Generative AI investors are shifting their focus from the companies building foundation models to the companies making applications built on top of those models.
A foundation model is a general-purpose AI model that can be instructed to generate content, such as text, code, images and synthetic data. Generative AI (GenAI) application companies leverage third-party foundation models to build software services for specific consumer or business purposes.
Eight of the top 10 largest GenAI deals in 2024 to date went to companies that built applications using third-party foundation models, including code generators Magic AI Inc. and Exafunction, which does business as Codeium, and search app startups Perplexity AI Inc. and Glean Technologies Inc. Together, those four companies raised more than $450 million. The companies use third-party foundation models, including those from OpenAI LLC, Mistral AI SAS, Meta Platforms Inc. and Anthropic PBC.
GenAI companies without their own foundation models attracted more than twice as many investing dollars in the first quarter of 2024 compared to the same period last year, according to data from S&P Global Market Intelligence and 451 Research.
Investors' interest in foundation model providers began to slip in late 2023. As of the first quarter of 2024, total funding received by the GenAI industry overall had fallen for two consecutive periods, largely due to the declining interest in foundation model providers.
The 10 largest GenAI funding rounds in 2023 all went to startups that built their own foundation models, reflecting the strong interest in this group of companies at the start of last year. Microsoft Corp., Alphabet Inc., Amazon.com Inc. and SAP SE were among the key backers for GenAI companies in 2023.
According to 451 Research, about 45 companies now have their own foundation models. Databricks Inc. is the latest to launch its own large language model, called DBRX.
As competition heats up, plowing money into foundation model startups is fraught with risk. Building foundation models from scratch requires enormous resources in the form of computing power and human talent — both of which remain scarce.
Given the number of foundation model startups in the market, a "consolidation wave" could be on the horizon, said Maximilian Freiermuth, founder of cloud company Genesis Cloud and a former McKinsey consultant. "Pure API-/Chat-based foundation models have weak moats, with users and developers often finding it easy to switch between models," Freiermuth said.
The difficulty of operating a foundation model was only displayed in the first quarter with Inflection AI, which switched its focus to creating GenAI models for smaller business use cases after key co-founders and employees decamped to Microsoft. Inflection AI has raised $1.5 billion in funding since its launch in 2022.
"We doubt that foundation models alone are a sustainable business model because the cost of training and maintaining the models is too expensive," said Nick Patience, lead analyst for AI and machine learning at 451 Research. "The availability of viable open-source models as alternatives will also put pressure on margins."
