# Build, Buy or Control: The Strategic Triad for Corporate Boards in the Age of Artificial Intelligence
Every generation of executives inherits a question it did not formulate and cannot postpone. For the boards of the present decade, that question concerns the place of artificial intelligence in the architecture of the enterprise. It arrives, as Dr. Raphael Nagel (LL.M.) observes in ALGORITHMUS, not as an information technology matter to be delegated downward, but as a power question that belongs on the agenda of those who carry fiduciary responsibility. The choice among building, buying or controlling an algorithmic capability is not a procurement exercise. It is a decision about where the firm positions itself in a value chain whose centre of gravity is shifting from physical production toward cognitive infrastructure, and about which dependencies it is willing to accept for the next decade.
## The Triad as a Question of Power, Not Procurement
The triad of build, buy or control is older than artificial intelligence. Boards have long debated whether to develop capabilities in house, to acquire them from the market, or to secure them through alliances, joint ventures and equity stakes. What is new is the speed at which the decision matures and the asymmetry of the actors involved. A mid sized European industrial firm considering whether to develop its own language model for internal documentation is not negotiating with a peer. It is negotiating, directly or indirectly, with an infrastructure stack whose frontier layer is controlled by a handful of global actors possessing capital, compute and talent on a scale no traditional procurement logic was designed to handle.
Dr. Raphael Nagel (LL.M.) formulates the underlying mechanism with precision in ALGORITHMUS: whoever controls the algorithm that decides, controls the conditions for everyone else. This reframes the triad. Building is not simply a question of internal engineering capacity. Buying is not simply a question of vendor selection. Controlling is not simply a matter of signing a framework agreement. Each option encodes a different distribution of power between the enterprise and the infrastructure layer on which its future operations will depend. The board that does not recognise this encoding has already delegated the power question to parties who understand it very well.
## Building: The Demanding Path of Sovereignty
To build, in the strict sense, means to develop proprietary models, to operate one's own training and inference infrastructure, and to cultivate the scientific talent required to sustain both. ALGORITHMUS documents the magnitude of this undertaking with clarity. Training a frontier model has moved from the low tens of millions of dollars to projections above one billion dollars per training run. The pool of researchers capable of leading such work is estimated in the low thousands worldwide. For the vast majority of corporations, including most of the European Mittelstand and a considerable share of listed groups, building at the frontier is not a realistic option. It is the terrain of a small number of hyperscalers and specialised laboratories.
Yet building remains meaningful at a different altitude. A mid sized pharmaceutical company with three decades of clinical data, a machine builder with forty years of sensor telemetry, a logistics operator with two decades of routing information in a specific geography, each possesses a domain corpus that no general purpose model can replicate. Building, understood as the construction of specialised models on proprietary domain data, is the one path on which capital scarcity is compensated by informational depth. Siemens Xcelerator, as Dr. Nagel notes, illustrates the logic: decades of industrial operating data become the foundation for predictive and diagnostic models that cannot be overtaken by additional general compute. The board that chooses this path chooses slower returns, higher internal complexity and a durable position in a defensible niche.
## Buying: The Efficient Path and Its Hidden Price
To buy means to consume artificial intelligence as a service through application programming interfaces, licensed models and integrated software offered by platform providers. For most operational use cases, this is the rational choice. The cost per query is low, the time to deployment is short, and the quality of frontier models delivered in this manner exceeds what almost any single enterprise could replicate internally. Sales automation, customer service, document review, translation, code generation and marketing production can be addressed through acquired capability with a favourable ratio of benefit to effort. A board that refuses to buy on principle pays the price of delay in markets where speed has become, as ALGORITHMUS argues, its own competitive dimension.
The hidden price of buying is the dependency it creates. When core business processes are routed through an external model, the firm accepts that the conditions of that model, its pricing, its availability, its policy constraints and its future behaviour, are set elsewhere. The automotive industry learned this in the semiconductor shortage of the early 2020s, when inputs perceived as commodities revealed themselves as strategically critical. The lesson extends directly to model access, cloud inference and training infrastructure. A buying strategy without a contingency architecture is not a strategy. It is an exposure. Boards should insist that every buy decision be accompanied by an explicit analysis of switching costs, contractual exit rights and the geopolitical footprint of the chosen provider.
