Moore’s death and the rebirth of vertical monopolies
As Big Techs integrate hardware, competition policy faces new challenges
Last Friday, 24 March 2023, Intel co-founder and microchip visionary Gordon Moore passed away at the age of 94. His prediction on the exponential economies of scale of chip production made him a symbol of the technological progress in the industry. Moore’s law states that the number of transistors on a microprocessor chip will double every two years or so. Such progress contributed to creating a modular industry structure whereby semiconductor companies took care of providing computing power to keep up with the extraordinary growth of the software industry. All modern-day Big Techs are software companies and they owe their success to the hardware innovation provided by semiconductor companies like Intel — at a pace predicted by Moore’s law.
The tech industry, however, has mourned the death of Moore’s law for some time already. In 2016, Nature published an article arguing that semiconductors are reaching a stage in which a doubling of their capacity is proving almost impossible as more and more components are crammed into chips. Moore’s law seems to have hit the wall of physics, as the most advanced transistors are currently only around one order of magnitude larger than the silicon atoms which they are made of. Moreover, the cost of manufacturing and cooling microprocessors is increasing exponentially with performance. Developing a new 10 nm chip costs around US$170 million, while for a 7 nm chip it is almost US$300 million and over US$500 million for a 5 nm chip. On top of this, demand for computing power shows no sign of slowing down. Virtually all projects in the innovation pipeline require massive levels of computing capacity. Innovations such as the Metaverse and AI bots seem to implicitly project into the future the same performance improvements experienced in the past decades. In other words, software ate the world so much that hardware has become the industry bottleneck.
Innovating through specialisation
How are Big Techs then planning to deliver on their software innovation promise? There is an answer: by vertically integrating hardware production and through chip specialisation. This addresses the performance growth problem in two ways. Firstly, it allows prosperous companies to directly put their financial weight behind the expensive development of new chips. Indeed, this is likely to be a major component in Tech’s huge R&D budgets. Secondly, as Moore’s law is breaking down with regard to transistor size, performance can be propped up by means of specialisation. While traditional chips tended to be general purpose, in-house chips tend to be co-designed with software, enhancing purpose-specific performances.
The process has already started, as all of the Big Techs are entering the semiconductor business. Amazon was the first one, entering the industry as early as 2015 through the acquisition of Annapurna Labs. Amazon’s specialised chips now sustain its massive, computing power-thirsty cloud business. In 2020, Apple replaced Intel’s CPUs with its own M-series in Mac computers, bringing about huge performance improvements in its new MacBook models. While the manufacturing stage of M-chips is still outsourced to semiconductor companies, bringing the design stage in-house is an important indicator of the underlying changes in the tech industry architecture, and the manufacturing stage might follow suit in the coming years. Google, Microsoft and Meta all undertook similar initiatives.
The integration of chip production is accompanied by Big Techs’ massive investments towards the purchase of data centres, in order to support cloud services. As the adage goes, “there is no cloud, it’s just someone else’s computer.” In the case of hyperscale data centres, which are the largest scale and most efficient types of data centres, that “someone else” is usually a Big Tech. Amazon, Microsoft, Google and IBM all have more than 60 data centres each. The combination of investments in chip production and data centres makes the trend inescapable: Big Techs are vertically integrating hardware. The architecture of the industry is changing, as modularity is reversing and vertical monopolies are making a comeback.
Implications for competition
This is likely to have profound consequences in terms of competition. As each Big Tech increasingly relies on its own specialised hardware, the walled garden of its ecosystems will have higher walls and wider moats. Firstly, inter-ecosystem boundaries are likely to accentuate, as common components are replaced with ecosystem-specific analogues. This means that, as a user, you are more likely to choose the same brand for all your devices. New entrants in digital industries will be penalised. Secondly, hardware production has huge fixed costs and a new entrant is unlikely to be able to sustain them. Thirdly, hardware specialisation will increase the company’s leverage over the complementors operating in its ecosystem, requiring them to integrate more tightly or face lower performance. Google’s Tensor Processing Unit (TPU), for example, is built around Google’s own TensorFlow software, therefore the distribution of the chip to third parties will also promote software adoption — and whatever other products are complementing them.
On the other hand, competition between ecosystems is likely to become fiercer. This will depend on the extent to which tech giants will converge in the provision of certain technologies. Recent industry developments are seeing neck-to-neck competition over AI-powered chatbots, as Google’s Bard tries to keep up with Microsoft’s Bing and Chat GPT. Finally, traditional semiconductor companies will have to face an identity crisis, for if they do not change their business model, they will become increasingly marginalised. Intel’s last quarterly results, which saw a year-on-year revenue decline of 32%, were a warning.
Conclusion
The power of the digital world arose from its edges: innovation, growth, and opportunities were brought about by users in a decentralised setting. Today, a handful of technological giants came to dominate the Internet, bringing about efficiency, but also threatening the initial promise of inclusiveness and empowerment. Competition authorities will have to be on alert. Depending on the circumstance, tools to regulate interoperability and data portability might come in handy to promote an open architecture of the digital space. Changes are happening fast and, in digital markets, antitrust agencies have a history of acting too little and too late. On both sides of the Atlantic, policy teams seem to be aware of what is at stake. Yet, to their ability to act will also depend on the availability of good theory to understand the economics of industrial modularity, system interoperability and coordination within digital ecosystems. Academic research has a key role to play in this and, indeed, at IIPP we have accepted the challenge.
