Factors That Could Tip The AI Market To A Few Dominant Players

We have seen the rise of megacloud providers who now dominate the computing market. Will the same happen with artificial intelligence? Could a few well-heeled tech giants corner the AI space?

Concerns about an AI monopoly, duopoly, or triopoly emerging was recently voiced in an analysis published by the Organization for Economic Co-operation and Development (OECD). The rise of AI is already creating fissures in the market economy, with potential — emphasis on potential — to freeze out companies from essential data and compute power needed to compete in this space.

It’s still too early to make predictions about market dominance of AI, the study’s authors, led by Richard May of OECD, conclude. “Market dynamics are still in a state of flux as are as companies rush to innovate,” they cautioned. “It is hard to predict how far capabilities will advance, and how quickly. It is too early to assess competition issues in the AI sector with authority, and to know how realistic fears on the future of competition are.”

Yet, there are patterns emerging that suggest potential competitive issues ahead. “While the deployment of generative AI for use continues to unfold, it is clear that it requires significant computing power to run queries on these models, meaning a certain level of access to IT infrastructure is required,” the researchers observed.

For example, “links between existing providers of cloud computing infrastructure services or within existing digital ecosystems may present situations in which competitors struggle to access requisite data, compute or end users,” they warned.

While the OECD co-authors hint throughout the paper that government regulators may eventually need to get involved in keeping AI environments open and competitive, they urge a go-slow-and-understand approach to any attempts to intervene. There are arguments that the risk of market monopolization is low at this time, “highlighting that the market is competitive, with many active firms and a dynamic state of innovation.”

Indeed, there are currently almost 77,000 AI companies in the world, according to estimates from Tracxn.

May and his OECD team outlined the factors that could “tip” the market toward a few dominant players:

• Economies of scale: “For foundation models, development costs appear high. If this were to emerge as an independent section of the market, for example if providers were to adopt licensing models, then there may be relatively few running costs, implying substantial economies of scale.:
• Economies of scope: These are “conferred through reductions in costs across business lines, the clearest potential comes from firms operating across adjacent markets, such as cloud computing services or other digital markets.”
• Access to compute power: “High costs of computing power may reduce the ability of firms to develop or train their own models. There may also be potential competition issues due to vertical relationships involving computing power, with suppliers of key computing hardware and cloud service providers also active in the development and deployment of foundation models.”
• Network effects: “There may be potential for feedback effects at certain parts of the generative AI value chain. For example, if user data allows the further refinement of models, then the value of a generative AI service may be higher to consumers as that service gets more users. If these effects are strong, it could provide substantial first-mover advantages, as entrants without an existing network of users struggle to compete by refining and improving models.”
• Mergers and partnerships: “Many acquisitions are pro-competitive and allow businesses to improve their offerings to consumers. This can be particularly true in the context of innovative markets, where the combination of different ideas allows previously unforeseen synergies to be realized.”
• Data volume: This is “likely to be a key requirement for foundation models. Much of the data used [in AI] appears to be publicly available.” However, “some large datasets are proprietary and may provide unique insights that others struggle to replicate, noting as an example Google’s ownership of YouTube and the potential to control access to its video transcripts.”
• Data access: This is predicated on “the extent to which suppliers of generative AI are able to capture and use data from other firms by supplying a service to them.”
• Talent and skills: “Another potential barrier to entry could be access to expertise and talent if these were difficult to obtain. Similarly, the ability for entrepreneurs to launch their own start-ups and innovate could be an important part of the competitive process going forward. There is high demand for the skills required across the AI value chain. How much of a bottleneck staff become is unclear, although one might expect demand and supply to balance over the longer term.”

The AI market is highly diverse, with innovations coming from many different directions. While consolidation is a natural effect with any emerging technology, it remains to be seen how this will play out.