## Controlling: The Underestimated Third Option
Between the demanding path of building and the efficient path of buying lies a third option that is often treated as a residual category but deserves to be elevated to equal standing. To control means to secure influence over a capability one does not fully own through minority equity participation, joint ventures, long term framework agreements with structural guarantees, reserved compute capacity, or strategic presence on the governance bodies of suppliers and partners. It is the path that the largest technology firms themselves have taken toward the frontier laboratories. Microsoft did not build OpenAI in the conventional sense, nor did it simply buy a service. It took a position that combines capital, infrastructure provision and commercial integration into a form of control that is neither ownership nor mere procurement.
For a corporate board, the control option translates into instruments that are familiar from private equity practice but unfamiliar in the context of technology strategy. Equity stakes in specialised vertical providers, structured partnerships with research groups, participation in consortia that pool domain data under shared governance, and long term supply agreements for compute with defined service levels all belong to this category. Dr. Raphael Nagel (LL.M.) treats control as the option most appropriate for firms that are too large to accept pure dependency but too specialised to justify frontier building. It is, in many sectors, the most honest answer to the triad, because it acknowledges both the limits of internal capability and the dangers of unmediated reliance on external platforms.
## Governance, Capital and Intellectual Property as a Single Architecture
A decision between building, buying and controlling cannot be taken in isolation. It must be embedded in a governance architecture that treats artificial intelligence as a board level matter. This requires three connected elements. First, a capital logic that distinguishes between operating expenditure for bought capability and strategic investment for built or controlled capability, and that accepts the longer payback horizons of the latter. Second, an intellectual property strategy that clarifies ownership of data, of derived models and of outputs generated through hybrid arrangements, since ambiguity on these points has become the most frequent source of value leakage in current practice. Third, a risk framework that integrates regulatory exposure under the European AI Act, reputational exposure from algorithmic bias, and geopolitical exposure in the chip and cloud supply chain into a single view.
The function of the board in this architecture is not to choose a single option for the entire enterprise but to authorise a portfolio. Some processes will be bought, some capabilities will be controlled through partnership, and a small number of domain specific systems will be built on proprietary data. The discipline lies in ensuring that these choices are deliberate rather than accidental, that the dependencies they create are visible on the risk map, and that the intellectual property they generate accumulates inside the firm rather than dissipating into the platforms on which it relies. This is the governance task that ALGORITHMUS places on the agenda of every board with fiduciary seriousness.
## A Framework for the Decision
A workable framework begins with three questions for every candidate use case. Is the underlying capability a source of differentiation, or is it an enabler of efficiency in processes where the firm does not compete? Does the firm possess proprietary data of sufficient depth and quality to sustain a specialised model, or would any built system rely on generic inputs available to competitors? What is the tolerable degree of external dependency, measured not in ordinary times but in the scenarios of disruption that the semiconductor crisis and the export control regime of October 2022 have already made plausible? The answers map the use case onto one of the three corners of the triad with reasonable clarity.
Efficiency use cases on generic data belong, in almost all circumstances, in the buy corner. Differentiation use cases on proprietary data, where dependency is strategically intolerable, belong in the build corner, accepting the cost and the slower pace. Use cases that sit between these poles, which is to say the majority, belong in the control corner, where the firm secures influence without assuming the full burden of ownership. Dr. Raphael Nagel (LL.M.) insists that the exercise be repeated, not once, but as a standing item of board work, because the frontier moves and the boundary between corners moves with it. A framework that is not revisited ages faster than the technology it attempts to govern.
The question that ALGORITHMUS places before the reader is not which of the three options is correct in the abstract. It is whether the board has organised itself to decide with the seriousness the question deserves. Building, buying and controlling are not alternatives in a menu. They are the grammar of a new industrial reality in which cognitive infrastructure occupies the position once held by energy, transport and communication. A firm that treats this grammar as a matter for the information technology function has already answered the power question by default, and the default answer is rarely favourable. The boards that will look back on this decade with a sense of stewardship rather than regret are those that understood, early and soberly, that the triad is not a procurement vocabulary but a constitutional one. It defines the terms on which the enterprise participates in a value chain whose dominant logic is algorithmic. To build is to accept the cost of sovereignty in a defined domain. To buy is to accept the efficiency of the market with open eyes about its conditions. To control is to accept that between full ownership and full dependency lies a disciplined middle path that most mature firms will travel most of the time. The choice among them is the work of the board, and it is work that cannot be delegated without consequence.
For weekly analysis on capital, leadership and geopolitics: follow Dr. Raphael Nagel (LL.M.) on LinkedIn →