There’s a lot of money being thrown into AI. Many analysts choose to focus on the funding within established companies, like Meta’s enormous investment, or the Project Stargate touted by partners OpenAI and Softbank and Oracle, with a $50 billion price tag. But venture capital firms are in the mix, too – they are out there looking for who to fund, as the space heats up. Some may think we’re in a bubble, but the VCs are undaunted, and scouting for AI plays.
In fact, what researchers are finding is that AI firms, and physical AI firms in particular, are winning the day. Joanna Glassner, writing at Crunchbase, has more on what kinds of companies VCs desire:
“The stereotypical seed-funded company may be a scrappy startup with a shoestring budget,” Glasner writes. “But in the age of AI, that’s not where investors are concentrating their bets. Instead, recent months have served as a busy period for big commitments to seed-stage companies that are short on operating history and long on ambition.”
Eze Vidra, at VCCafe, takes a slightly different tack, referencing a “distortion field” laid on the market by AI.
“For founders and investors alike, the ‘standard’ rules of the seed game are being rewritten, driven largely by an unprecedented distortion at the top of the market caused by Generative AI companies, across all stages,” Vidra writes.
More from Boston
Our Imagination in Action event in April actually had a panel on tis exact topic. Participants talked about seed money, where it goes, and ow to get it, with insights on why and how AI is revolutionizing things. Boaz Fachler of Link Ventures interviewed Rudina Sesari of Glasswing Ventures, Heena Purohit of Microsoft, Yossi Hayut of Wix and Krishna Gupta of Presto.
One of the early questions was around how investors can tell a real AI-native company wit a fundamental technology from a “wrapper” around a model or LLM that doesn’t really deliver novel use cases.
Hayut suggested that marketing plays an integral role.
“Part of it is putting it out there,” he said. “I’m okay if they’re doing their tech right, but they also need to do the marketing right, so I would look for both. And I think building in public has become such a thing these days, that you need to put the product out there and move as fast as you can while you’re building an AI-native company.”
Sesari had a different take on the word “wrapper,” while agreeing that the product of a startup should be AI-native.
“I think every company will have a wrapper if they have a prayer of existing in this day and age, in terms of just humanizing the interaction, of the UI, UX, being more prone to human-like interactions,” she said. “I think that’s checking the box, and table stakes.”
Purohit agreed.
“LLM wrappers can be real businesses that make money,” she said, “but the way we traditionally look at it, when you talk about investable and defensible companies, my tell is typically startups that want to own a particular problem or a workflow, because that’s more than just the model, expanding on the tech stack elements, in order to do that well, in enterprise environments, you need multiple models.”
Who Started It
Gupta pointed out that a founder’s identity and track record, in this game, can matter, too.
think the we live in a world where there’s like a divergence of founders, but it’s not unique, right? We’ve always had founders who are, I think, authentic to building value over a long term, and solving a problem, and then you’ve got founders who are looking to capitalize on the moment, or the momentum in front of you. It’s easier to do that than ever.”
He talked a little bit about the process.
“If you capture lightning in a bottle right now, whether it’s a wrapper or not, you can go raise money,” he said. “You can go monetize. You can sell secondary, and so for me, it’s about really understanding the founder, and reading what their real motivation is. Because someone who wants to build value for the long term will not just be building a wrapper, because they will understand that they’re solving a problem that needs to either integrate into the workflow, or needs to do something more than something that is trivial.”
Is It Dead?
Later, Fachler mentioned the oft-cited ‘SaaS-pocolypse,’ and the replacement of services with subscriptions.
“If you’re building at the last mile, I think you understand that we’re not there yet, where you can just swap one out for the other,” Gupta said.
“The exercise is not to replace humans with tech,” Sesari added. “AI is supposed to deliver orders of magnitude better outcomes. So the focus should be on the outcomes.”
Fachler asked about scenarios where vendors or third parties may just take the regular money for a service, and then use AI agents to do it, rather than humans, for extremely miniscule time and money spent.
