceX, etc.) and software.

Quick breakdown of full-stack vs RaaS approaches

Further Reading

https://mhdempsey.substack.com/p/full-stack-deep-tech-startups-and

https://www.michaeldempsey.me/blog/2024/06/20/robotics-fomo-scaling-laws-technology-forecasting/

https://www.notboring.co/p/vertical-integrators

https://hardwareishard.com/freeform

https://www.michaeldempsey.me/blog/2025/10/03/sequencing-vs-equal-odds-applied-research/

Always remember that SpaceX achieved its first successful launch on ~$90M