The field of generative AI has been rapidly evolving, and the question of who owns the generative AI platform has become increasingly important. While there are several players in the market, it is the infrastructure vendors who have emerged as the biggest winners so far. They have captured the majority of dollars flowing through the stack, while application companies are growing their revenues quickly but facing challenges with retention, product differentiation, and gross margins. On the other hand, model providers, who are responsible for the very existence of this market, have yet to achieve large commercial scale.

The field of generative AI has been rapidly evolving, and the question of who owns the generative AI platform has become increasingly important. While there are several players in the market, it is the infrastructure vendors who have emerged as the biggest winners so far. They have captured the majority of dollars flowing through the stack, while application companies are growing their revenues quickly but facing challenges with retention, product differentiation, and gross margins. On the other hand, model providers, who are responsible for the very existence of this market, have yet to achieve large commercial scale.

The first wave of generative AI apps are starting to reach scale, with some product categories, such as image generation, copywriting, and code writing, already generating over $100 million in annualized revenue. However, these apps still struggle with retention and differentiation. It is worth exploring whether there are any other use cases that have also reached this scale and have a large user base.

Gross margins vary widely across app companies, ranging from as high as 90% to as low as 50-60%. The cost of model inference plays a significant role in determining these margins. This raises questions for generative AI app companies regarding vertical integration, building features versus apps, and managing through the hype cycle.

For model providers, the key takeaway so far is that commercialization is likely tied to hosting. They need to address questions related to commoditization and graduation risk. Additionally, the importance of money in this market cannot be understated, as startups training their own models have raised billions of dollars in venture capital, a significant portion of which is typically spent on cloud providers.

While cloud providers currently dominate the market, there are other hardware options available, including Google Tensor Processing Units (TPUs), AMD Instinct GPUs, AWS Inferentia and Trainium chips, and AI accelerators from startups like Cerebras, Sambanova, and Graphcore. Intel, although late to the game, is also entering the market with their high-end Habana chips and Ponte Vecchio GPUs. However, these new chips have yet to gain significant market share, with the exceptions being Google TPUs and TSMC-manufactured chips.

One notable trend in the generative AI space is the lack of systemic moats. Applications lack strong product differentiation as they use similar models, and models face unclear long-term differentiation as they are trained on similar datasets with similar architectures. Even cloud providers lack deep technical differentiation as they run the same GPUs. This suggests that there is still room for innovation and differentiation in the market.

In conclusion, the generative AI platform ownership is still up for grabs. Infrastructure vendors have emerged as the biggest winners so far, but there are opportunities for application companies and model providers to carve out their space in the market. To succeed in this space, companies should focus on vertical integration, building differentiated features, managing through the hype cycle, and exploring alternative hardware options. By addressing these key areas, they can position themselves for success in the evolving landscape of generative AI.