Vertical vs. Horizontal Integration: Choosing Where to Compete
- vpeng2
- 3 days ago
- 3 min read
By: Stefan Seltz-Axmacher, CEO & Co-Founder, Polymath Robotics
Two Common Myths
Founders often boast about their technology being “full stack,” as if building everything in-house is a badge of honor. (This is particularly true in robotics, where you’ll see some companies build everything from scratch, down to the actual computer at the base.) Others assume the only way to scale is to build as broadly as possible, touching every market at once. Both are myths. For industries like mining, where integration decisions impact safety, uptime, and ROI, these myths can be costly.
In reality:
Vertical integration can be powerful, but only when every piece of the stack is truly critical to delivering value.
Horizontal plays can be resilient, but only if the layer you build is genuinely reusable across markets.
Understanding when to go vertical and when to stay horizontal is the difference between building a company and building a cautionary tale.
Vertical Integration in Practice
From Polymath’s founder & CEO, Stefan Seltz-Axmacher:
At my first company, Starsky Robotics, we went completely vertical. We retrofitted trucks, wrote teleop and autonomy software, built safety systems, ran fleet operations, and even became our own trucking company. The theory was simple: control everything, maximize performance.
The reality was more sobering: the weakest piece of the stack became the weakest link of the company. A lean engineering team was suddenly responsible for doing the jobs of dozens of companies, and “world class” quickly turned into “spread too thin.”
Investors wanted the story of full autonomy. But while shaving the last 0.5% of human intervention might cost billions, a single data scientist could improve pricing models enough to generate thousands in additional monthly revenue per truck. The “obvious” frontier problem was not the most valuable one.
What Customers Actually Value
Customers do not reward effort; they reward outcomes.
A grower cares about cost per strawberry picked, not how many localization vendors were tested.
A mining executive may choose one provider over another because the dashboard is clearer, not because the perception stack has one more research paper behind it.
The danger of vertical integration is forgetting that engineering miracles rarely matter if they don’t move the customer’s needle.
Buy First, Build Only What’s Missing
A simple principle: buy unless it truly does not exist.
In agriculture, you can buy arms, grippers, and vision models. What doesn’t exist yet is reliable coordination between multiple arms working in the same row. That’s where building makes sense.
In mining, you can buy individual collision avoidance systems and proximity detection hardware, but if comprehensive safety coordination across mixed autonomous and manual operations isn't available as a standalone product, building that safety orchestration layer becomes necessary.
In our own work, we tried to buy a robust IP-controlled remote start for vehicles. Nothing suitable existed. Reluctantly, we built it. That gap became a wedge.
If nothing in your stack can be bought, you are not in product. You are in R&D. That requires a different mindset, timeline, and funding model.
Knowing When to Let Go
Many teams cling to internal tools long after better options appear. The right discipline is to kill them early.
Testing a $5,000 sensor for a week is cheaper than tying up engineers for months to maintain a less capable homegrown solution. Switching hurts in the short term, but the cost of delay is almost always higher where custom parts might mean longer lead times and thus, unacceptable levels of downtime.
The Case for Horizontal
At Polymath, we deliberately chose a horizontal strategy: one autonomy core, many vehicles. That focus allows:
A larger share of the team devoted to autonomy rather than hardware or operations.
Reusability across industries without rewriting the stack.
Investment in unglamorous but essential infrastructure – for example, reliable over-the-air updates across fleets with unreliable networks.
Horizontal integration works only if the layer is truly common. If every deployment requires its own fork, you are effectively doing vertical five times.
A Practical Framework
Identify the outcome your customer values. Tie every decision to that metric.
Default to buy. Only build what cannot be purchased.
Build for differentiation. Focus your limited resources where the gap directly affects customer value.
Continuously test and replace. Kill in-house tools the moment a viable option appears.
Minimize the need for miracles. Robotics already has enough of them.
Conclusion
Vertical and horizontal integration are not identities. They are tools. Vertical makes sense when control is essential and no alternatives exist. Horizontal makes sense when a reusable layer spans industries.
The mistake is treating either as a philosophy. The right choice depends not on what feels boldest, but on where the company can actually deliver unique, defensible value.
For more on this topic, listen to Polymath Robotics’ podcast episode on Spotify, where the team discusses integration strategies shaping autonomy across industries.
