If AI is a commodity, where is the value?
AI companies have to avoid the commoditization trap by finding revenue streams with sustainable pricing power
Introduction
An interesting thing about AI is that all of the main AI companies, including OpenAI, Google, Anthropic, and Mistral, have repeatedly reduced the cost developers pay for processing data through their models. These costs are referred to as token costs, “token” meaning the smallest unit of data that these companies’ large language models (LLMs) process. As these token costs decline to zero, one has to wonder whether artificial intelligence is being commoditized. If customer costs are declining toward zero, how can OpenAI and its competitors maintain the valuations at which investors most recently invested in them, let alone increase those valuations over time?
One of the main reasons that investors have bid up OpenAI’s shares to nearly a $100 billion valuation is because investors believe that OpenAI is closer to developing artificial general intelligence (AGI) than its competitors. The problem, then, is this: if the cost of tokens declines towards zero, we’re rapidly approaching a world in which the cost of artificial intelligence is negligible. It’s hard to generate billions of dollars in revenue when the price your customers pay is near zero!
And, to the extent that commoditization does occur with artificial intelligence, how can companies avoid the commoditization trap? If all a company does is sell a commodtized product or service, the company will quickly find that its revenues, and so its market valuation, decline to zero. How does a company retain its value when the price of the product it sells inexorably trends towards zero?
There are two questions we need to answer:
Why does commoditization occur in the technology industry?
How can a company avoid this commoditization trap?
Why commoditization occurs in the tech industry
Commoditization in the technology industry refers to a process where products or services that were once unique and high-margin become indistinguishable from others in the market, leading to increased competition and lower prices.
Within AI, we see this happening very rapidly: OpenAI, Meta, Google, Anthropic, Mistral, and a variety of open-source projects, all compete with one another, offering features that vary little from provider to provider. Yes, OpenAI’s GPT4 is presently the most capable of the various large language models, but its cost on a per-token basis suggests that the market values it basically at parity with the other companies’ large language models. Intelligence has become commoditized, for better or worse.
The commoditization process occurs in several stages:
Innovation phase: initially, a new technology or product is developed. It’s unique, has little competition, and offers significant value, allowing the manufacturer or service provider to command high prices and enjoy high profit margins.
Adoption and growth: As the technology gains popularity, more competitiors enter the market. They often innovate slightly on the original idea, but the core technology or concept remains largely unchanged. This phase sees rapid growth, with increasing sales volumes but the beginnings of price competition.
Maturity and standardization: The technology or product becomes well-understood and widely used. Differences between competing products become minimal, often limited to brand name or minor features. In this phase, the technology or product becomes a standard in the industry.
Commoditization: At this stage, the market perceives little to no difference between products from different providers. Price becomes the primary basis for competition, as consumers see all products as essentially the same. Profit margins shrink as companies cut prices to maintain or grow market share.
Decline: Eventually, the commoditized product may face obsolescence, either due to market saturation or replacement by newer technologies.
The process of commoditization is easy enough to understand, but why does it happen? What features are inherent to technology which allow it to happen? What follows is a non-exhaustive list of these features.
Rapid knowledge transfer and iteration: Software allows for knowledge transfer and iterative developments to happen at an accelerated pace across companies. Knowledge transfer and iteration does just happen within a company. These things occur across multiple companies competing for the same customers. To some extent this can be mitigated by patents, NDAs, corporate security initiatives, etc. Nonetheless, information is leaky, and knowledge tends to diffuse, fairly rapidly, over time. It’s worth noting that the research which OpenAI used to build its various large language models, came from a Google research paper titled Attention is all you need.
Open source and collaborative development: The culture of open source software and collaborative development in the tech community leads to rapid commoditization. One reason that Steve Ballmer was so opposed to the open source Linux operating system was that he was worried it would commoditize Microsoft’s bread and butter—its Windows operating system. Why, Ballmer reasoned, would MIcrosoft’s clients pay a licensing fee to Microsoft for Windows, and all of its ancillary products, when they could just install Linux? OpenAI is not an open source project, but a number of its competitors, including Facebook’s offering, Llama, are some flavor of open source.
Intellectual property challenges: Protecting the intellectual property rights of something as complex and potentially ground-breaking as AGI could be extremely challenging. This is particularly true if AGI is based on widely known techniques or incremental advances over existing AI technologies.
Regulatory and ethical pressures: Given the potential impact of AGI on society, there might be significant regualtory and ethical pressures to prevent any single entity from monopolizing this technology. Governments and international bodies may intervene to ensure widespread access, or to control the development and deployment of AGI for the greater good.
Rapid advancements in complementary technologies: The development of AGI might spur advancements in other complementary technologies, which in turn could make the development of AGI easier for other companies. This could include improvements in computing hardware, data storage, and algorithms.
