Direct answer
Vertical integration means owning a stage of your business that you used to buy. For decades it paid off only at great scale, which is why the famous examples are companies like Tesla and SpaceX.
AI changes that math. It lowers the cost of building and running your own software. A company with €10 million to €100 million in annual revenue can now own infrastructure that used to be out of reach.
It does not make sense for every function. It makes sense for the functions your business depends on. When it fits, it pays twice: your costs go down, and you build a second company with a value of its own.
What vertical integration looked like before AI
Tesla is my favorite example. It does not sell through franchised dealers. It sells through its own website and its own stores, and every purchase is completed online.
Since 2023, Tesla has let customers in some European markets book a test drive in its app. They unlock the car with their phone and drive it without a salesperson.
Compare that with Volkswagen. Volkswagen sells through independent dealers. It tried an agency model for its electric cars in Europe from 2020, and it returned to the dealer model at the end of 2025.
The point is not that dealers are bad. The point is that Tesla took a cost driver, the showroom, and rebuilt it around its own technology. That only works because Tesla owns that stage. A brand that sells through someone else’s showroom cannot redesign it the same way.
This is what I mean by vertical integration: building the infrastructure your business depends on yourself. It gives you room to move that no supplier contract will. For most of history it also took a lot of capital and scale.
What AI changes
For a one-person business, building your own software is almost never worth it. Paying €40 a month for someone else’s product is the better deal.
For a company with €10 million to €100 million in revenue, building was usually not worth it either. It was hard, slow and risky. The default was to buy an ERP system and pay an outside firm to set it up.
AI lowers that bar. It makes software much cheaper to build and to run, and AI agents can take over much of the administrative work around it. Scale still matters, but the threshold is lower than it has ever been.
Example: a niche manufacturer
Take a German manufacturer of complex parts. Its business depends on production, but production is not its only cost driver.
Its internal sales team takes calls about orders and builds custom orders. It checks stock, calls production and estimates how long each order will take.
The usual answer is to buy SAP and have an outside firm implement it. That works. If the company leads a niche, there is a second option: build its own technology company.
That company builds the ordering and machine management software for the parent. It can then supply the same software to the parent’s partners and suppliers in the same industry. If ten, twenty or a hundred of them already depend on the parent, they gain from being connected to it natively.
The deciding factor is whether the parent can guarantee the first customers. If it can, the new company is a startup with its customers signed on the day it incorporates. It does not need venture capital to survive, because it earns from the first day.
Example: a restaurant group
Take a holding company that owns 100 restaurants in one region, such as Berlin, or Cologne, Düsseldorf and Wuppertal. Every restaurant buys the same services: pest control, repairs, renovation and more.
If the group plans to keep buying restaurants, it can form service companies that are partly independent. They serve the group’s restaurants first, and they are free to serve other restaurants too.
Built AI-native, such a company runs with less administration and less middle management. A pest control company can write the report for the health authority automatically, so the technician does not have to. It can run booking, scheduling and route planning in one system. With the group’s restaurants as its base, it can undercut its competitors.
The money works differently too. Say the group pays a contractor €200 a month per restaurant. That money is gone.
Paid to its own company, part of it comes back as margin, perhaps €50, like cash back. And the company it built has a value on the balance sheet. It can grow on its own, be spun off or be sold. A contractor’s invoice never becomes an asset.
Where it works, and where it does not
It works best where a company leads a niche. That can be a global niche, like a mid-sized manufacturer that leads the world in one ingredient. It can be a local market, like restaurants in Berlin or Dublin. In both cases, the company already has the demand that a new service company needs.
It works less well in commodity businesses such as dropshipping, where there is little to own.
Some functions will not make sense at your current scale. Integrating the right ones first can grow you into the scale where the others do.
How to decide, function by function
- List every function in the company.
- Mark the core functions.
- Record how each function runs today: outsourced or in-house, and how well.
- Estimate the direct savings: if it were built today, what would it save in the first year?
- Estimate the growth case: what could it open up? More acquisitions, for example, or a new business such as pest control or waste management.
- Estimate the technical cost of building it.
- Start with one pilot, and bring in one function at a time.
Get the core right first. It makes no sense to start a pest control business before your sales and marketing work. Then integrate one function at a time, where it makes sense.
Key takeaways
- Vertical integration used to need the scale of a Tesla. AI lowers that bar.
- The best candidates are the functions your business depends on, in a niche you lead.
- An in-house service company pays twice: part of the cost comes back as margin, and the company has a value of its own.
- Guaranteed first customers, your own company and its partners, remove most of the risk.
- Decide function by function, on savings, growth case and technical cost, and start with one pilot.
Sources
- Wikipedia, “Tesla US dealership disputes” (Tesla sells through its own website and stores, not franchised dealers)
- Marketplace, “Tesla fought to sell cars direct to customers, and now more carmakers want in”, 22 April 2026
- Fleet News, “Tesla launches remote app-based test drive programme”, 2023, and “Tesla expands self-serve test drive programme”
- electrive, “VW abandons agency model across Europe”, 22 December 2025
Questions this essay answers
What is vertical integration?
Vertical integration means owning a stage of your business that you used to buy from someone else. Tesla sells its cars through its own website and stores instead of franchised dealers, so it can redesign the whole sale around its own technology.
Why did vertical integration need so much scale?
Building and running your own infrastructure was hard, slow and risky. For a small business, paying for someone else's product was the better deal. Even a company with €10 million to €100 million in revenue usually bought an ERP system and paid an integrator.
How does AI change vertical integration?
AI makes it much cheaper to build and run your own software, and AI agents can take over much of the administrative work around it. Scale still matters, but the threshold is lower than it has ever been.
Should a mid-sized company build its own software company?
It can make sense when the company leads a niche and can guarantee the first customers, for example its own operations plus the partners and suppliers that already depend on it. Then the new company starts with customers on the day it incorporates.
What is the financial case for an in-house service company?
Money paid to a contractor is gone. Money paid to your own company partly comes back as margin, and the company itself has a value on the balance sheet. It can grow on its own, be spun off or be sold.
How do you decide which functions to bring in-house?
List every function, mark the core ones, record how each runs today, and estimate its first-year savings, its growth case and its technical cost. Get the core right first, then start with one pilot, one function at a time.