Investing in Generative AI with Lotan Levkowitz

Lotan Levkowitz

In the high-stakes world of early-stage investing, few navigate the biases and fallacies that plague your critical thinking. From pareidolia to confirmation and survivorship bias, picking sprinter-winners in the first few meters of the race is nearly impossible. Look beyond the horizon, lock in your bets and swing on the long-run pendulum. Convince your Limited Partners and board that your investment thesis is right and not a misread opportunity or errant gut feeling. That’s the role of early stage investors, a job for those who must stomach the risk. Of course, there’s rewards to reap, but only for the prescient and patient. Lotan Levkowitz is a General Partner at Grove Ventures and his job very much epitomizes that characterization. He’s a bona fide data-driven investor and his active portfolio spans six Israeli startups (with additional investments he led soon to be revealed). Now he’s on the lookout for the next Generative AI unicorn and we sized him up in an interview to understand where his next bets will be cast. 

Grove Ventures raised $185 million for their third fund, closed last year and focused on Seed to Series A funding rounds. Managing over $500 million in assets are General Partners Dov Moran, Lotan Levkowitz, Lior Handelsman and Renana Ashkenazi. They’ve notably backed Williot, ActiveFence, Navina, and NeuroBlade, among others, focusing on wide ranging sectors, from software infrastructure to deep technologies like semiconductors for data centers and outer space computing, to next-generation bluetooth communication and IoT sensors. 

Like many, Levkowitz is unequivocal about the transformative power of Generative AI. His insights offer a comprehensive roadmap for anyone looking to navigate the complex and rapidly evolving landscape of Generative AI. From investment strategies to real-world applications, his perspectives serve as a valuable guide to building a successful startup that fits the Venture Capital investment model.

Has Generative AI influenced your investment strategy? How do you approach such a revolutionary technology to capitalize on seeding the next unicorn, but also avoid quick shots from the hip?

“When we started our fund, our overarching thesis was around data and the transformation it’s having on the world. Now it may seem obvious, but it was still in its infancy. Our thesis was built on three pillars. First, the creation of data, particularly in edge IoT and the real world. This is where data is being generated at an unprecedented scale, from sensors to user interactions. Second, data infrastructure, which includes the software and hardware needed for storing, analyzing, and managing data, like data pipelines, DevOps, and DevTools. Third, Artificial Intelligence/analytics – making sense of the data through vertically integrated solutions, turning raw data into actionable insights that can disrupt industries.”

“We faced a lot of skepticism initially. The term “AI” was often associated more with Steven Spielberg movies than with real-world applications. We even coined the term “applied deep data” to make our focus more palatable and digestible to LPs.”

“Generative AI is not just a buzzword; it’s a seismic shift in technology, and we evaluate its impact through three lenses. Its effect on our existing portfolio’s daily operations. We’re talking about real-time changes that can make or break a business. The influence on market dynamics: including market size, competition, and the overall landscape in which our portfolio companies operate. And finally, the new doors it opens for fresh investments. Generative AI is creating entirely new subfields and opportunities that didn’t exist before.”

So how has it impacted your portfolio companies? Does it level the playing field for newcomers? It’s an easier process today to integrate LLMs or train your own models, anyone could enter a market today and make a similar offering to what’s out there.

“For the application layer, for the most part, it’s really a means to an end. Datasets and product market fit are the basis upon which Generative AI can boost your offering and service your clients better. Generative AI can certainly bolster our portfolio companies’ offering. But it’s a double-edged sword: it can also exacerbate the problem they attempt to solve.”

“Take Navina, for instance. They assist physicians by sifting through mountains of historical medical records to present only the most relevant data so that the physician can make the best possible decision. This not only aids physicians but also leads to healthier patients and a more efficient healthcare system. The physician actually gets better compensated in that case, resulting in a win-win-win. Generative AI has supercharged their capabilities, allowing them to provide even more precise summarizations. They already have a monstrous proprietary dataset. It’s cleaned and structured in the right way. And the addition of Generative AI capabilities is just a killer feature as a part of their product. They already knew their customer, knew their needs and became integrated into the physician’s workflow. It’s hard for newcomers to penetrate that, even with Generative AI LLMs. This is one example of a company that has product-market fit, has data, and Generative AI enhances those capabilities.”