Economic incentives for fast adoption: The immense potential value of AGI in various sectors would create strong incentives for companies to adopt or develop their own AGI solutions as quickly as possible. The competitive advantage provided by AGI could be so significant that companies would invest heavily in catching up. Meta recently announced that it is spending tens of billions of dollars on NVIDIA GPUs, in an effort to speed up its AI research.
Talent mobility: The mobility of talent in the tech industry also plays a role. Experts and key personnel involved in the development of the first AGI might move to other companies, taking their expertise with them and thereby accelerating the spread of knowledge. It’s important to remember that, in California, where many technology companies are based, non-compete agreements are illegal. As long as talent can move easily from one company to another, knowledge will diffuse, even if employers require employees to sign NDAs.
Network effects and ecosystems: The value of AGI could be heavily dependent on network effects and the ecosystem in which it operates. Companies that can quickly integrate AGI into existing ecosystems or platforms could commoditize the technology by making it a fundamental, yet not exclusive, part of a larger system of products and services.
Reverse engineering and espionage: In a competitive global market, the risks of reverse engineering and industrial espionage are significant. These practices could lead to the rapid dissemination of AGI technologies among competitors.
Limitations of first-mover advantage: While being the first to develop AGI provides a temporary advantage, the history of technology shows that first movers don’t always dominate the market in the long term. It’s often those who improve upon the intiial invention and better understand market needs that eventually capture the most value.
How companies in traditional commodities markets avoid commoditization
One way to understand how technology companies can avoid the commoditization trap is to examine how companies which operate in traditional commodities markets avoid it. A bushel of wheat is a bushel of wheat no matter which company sells it to you, and a barrel of oil is the same whether one company or a different one sells it to you. Traditional commodities are fungible, which means that there’s zero difference between bushels of wheat or barrels of oil.
If traditional commodities are fungible, why, then are companies like Archer Daniels Midland (agricultural products) and Exxon Mobil (petrochemicals) worth tens of billions of dollars? Surely, if wheat and oil are fungible then a customer would just buy from CheapestWheatProducer and CheapestOilProducer. What do these companies do to sustain value?
Answering this question, it turns out, provides us with guidance about how AI companies can avoid the commoditization trap. So let’s take a look at this in more detail. Here’s what companies like Archer Daniels Midland and Exxon Mobil do to sustain their valuations:
Control over supply chain: Companies like ADM and Exxon often control significant portions of the supply chain. This control extends from cultivation or extraction to processing and distribution. Such vertical integration allows these companies to maintain profitability even when the underlying commodity is widely available.
Scale and efficiency: These companies operate at a massive scale, allowing them to benefit from economies of scale. This means they can produce or procure commodities at a lower cost per unit than smaller competitors, which helps maintain their profitability even in a commoditized market.
Brand and market position: Although they deal in commodies, companies like ADM and Exxon have established strong brands and market positions. This can provide a competitive edge, such as preferred supplier status, long-term contracts, or the ability to command a slight premium based on perceived quality or reliability.
Diversification: Companies like ADM and Exxon often diversify their product offerings. They don’t just sell agrilcutral products or crude oil; they also sell refined products, specialty chemicals, processed food ingredients, and other value-added products, which have higher margins.
Logistical and technological advantages: Big players in commodity markets usually have advanced logistics and technology. This can include everything from superior shipping and distribution networks to advanced techniques in exploration, extraction, and cultivation, which smaller competitors may not have.
Regulatory influence and market access: Large, established companies often have greater influence in regulatory environments and can shape policies in ways that favor their operations. They also have better access to global markets, including emerging markets, which can be more challenging for smaller companies to penetrate.
Adaptation and innovation: Despite dealing in commodities, these companies often invest heavily in research and development. This R&D is not just about finding new sources of commodities, but also about improving efficiency, developing new uses for their products, and even exploring entirely new lines of business.
Customer relationships and service: Long-standing customer relationahips and a reputation for reliability can be significant assets. These companies often provide a level of service and consistency that newer entrants or smaller competitors struggle to match.
So let’s bring this back to AI. While the core technology of AI looks like it will be commoditized, a company can still maintain and add value through various strategic initiatives, much like companies in traditional commodities markets. The key would be to leverage first-mover advantage, scale, market knowledge, and continuous innovation to stay head of the curve.
Conclusion
So we’re left with the following conclusion: it’s likely that artificial intelligence, whether of the general form or not, will be available for near-zero, or zero, cost. In order for companies like OpenAI and its competitors to sustain their current valuations, and grow them over time, they will have to diversify their businesses into ancillary revenue streams which have sustainable pricing power. Apple, Microsoft, Google, Nvidia, and scores of other technology companies have successfully navigated this, and avoided the commodification trap, even though many of the technologies these companies develop are themselves commodified.