Can LLM AI Be a Source of Competitive Advantage?

Let’s start with the optimistic “yes”, and see if it remains acceptable. Before we get carried away, a few reminders.

Competitive advantage: “The term competitive advantage refers to the ability gained through attributes and resources to perform at a higher level than others in the same industry or market.” [1]
LLM, for Large Language Model: “a language model notable for its ability to achieve general-purpose language generation and other natural language processing tasks such as classification. LLMs acquire these abilities by learning statistical relationships from text documents during a computationally intensive self-supervised and semi-supervised training process. LLMs can be used for text generation, a form of generative AI, by taking an input text and repeatedly predicting the next token or word.” [2]

For an LLM to be a source of competitive advantage, it needs to be a resource that enables products or services of a firm “to perform at a higher level than others in the same industry or market” [1].

Starting from the optimist’s position, try to find ways an LLM could be a source of competitive advantage.

For companies such as Open AI, Mistral, and more generally, those whose product/service is an LLM itself, the LLM needs to perform better than substitute services. It can do so in various ways. It may be first to market, which is Open AI’s case, which will only be sustainable if it manages to increase switching costs. It may be cheaper to scale up, that is, have a favorable ratio of computational resources invested to specific performance thresholds across relevant benchmarks. It can be trained on proprietary data, or using proprietary algorithms, which lead it to demonstrate performance, subjective or measurable, that appeals to its target market. It may achieve trustworthiness in a way that is hard to replicate, for a target market this matters to, for applications where that makes a significant difference.

For firms who make AI derived from an LLM, competitive advantage cannot come from the LLM itself, but from how it integrates with other components of the product and/or service. The LLM needs to enable the customer to get more, or the same at lower cost, because the LLM is part of the service and/or product. The likely approach will be to use the LLM to create shortcuts, simplify workflows that users already had. Innovation will come from power users.

What would be grounds for pessimism that firms offering proprietary LLMs can sustain their competitive advantage?

One, LLM performance depends on the amount and quality of training data, and it is questionable whether it is possible to create proprietary datasets of that scale – most current LLMs are trained on, roughly speaking, similar data [3]. So seeking competitive advantage in training data is going to remain a significant challenge [4].

Two, making a significant advance in algorithms requires significant research, most of which requires substantial resources – a combination of talent and computational and data management infrastructure that few firms globally are able to build and sustain [5].

The two factors above will make it unlikely that many firms will be able to deliver LLM AI products and/or services that ground competitive advantage in data and/or algorithms. Ability to make capital investments may end up being a bigger factor, and given its scale, talent costs pale in comparison. In other words, it is not that LLM AI cannot be the primary source of competitive advantage, but that they will be so for very few massive firms, which if they don’t engage in explicit oligopoly (if anything to steer talent spend), will have significant incentives for a covert oligopoly, or at the very least some loose forms of coordination.

There are grounds to be pessimistic that derived LLM AI can be a source of competitive advantage – that because a firm integrates available LLM AI into its own products/services, it will achieve competitive advantage. There are incentives for companies with proprietary LLM AI to make them cheap to integrate – consequently, competitive advantage will be from the combination of an LLM AI with the features of the existing product and/or service, which in turn means that these features are what will make the difference. Integrating LLM AI with mediocre products and services will not make them by default any better.

References