ActiveFence is another case. They’re essentially the guardians of the online world, protecting platforms from harmful behaviors, like misinformers, sexual predators, election integrity compromisers and the like. They built what they call a “database of evil,” updated multiple times a day. They generate insights from this database and inform their customers to help them stop these bad actors from participating in the digital world. Generative AI has significantly exacerbated their need and at the same time enhanced their ability to extract insights from this database, making the online world a safer place. Content is easier than ever to produce and distribute. Going back to the dawn of humanity, we started writing on stone, then writing on paper and books, and now the internet. Distribution started with pigeons, to newspapers in print to online forums, and now automatic website creation with best SEO practice. The barriers that once existed have been obliterated. Generative AI can now fool us and distribute news indistinguishable from the real thing. The computer can write by itself. Now, because of the way LLMs are structured, funny enough, they’re very good at giving you the average answer.”

“What Activefence excels at is going to the niche areas, the long-tails of data of bad behaviors. And they excel at identifying these edge cases because their dataset is rich in the long-tails and edge cases. They also have a unique data collection methodology in sourcing text. So with Generative AI, the problem has certainly grown, but Activefence’s unique positioning enables them to leverage the technology and combat bad digital behavior even better. They’re data experts and the data explosion, thanks to Generative AI, has made them relatively better.”

“If you have proprietary datasets, knowledge and understanding of your customers, and product market fit, I think Generative AI will really accelerate your capabilities and the value you’re providing to your customers.”

“Every company leader today should understand the following questions: are our customers dramatically affected by Generative AI; can we really expand the value we’re delivering to our customer using these new Generative AI architectures; and can we improve the way that we’re developing our product with Generative AI. These are three answers that you must be opinionated about.”

What’s your approach to investing in Generative AI startups? What do you look for and what are you seeing in your deal flow?

“Generative AI is evolving so rapidly that what’s relevant today may be obsolete tomorrow. This makes early-stage investments extremely difficult. All the world’s focus was changed in a year with one very deep technology, and it keeps changing. Everybody is chasing it, for better or for worse.”

“When we look at new investments, we divide it into two segments. Generative AI infrastructure and Generative AI applications. Infrastructure has always been and still is a big part of our investment thesis, like data engineering, software engineering. It’s from the infrastructure itself, database centric and the operating piece on top of that. Like how to clean the data, how to monitor it, how to connect it, all of that. This is a big pocket of what I’m seeing in the new companies seeking capital. It’s people creating the next tools, the next capabilities of Generative AI tools. And the need is growing because so many companies are running as fast as they can to make use of Generative AI technology, but they’re improvising a lot. It’s a lot of early adopters using open source software, prompt engineering… and because the capabilities of Generative AI are changing at a record pace, if you use one model today, by the time you finish implementing it, you see the next best thing is already out there.”

“But I think that the biggest beneficiaries of the first generation of Generative AI, as it relates to infrastructure, are already out there. In our portfolio at least, the relevant infrastructure companies such as Deepchecks and Unifabrix are ready. They were in the right place at the right time when the storm started. The challenge in infrastructure for Generative AI today is that it’s going to be very hard for you to understand which problem you’re solving. Because when you talk to your customers today, return in two months and the problem is already different. It’s really tough to understand your competitive landscape because the market is moving at a unprecedented speed. So I believe that actually now is one of the most challenging times to do very early stage investments in infrastructure for Generative AI. Everybody is chasing the same target and because of the nature of infrastructure, it’s either one platform rules them all, like OpenAI or HuggingFace, or it has a lot of layers of integrations to create the next best thing. I’m not sure that now, if an infrastructure company starts today, they’ll understand the other lego parts around them.”

“Software infrastructure founders should be very opinionated of what’s out there. Trying to go too wide on many use cases and needs will make you run into vicious competition. My job is to pick the winner in a sprint in the first few meters. You might see traction in the Series A or B round. And there might be an Israeli winner in the software infrastructure space, but there will be many losers because they didn’t get funded and the landscape is already filled. Pinecone is a good example of being a bit earlier than others. They had the right intention at the right time: starting a vector database in 2019 vs 2023. It sounds like nothing, but those four years at that time means everything.”

