Should we build it ourselves, or depend on a partner?

The question behind the question

The make-or-buy decision looks like a comparison of costs: what would it cost to build, against what a partner charges. The record suggests that costs are rarely what decides the outcome. What decides it is who learns, who controls the future of the capability, and whether the partner’s interests will stay aligned with yours once you depend on them.

Oliver Williamson’s work on transaction costs, discussed in the site’s idea Why Firms Exist, explains the danger. When one side must invest in something valuable only within a relationship, it becomes exposed to the other side’s opportunism. The traditional remedy is to bring the activity inside the firm. The modern remedy, where it works, is a partner that has bound itself not to exploit the dependence.

When owning it worked

TSMC’s history is a sequence of decisions to keep learning inside. In 2000 it declined IBM’s offer to share its copper wiring process and developed its own; owning the process meant owning every improvement, in an industry where cost falls with accumulated experience. In 2012 it paid €838 million for a stake in ASML and funded its research into extreme ultraviolet lithography, keeping close to the one supplier whose breakthrough it would need. Google, after years of licensing street data from Tele Atlas, began building its own map in 2008 with Street View cars and human editors, removing a dependency on a licensor that had been bought by a competitor and making free navigation possible.

When depending on a partner backfired

In 2000 Yahoo, then the web’s leading portal, chose to use Google’s search results on its site, with Google credited on the page. It saved Yahoo the cost of search at the moment search was becoming the internet’s most valuable activity, and taught millions of Yahoo users to use Google. In 2011 Nokia announced it would abandon its own smartphone platforms and build phones on Microsoft’s Windows Phone, a decision prepared by its chief executive’s “burning platform” memo. It handed the software, and with it the future of the product, to a partner whose platform had little support from app developers, and Nokia’s phone business did not recover.

Apple’s maps show that taking a capability back has its own risks. Apple was right that depending on Google, a rival in phones, for maps on the iPhone was strategically weak. It replaced Google’s maps in 2012 with its own, assembled from licensed data, before that data was good enough, and the errors forced a public apology.

When depending on a partner worked

The counterpart to these failures is TSMC’s own founding decision: it promised never to design or sell chips under its own name. That promise made TSMC a partner chip designers could depend on without fear, which is why companies from Nvidia to Apple outsourced their manufacturing to it. Dependence is safe when the partner has structured its business so that competing with you cannot pay.

A diagnosis to start from

Three facts carry most of the weight. Does the capability improve with use, so that whoever performs it learns and pulls ahead? Could the provider become a competitor, or does it already serve your competitors? And how reversible is the dependence: how long would it take to rebuild the capability if the relationship failed?

The guiding policy is to own capabilities that compound and that sit close to what customers value, and to buy the rest from partners whose incentives are bound to yours. The coherent actions include building internal capability before switching away from a partner, as Apple learned the hard way, and writing the partner’s commitments into contracts or structures rather than relying on goodwill.

The strongest counter-case

The strongest objection is the success of outsourcing in the same industries. Apple designs chips and outsources their manufacture to TSMC; Nvidia, AMD and Qualcomm own no factories. Many of the most valuable technology companies are built on depending on partners for capabilities they could never afford to own. Gary Pisano and Willy Shih argued in 2009 that such outsourcing can hollow out the industrial commons of a country, but for the individual firm it has often been the right choice.

The objection is right, and it shows that the decision turns on the partner’s position, not on whether outsourcing is good in general. Fabless chip designers depend on a foundry that has promised not to compete and that serves hundreds of customers; Yahoo depended on a search company whose own product competed directly for its users.

What the wider research says

Kenneth Arrow’s 1962 paper on learning by doing explains why owning a capability that improves with use matters: knowledge is a by-product of production, and those who produce learn. Christensen and Raynor argued in The Innovator’s Solution that integration wins when products are not yet good enough and the interfaces between components are not well understood, while modular outsourcing wins once performance exceeds what customers need and interfaces are standard. The record fits that view: the capabilities worth owning were those still improving fast.