“We also use the agents as if they are sort of synonymous with humans,” Sesari said, while at the same time explaining about different kinds of human-in-the-loop frameworks.
Hayut pushed back on the idea of a SaaS-pocolypse.
“SaaS will not die,” he said. “SaaS will have to evolve, because we’re using a lot of AI right now, so the margins are getting lower and lower, so we need to find better ways for our customers to pay for it.”
“I don’t know what the ultimate outcome will be in terms of business model,” Sesari said. “I think services are here to stay for the foreseeable future.”
Paying Attention to Cost
“Why do you think there’s a paradigm shift now?” Fachler asked the panel, referencing some of the realities around cost.
“I think with the cloud, we understood the costs better, and so the margins are very, very healthy, and there is passing of costs, but there’s also quite a bit of margin that can be captured, and therefore we don’t think of it as passing along costs,” she said, guessing at an estimated 60% to 65% in gross margins. “I think with AI, you might be considering passing along costs, because we don’t know what the ultimate cost and efficiency of these models are right now.”
She cited what my colleague Daniela Rus, director of MIT’s CSAIL lab, and others have been using with new kinds of liquid networks, and for full disclosure, I’ve been affiliated with this effort. In addition, Fachler is with Link Ventures, an organization I am also a part of.
Purohit mentioned the Nvidia GTC, noting that Microsoft has taken a page from the firm’s playbook.
“Jensen made the bold claim that every employee is going to have a token budget,” she said. “And at Microsoft, we have given Copilot Cowork to every employee who wants access to it.”
Where the Value Lies
“I think domain expertise is even more valuable today,” Gupta said. “You know, the kinds of things you can do now are not necessarily like what we’ve interacted with in the past. So I love founders that are, A, an expert in some domain, but also B, just are deeply immersed in understanding how that domain may or may not change at the last mile.”
Purohit talked about stealth, and saturation.
“There are a lot of players out there, but there’s so much innovation happening out there, across different types of modalities,” she said.
“I think there is real opportunity at the application layer, especially in the vertical setting,” Sesari added.
Let the Agents Work
“The next thing will be how to use the agents to run your businesses,” Hayut said. “If you can use the agents, and democratize that to run your business, an SMB can start running as a medium level enterprise. And this is like something that I feel will truly be the next thing, how you set it up, how you use it, how you do dynamic pricing, how they do the marketing for you. There’s going to be a lot of direct to SMB, but I think you’ll need implementers. You’ll still need implementers, because there’s a lot of non-technical people who are not interested of being AI native, or know about AI. So I think they will, they will need it.”
Defensibility with Verticals
Speaking on the idea that firms need to defend against the big players with verticals, the panelists provided a bit more on differentiation for funding and ultimate success.
“If you can own that last mile, you can somehow protect yourself,” Gupta said.
“I would agree,” said Sesari. “I would also add that it’s quite interesting in terms of the verticalization of domain expertise of founders. You know, technical expertise is almost secondary to the domain expertise. Because I think the key word, at least in this stage, for these companies, is workflows.”
“You have to be unafraid to go back to the drawing board and completely rethink the way you’ve been thinking through your product and go-to-market,” Purohit said, noting the advancement of technologies like Claude and OpenClaw over the past year. “You have to go back and incorporate that into your processes. So it’s a lot of iterating from scratch.”
“You want to start, first of all, come with a product, not a presentation,” Hayut said. “Build in public. Show what you’re learning through the process. Show us that you have a community of users, that you’re learning from them, and iterating, building and iterating your way forward. I think that’s the main thing.”
“Dare to imagine, not where you can go from state A to some in-between state, but reimagine entire verticals and industries,” Sesari added, “and secondly, find investors with capital.”
That’s a good concise way to say it. There was a bit more in the panel, but I thought this provides a good survey of the knowledge that this crew has on VCs, startups, and funding. Stay tuned.
