Sooner or later, an investor, a member of the executive team, or a partner will likely ask the question: “Should we build an API for our product?” . The decision on whether to invest efforts to develop a product interface is often based on opportunity and cost. Many companies go on building APIs because their competitors also have them, or because customers ask for integrations.
However, modularity theory from Clayton Christensen suggests that such a decision should be more strategic. Creating external product interfaces or, more broadly, building a modular product whose components are interchangeable, is a product architecture decision that can strongly influence the success of the product.
There is nothing wrong with asking the question “Should we build an integration or not?” . But, the real question a product leader should ask is “What product architecture best fits the market where we (want to) compete?”. Reframing in this way allows to consider strategic aspects such as industry maturity and customer expectations. From this perspective, interfaces are not simply product features. They are a consequence of choosing the right product architecture.
There are two primary product architecture styles, each coming with its own trade-offs:
- An integral product presents itself as a set of tightly coupled components. It is optimized as a single stand-alone system with no external dependencies. The high interdependency between its components allows to maximize product performance and meet demanding market requirements. This is because its components are optimized to work together in the best way possible only to address exigent market requirements. Although an integral product achieves high performance, its components are by design not reusable. This means that, once standardized components become available on the market, the product risks becoming less competitive, in terms of cost and versatility, than modular alternatives built from interchangeable parts.
- A modular product is, on the other hand, built from interchangeable components connected through well-defined interfaces. A modular product also exhibits external interfaces to work with other products agnostically. A modular product usually has shorter time to market, higher flexibility and lower production and maintenance costs. This is because its components are typically designed for reuse and compatibility more than overall performance. Its sub-optimal performance comes with a lower price premium compared to an integral product.
Choosing the right architecture style
The choice on whether to develop an interdependent or a modular product architecture strongly depends on market positioning. More specifically, it comes down to how well product performance meets market demands.
When an integral architecture is the only choice
In highly demanding markets, where existing offerings are still “not good enough” to meet the performance the market requires, an interdependent architecture is usually superior.
This is because, early in an industry’s life cycle, product developers must control every proprietary detail to optimize performance. Because components are tightly coupled rather than isolated, a change to one piece requires adjusting the entire system, making it nearly impossible for competitors to replicate without building the product from scratch.
An important caveat is that choosing an integral architecture is not a deliberate decision to vertically integrate and create barriers to entry. More often, the choice is a necessity. In new markets, an integral architecture is often the only way to achieve the performance levels the market demands. Product developers simply cannot rely on off-the-shelf components to reach the required level of performance.
For example, SpaceX adopted an integral architecture because the existing aerospace ecosystem could not provide off-the-shelf components that met its performance requirements. Reusable rockets required a fundamental redesign of propulsion, materials, avionics, and flight software. Because these components were tightly coupled, SpaceX had to optimize the entire launch system rather than assemble it from existing modules.
When a modular architecture works best
While the integral architecture is the most appropriate choice for products in new or early-stage markets, a modular architecture typically works best as the industry matures. This is the stage when off-the-shelf components become sufficient to meet market performance requirements.
Thanks to the availability of interchangeable components, a modular architecture allows faster deployment, higher scalability, and lower production and maintenance costs. As the technology frontier advances and components mature, product developers can swap individual components without re-engineering the entire product.
The premium bicycle market is a classic example of an industry that transitioned from extreme, hand-built integration to highly optimized modularity.
When high-performance bicycles first emerged, manufacturers had to design frames and components together because industry standards were still evolving and many components were not interchangeable. As drivetrain interfaces, wheel dimensions, brake mounts, and other standards matured, manufacturers increasingly specialized. Today, bicycle brands routinely combine modular groupsets, wheels, brakes, and suspension systems from specialized suppliers while focusing their own innovation on frame design, geometry, and rider experience.
Evolution of product architecture in the humanoid robots market
Within industries, products tend to evolve from integral to modular architectures. A market originally dominated by highly integral products, now gradually becoming modular, is the humanoid robots market.
Building a humanoid robot is similar to engineering a Formula One car. Performance depends less on the quality of individual components than on how tightly the entire system is optimized. The hardware and control software continuously influence one another, making system-level integration the primary source of competitive advantage.
Software interdependence on modular hardware
A humanoid robot is a highly complex system, made of many leading-edge components, many of them custom-made. These components must work together seamlessly for the overall system to perform effectively.
However, what increasingly differentiates humanoid robots is not so much the hardware as the software. Hardware components, such as motors, sensors, and actuators, are gradually becoming modular. China, the United States, and Europe now have many suppliers specializing in high-performance modular components. On the other hand, software remains the primary source of differentiation. This is true in application domains, such as healthcare, where human-machine interaction is critical. For such applications, user experience and fault tolerance mainly determine product performance.
This different impact between hardware and software is why the market for humanoid robots is actually splitting into a two-tiered product architecture. The hardware layer is rapidly decoupling into modular components. Actuators, frameless servo motors, harmonic drives, and sensors (like IMUs or depth cameras) are no longer bespoke engineering projects. In fact, industry development has pushed the the technology frontier forward.
By contrast, humanoid intelligence and physical execution have not yet reached the same level of modularity. Current technology is still not good enough to provide modular, cost-efficient, application-ready intelligence and whole-body control. The AI brain and whole-body control systems still require extreme interdependent integration. Leading humanoid robotics companies, such as Figure AI and Unitree, do not simply use off-the-shelf AI models. They invest billions in tightly integrated, proprietary end-to-end neural networks, where hardware design influences software training, and software requirements shape hardware design.
This tight integration of AI models for reasoning and physical execution is where humanoid robot manufacturers and application developers create most of their differentiation, and capture most of their profits. Nevertheless, the first modular components for physical AI are already emerging. This transition will take time. But it marks the beginning of the humanoid robots industry’s shift from highly interdependent products toward highly modular architectures.
The product architecture choice
Choosing one architectural style over the other comes down to three factors:
- Whether the market requires performance beyond what available products can deliver.
- Whether you possess a capability advantage over competitors.
- Whether you have enough time to develop an interdependent solution.
If you have a capability advantage, the market is willing to pay a significant premium for it, and the window of opportunity is long enough to support development, an integral architecture is usually the right choice. It allows you to build a first-mover advantage, learn from the market, and iterate.
If suitable components are already available for integration, your time to market is limited, or you do not possess a capability advantage that customers recognize as superior, a modular architecture is likely a better choice. It helps avoid the “not invented here” syndrome and accelerates time to market.
The humanoid robotics industry is not the only one undergoing a strategic transition from interdependent platforms to modular architectures. Other high-tech industries, including smart glasses, satellites, electronic health record systems, and many others, are seeing the emergence of specialized modular component suppliers as supply catches up with demand. At the same time, new integral products continue to emerge to serve entirely new markets. Over time, these will become modular too.
Post image by Hithesh Shivakumar on Unsplash.