“For hardware infrastructure, Nvidia is the biggest winner in Generative AI, and they’ve been ready for over a decade. There’s an opportunity to take market share from Nvidia considering people are lining up to buy their products, making them wait. Demand is over supply. But new companies won’t be able to penetrate that market, it will be existing companies with similarly long history of operations.”

What about foundational model startups?

“I actually had a chance to know Ori Goshen (Co-Founder and Co-CEO at AI21 Labs) early on as his wife Noga is a friend of mine. He was talking about building a startup in the realm of language processing, amazing foresight for the future. Seeing his journey is a great inspiration for Israeli founders that want to be one step ahead of their market. Ori’s a great founder and it’s really rewarding to see him flourish. But we won’t see other foundational model startups created and rise to the top like AI21 Labs has; training costs are just too capital intensive at this point.”

How do you evaluate new opportunities in the Generative AI space? 

“In a simplified formula, we boil it down to two factors: their success in distribution multiplied by their product quality. Many startups have groundbreaking products but no answer on how to get significant  distribution. They find themselves at a steep-inclined battle against incumbents. My investment model is geared towards category leaders—those who can dominate a particular niche or create a new one – who benefit from considerable product quality and distribution wherewithal together.”

“For those with an answer to distribution, it usually comes in two flavors: an area where no incumbents operate, or a full-stack solution. Instead of a solution like software as a service, it would be a knowledge worker  in a box. Think of automated sales representatives, automated call centers, automated software engineers, automated quality assurance… They’re doing work that is currently done by humans manually, but they can take on the same job, recruited theoretically with half the salary and enjoy SaaS profit margins.”

“I’m also interested in startups that integrate Generative AI, but aren’t Generative AI startups per se. They have a great offering and added Generative AI as a killer feature. But without it, they’re still a great startup.”

How has Generative AI impacted the value of data? Using ChatGPT and for example, coupled with low-code automation tools, anyone could gain access to public datasets, clean them and structure them for training a model.

“It’s true, Generative AI has democratized access to good data, but it’s also raised the stakes for what constitutes premium data. In marketing, for example, consumers are becoming increasingly savvy. They can spot automated, Generative AI based outreach a mile away. This has made high-quality, personalized data more valuable than ever. Good data is available through Generative AI techniques, but premium data isn’t. And those that get premium curated data are steps ahead of the competition. How can you leverage access to premium data? You can perform high cost laser focus actions such as going to your customer in person, talking with them, learning their preferences and situation. It’s archaic and reverting to the door-to-door method, but it’s superior now and it will result in building a better model or product. At least a cut above that built on Generative AI data collection methods.”

Fortune indeed favors the bold, but boldness alone is not a strategy. Lotan Levkowitz’s approach is a blend of high-risk taking and calculation. “In the next 5 years, there will be more earthquakes similar to that of ChatGPT and GPT-4’s impact. You must be opinionated on what’s out there. The world is changing, and you must react,” Levkowitz advises.

A testament to his data-driven investing style and maintaining a market pulse, Lotan Levkowitz recently authored a comprehensive report on the changing landscape of the software development lifecycle. The report, ‘Shift Happens: The State of Software Development Life Cycle 2023,’ demystifies the shifting paradigms of the changing industry, conducted through 70 in-depth interviews with industry experts. It uncovers key trends and challenges that are shaping the future, including the decline of product-led growth, a cautious approach to the adoption of Generative AI and a significant shift in budget allocations towards DevOps and infrastructure tools.

The report ‘Shift Happens’ can be read here.

Mentioned in the article

Continue Reading, the hub of the AI startup ecosystem.

Add or Claim Your Profile

Submit the form below to add or claim an existing profile. Within 24 hours, we'll review your request and connect you to your existing profile for full editing.

Pro Lite Subscription Required

Unlock all search results

With a Pro Lite subscription, you can enjoy unlimited profile viewing. Power your innovation efforts, uninterrupted.


cancel anytime

Pro Lite includes: