The Future of Generative AI: Ownership, Business Models, and User Interfaces

The Future of Generative AI: Ownership, Business Models, and User Interfaces

Introduction:
Generative AI has revolutionized the tech industry, offering novel solutions and transforming the way users interact with computers. In this article, we will explore the ownership dynamics, business models, and the evolution of user interfaces in the generative AI landscape.

Ownership and Business Models:
In the generative AI market, infrastructure vendors have emerged as the biggest winners, capturing a significant portion of the revenue. While application companies experience rapid revenue growth, they often struggle with retention, product differentiation, and gross margins. On the other hand, model providers, responsible for the existence of this market, are yet to achieve large commercial scale.

The growth of generative AI applications has been staggering, with image generation, copywriting, and code writing being the first product categories to exceed $100 million in annualized revenue. However, in the absence of strong technical differentiation, long-term customer value is driven through network effects, data retention, and complex workflows.

It is important to note that selling end-user apps may not be the only path to building a sustainable generative AI business. As competition and efficiency in language models increase, margins are expected to improve, and retention is likely to increase as AI tourists leave the market. Vertically integrated apps also hold an advantage in driving differentiation.

Moreover, model providers have realized the potential harm and benefits of generative AI and have incorporated the public good explicitly into their mission. Many have organized as public benefit corporations (B corps) or issued capped profit shares without hindering their fundraising efforts.

The Role of Infrastructure Companies:
A significant portion of the revenue in the generative AI market flows to infrastructure companies, particularly cloud providers. On average, app companies spend 20-40% of their revenue on inference and per-customer fine-tuning, which is paid either to cloud providers or third-party model providers. Thus, approximately 10-20% of total generative AI revenue goes to cloud providers.

Nvidia stands out as a major winner in the infrastructure space, reporting billions of dollars in data center GPU revenue, including a significant portion from generative AI use cases. Infrastructure companies possess lucrative, durable, and seemingly defensible positions within the generative AI stack, supported by scale moats, supply-chain moats, ecosystem moats, algorithmic moats, distribution moats, and data pipeline moats.

The Evolution of User Interfaces:
AI is introducing the third user-interface paradigm in computing history, shifting the locus of control from users telling computers how to do tasks to users stating what they want. This represents a significant shift in interaction mechanisms and marks the first new interaction model in over 60 years.

The first UI paradigm, batch processing, involved users specifying a complete workflow without back-and-forth interaction. The second UI paradigm, command-based interaction, allowed users and computers to take turns with one command at a time. The graphical user interface (GUI) dominated for about 40 years until the rise of AI.

The current chat-based interaction style in generative AI has its limitations, requiring users to articulate their problems as prose text. Research suggests that a significant portion of the population may not be articulate enough to effectively use current AI bots. To overcome this, better usability of AI becomes a competitive advantage.

The future of user interfaces in generative AI lies in a hybrid approach that combines elements of intent-based outcome specification and command-based interaction. While the third UI paradigm focuses on intent-based outcomes, the second UI paradigm, GUI, will continue to play a role, albeit in a less dominant capacity.

Actionable Advice:

1. For model providers, commercialization is likely tied to hosting. Emphasize the demand for proprietary APIs and consider offering hosting services for open-source models to facilitate sharing and integration.

2. App companies should focus on building differentiation through vertical integration and complex workflows. Explore partnerships and network effects to drive long-term customer value.

3. Usability should be a significant focus for AI companies. Improve the user experience to cater to a broader population, ensuring that AI tools are accessible and effective for a wide range of users.

Conclusion:
Ownership dynamics in the generative AI market favor infrastructure vendors, while the success of app companies and model providers depends on differentiation and long-term customer value. The evolution of user interfaces introduces a new paradigm, where users state their desired outcomes instead of providing explicit instructions. As the generative AI landscape continues to evolve, a combination of intent-based and command-based interfaces will shape the future of user interaction, ensuring usability and effectiveness in AI applications.