A barbell strategy in generative AI points to an overlooked middle

No Content

Deals within generative AI have been hot—depending on where you look. A barbell is forming within the vertical, with more deals flowing to the underlying technologies and vertical applications.

Meanwhile, investment in middleware applications—software for audio, language and video—has been declining, according to PitchBook’s recent note on generative AI VC trends. These lagging categories present a potential opportunity for VCs and startups to build up underdeveloped areas within generative AI.

Generative AI investment this year has concentrated on the natural language interfaces segment and AI core segments, according to PitchBook data. These types of startups are focused on creating chatbots and personal assistants, deploying and orchestrating large language models, and maintaining model architecture. This includes OpenAI and its ChatGPT service along with Anthropic‘s Claude.

The hype surrounding these kinds of generative AI startups stands in contrast to the general-purpose middleware segments like audio, video and content creation. Still, some VCs see opportunity in the underrepresented middle.

“It’s possible for startups to leverage chat, video, and even motion to displace incumbent software,” said James Murphy, general partner at Forum Ventures. “The entire way in which humans interact with software has opened up an opportunity.”

Just last year, visual media was a standout category within generative AI. Startups like Midjourney and Stability AI, the creator of Stable Diffusion, were breakout stars within image generation software. Runway, a video generation startup for creatives, did raise a $141 million Series C extension from investors like NvidiaSalesforce Ventures and Google in June, but overall, deals for visual and audio applications have been lower in 2023.

Brendan Burke, a senior research analyst at PitchBook covering AI and machine learning, says that one of the reasons AI core and certain application startups are doing so well is their lack of reliance on one single large language model, unlike other startups within the middleware section that require specialized models.

“Vertical applications that focus on a specific use case are growing more quickly in deal count given their competitive insulation from foundation model providers,” Burke said.

Similar Posts