附近上门

Editor鈥檚 note: This article is part of an ongoing series in which 附近上门 News interviews active investors in artificial intelligence. Read previous interviews with Felicis, Battery Ventures, General Catalyst, Bessemer Venture Partners, Accel, Insight Partners, Index Ventures, Sequoia Capital, Section 32, M12, Sapphire Ventures, Bain Capital Ventures, Menlo Ventures and Scale Venture Partners, as well as highlights from these stories from 2023.

Two years into the artificial intelligence wave, we thought it would be helpful to speak with an early-stage investor in the sector to get a perspective on what it鈥檚 been like to invest in generative AI startups from the get-go.

To that end, we caught up with founder and general partner . The Palo Alto, California-based firm invests at pre-seed, seed and Series A, and closed on its fifth fund of $275 million (its largest), and a third opportunity fund of $119 million to invest in its winners 鈥� both in September.

From its first $100 million fund raised in 2012, the firm has built its practice in enterprise cloud-based services leveraging data and analytics.

Costanoa鈥檚 focus today is applied AI and AI infrastructure, and B2B fintech. With generative AI, 鈥渨e can solve problems that were unsolvable two years ago,鈥� said Sands.

Costanoa is also doubling down on cybersecurity 鈥� as the threat instance has increased with AI as well as the potential to improve security operations 鈥� and national security. The firm incubated in 2019, before defense tech and AI were as popular as they are today.

New capabilities, not imagined before

The advancements in AI in recent years mean that through natural language processing, written language can be easily digested, consumed, organized and even reasoned with. Computer vision allows images to be evaluated and understood. Combine those technologies, and you start to get some very interesting and complex outputs, Sands said.

Going vertical

鈥淲e have literally solved most of the problems that could be solved by that generation of technology,鈥� Sands said of the SaaS revolution of the past 20 years.

The firm believes that generative AI has now opened up vertical opportunities that might have been too small in the past.

鈥淭he knock on vertical SaaS has always been: it’s a smaller total addressable market because you limit yourself to one vertical, rather than building something that can apply to every vertical,鈥� said Cowgill.

The firm is looking closely at each vertical to understand how big it can be with the rise of AI.

In that vein, Costanoa backed , which provides computer vision for fish farming.聽 It also invested in for pilot assistance in aviation, and , a service to give first responders situational awareness by consolidating and presenting data. It also made a stealth investment in a company in construction permitting. Though that may seem narrow, there’s a big opportunity in owning an end to end workflow in that space.

Bullish on native AI

The big question for investors is: Does this vertical opportunity benefit existing mid-size portfolio companies embedding an API? Or is it like the cloud and mobile era where it was challenging for incumbents to rebuild their business models?

鈥淚’m more bullish on AI native application layer companies than I was two years ago,鈥� said Cowgill, who has seen firsthand how hard it is to integrate AI and where the value lies.

The firm has a handful of later-stage vertical SaaS companies working to integrate AI. 鈥淭he rate at which they will integrate AI will be slower,鈥� said Cowgill. 鈥淭hey might actually end up delivering different services and different products, different business models.鈥�

鈥淭he real value in AI comes when you can own an end-to-end workflow,鈥� he said. 鈥淭his is the idea of an agent moving from just applying AI, pointing at something and saying, 鈥榮earch across it with AI, summarize with the AI鈥� 鈥� to own a job to be done with AI. Getting agents to work is incredibly difficult.鈥�

Nuance in picks and shovels

Costanoa has made several bets in AI infrastructure. It invested in , creator of an AI junior data scientist to assist teams. It also backed , an open source visualization stack for multimodal data including audio, image and video, and , which is used to fine-tune models.

Still, Sands noted the infrastructure for model management broadly has not grown as much as was expected 18 to 24 months ago, in part due to the fact that companies in applied AI are not managing models themselves, but purchasing from hyperscalers.

鈥楾hings that matter鈥�

But some things have not changed.

In a 2018 blog post 鈥溾�� Sands noted that AI does not replace the need for 鈥渆xcellent product management鈥� and that 鈥渄ata is as important as the algorithm.鈥�

Through the firm鈥檚 12 years, Sands said he has focused on eternal principles: 鈥渇ind extraordinary people, pour time and energy and capability into them so that they have the best chance of success,鈥� he said. 鈥淧ick things that matter 鈥� both technologies that matter and problems that matter.鈥�

Related 附近上门 Pro list: