The emerging artificial intelligence (AI) supply chain is effervescent and unstable. Startups, digital platforms, software suppliers, and hardware manufacturers are struggling to offer products, solutions, and business models to serve a growing demand from end users.
This is the beginning of a new industry, although it is unclear how AI will reach the last customer mile and a mass market. AI is composed of a series of sub-technologies capable of interpreting external signals; extracting patterns; generating outputs such as speech and text synthesis; evaluating outcomes; and autonomously reprogramming its own logic to learn from experience. A core set of AI technologies (machine learning) enables machines to self-adjust their programmes through experimentation in pursuit of a given objective.
A huge market is being created around this new technology. PricewaterhouseCoopers (2017) estimated that AI will add $15.7 trillion to global GDP by 2030. The impact of AI may be comparable to the other digital waves that penetrated every part of former business models and user appliances: PCs in the 80s, the internet in the 90s, or smartphones in the 2000s. However, it is not yet clear how AI supply chains will be structured, what will be the dominant technological standards, or which final business models will emerge. How will AI reach the mass market? How, ultimately, will AI be democratized?
The emergence of new industries due to disruptive technologies is characterized by initial ‘effervescent’ phases, during which entrepreneurs, former leaders, and new entrants compete for market opportunities. Technological discontinuities generate an initial messy moment of experimentation and competition between different players and their various technological approaches, products, and market solutions. Finally, a ‘dominant design’ for the configuration of a product or service appears that offers optimal efficiency and convenience to the market. Examples of dominant designs include the Ford Model T (automobiles), the Remington typewriter, the IBM PC (personal computers), or Google (search engine).
During the formation of the PC industry, a swarm of startups competed to define the standards (Atari, Amiga, Commodore, Sinclair, Next, and Apple), and it was the IBM design that finally prevailed for two decades (with more than 90% of the market). In the case of the automobile market, a small group of big brands (including Ford, GM, Chrysler, Toyota, Volkswagen, and Fiat) dominated the market for a whole century in a classical example of an ‘oligopoly’.
If we assume the former models of industrial innovation, then the AI industry is in its fermentation phase. Yet the theory predicts the advent of some type of dominant design that will enable the survival of a few players who achieve industry leadership. Fine (2000) states that the creation of industrial supply chains follows a cyclical process (a kind of ‘clockspeed’), between modular/horizontal and vertical/integral architectures. Every modular and disintegrated industry suffers competitive pressures towards vertical integration while searching for efficiency and market control. In contrast, every verticalized industry suffers competitive pressures to disintegrate under the weight of its own bureaucracy and core rigidities, and the lateral expansion of competitors from adjacent market niches. Assuming this model, we are likely to witness a verticalization of the AI industry, with the appearance of one or several integral, end-to-end AI supply chains (from computing hardware to the customized end-user solution). In the same way as any digital industry, AI follows a ‘winner-takes-it-all’ game. Only a few players will survive and dominate the industry, which will probably be vertical and defined by a dominant architecture.
Amazon has shown the path to the end user for AI (the emerging dominant architecture). Years ago, Amazon’s hyper-capacity in computer services and memory storage led the company to offer computing power and memory storage to third parties through the cloud. Amazon followed a ‘SaaS’ model (software as a service) with Amazon Web Services (AWS). It was a huge success. AWS is now a 16 billion-dollar business that is growing 42% annually (McCracken, 2017) Microsoft and Google have followed Amazon’s footsteps. The three firms are expected to capture 80% of all cloud platform revenue by 2020 (Forrester, 2017). With this strategy, businesses can connect to an AI ‘hose’, parameterize their needs, train remote state-of-the-art processors, and access the results through cloud services, while paying for the time and computing capacity used. According to the Financial Times, ‘AI in the cloud will be the next great disrupter’ (Waters, 2016).
In the light of how previous industries have been created, a few AI players may impose their standards, gain market power, and expand across integrated supply chains. But only huge players will be able to provide the computational power, hardware, and R&D investments to create global AI standards. The best positioned are the digital leaders: platforms like Amazon, Google, Microsoft, or IBM. Centripetal forces tend to concentrate resources and gain critical mass backwards in their value chains, with huge R&D investments and supercomputing power; while, at the same time, AI suffers strong centrifugal forces forward, driven by the growing demand for final user applications or products. If concentration dynamics take off, it will enhance the natural competitive advantages of digital platforms, such as Google or Amazon, by absorbing even more massive data flows from thousands of remote and small clients and applications – and thereby gaining more specialized and segmented knowledge, better algorithms, and unbeatable AI competences through spillovers. The AI knowledge created by digital platforms for a set of users in a given industry (such as retail) can be used for other consumers in a type of cross-spillover effect in data management. Digital platforms will be the engine of future AI supply chains and act as global drivers of AI through the cloud.
As in the 80s, when extraordinary computing power was mass marketed and presented to the world in IBM PCs and their clones (and became a commodity), so AI supercomputing power may be mass marketed by digital platforms through cloud services. It will be the final dominant design of the industry. AI will be commoditized and democratized. Like the internet, a standard AI connection will be available for every business. True competitive advantage will be provided by the quality and quantity of data (reflecting the data strategy) and superior customer experience. And, like years ago, when an ‘Intel Inside’ phenomenon emerged (indicating that a PC was no more than the user’s interface for an Intel processor inside), we may soon find a startup, supermarket, hospital, hotel, bank, or car, labeled ‘Google Inside’ or ‘Amazon Inside’. They will be powered by a Google or Amazon AI brain, and will use the same AI hardware and algorithms as the leaders.