AI's Infra, Application Layers are Converging: Boundless Ventures' Natasha Malpani

I mean, that's the job, right? First is sourcing: if thousands of companies approach you, how do you find the outlier teams? Second is selection: how do you decide what to invest in? Third is portfolio management: how do you support founders to achieve the best outcomes?

In terms of selection, there is no single right answer. Every investor has a different framework. At Boundless Ventures, we focus on multiple aspects. First, we are founder-first. We spend significant time understanding founders, their background, psychology, thinking, and past actions. We do not rely only on standard metrics like education or work experience. Our portfolio includes founders who are former CROs of hospitals, CMOs of companies like Ola Electric, as well as university dropouts with no work experience. There are founders from IIT Kanpur and IIT Bombay, people who have worked at Unacademy or spent a decade at ISRO, and others who started companies in college. There is no single visible pattern, so we focus on mindset.

We look for belief, the ability to shape the world, a unique vision, and resilience, the capacity to improve after setbacks. We also assess whether founders combine technical depth and customer understanding with the ability to sell. Selling is storytelling: convincing employees, investors, and customers at every stage. At the early stage, founders have very little and must create their own reality, so we look for people who can do that.

Second is the market. We spend time at the frontier because we are a specialized fund, an AI-native, cross-border fund that invests early, doubles down at each stage, and focuses on consumer AI, vertical AI, and deep tech. We engage with universities and labs, including Stanford, IISc, and IIT Madras, to understand developments in areas like physical AI, robotics, space tech, and world models. This is how we form market views.

As a result, our portfolio is differentiated, spanning areas like space, travel, AI security, and healthcare, along with emerging work in fields such as immunology and world models.

Finally, there is traction. At our stage, we are not focused on unit economics like a growth investor. However, raising money purely on an idea is no longer sufficient. Since anyone can build an MVP today, we expect founders to have built one and demonstrate early validation.

As a cross-border fund, we evaluate companies globally, which gives us a clear view of different markets. In India, we are particularly bullish on consumer AI and full-stack physical AI.

One contrarian view we hold is that AI infrastructure and the application layer are converging. Many of our consumer companies are not just building on top of models like ChatGPT or Claude; they are also solving for memory, context, security, and governance. They are effectively building both infrastructure and applications. Similarly, companies like Alter, which focus on AI agent access, security, and identity control, are also building infrastructure alongside applications.

We believe the winners at the application layer will have infrastructure built into them. AI cannot be deployed in the real world without reliability, and that reliability must be engineered. Having access to powerful intelligence alone is not enough. The most successful outcomes come from systems that are reliable, trustworthy, and dependable. The same principle applies to AI, intelligence must be translated into real-world impact.

The consumer versus enterprise debate is a false one. There is opportunity in both segments. In consumer AI specifically, I am bullish. We have an exclusive program with Anthropic that we believe will help surface the next generation of AI breakouts.

Consumer AI is still very early globally, with few clear winners. You could argue that ChatGPT and Claude are the primary consumer AI breakouts. It is notable that foundational model companies themselves are among the largest consumer-facing players. However, at a vertical level, across sectors like health or travel, there are no clear breakout winners yet.

No, no, nothing like that. They'll go off to all markets. Claude began with an enterprise focus and is now moving into consumer, while ChatGPT started as a consumer product and is now expanding into enterprise. At that scale, the distinction becomes artificial. Once a company has sufficient capital, strong technology, and a viable business model, it can move across segments. The same is true for companies like Microsoft, Apple, and Tesla, which operate in both consumer and enterprise markets.

However, at the start-up stage, founders must make choices. It is not practical to target both segments simultaneously. Globally, consumer AI is still early, with few vertical breakouts so far. That said, this is likely to change. The absence of clear winners today does not mean they will not emerge.

In consumer AI, the path to success remains similar to traditional consumer tech. Companies must deeply understand their users, build around their workflows, and create sticky products with strong habit loops and retention. What AI adds is the ability to leverage context and memory. The more a product understands a user, the more personalized and engaging it becomes. This level of depth and personalization was not possible before generative AI, and it can make products extremely difficult to switch away from.

This also changes how sectors like education evolve. Future education companies may not resemble traditional models. With generative AI, it becomes possible to fully personalize learning, understanding a student’s current knowledge, identifying gaps, adapting to their learning style, and tailoring both content and format. This level of customization can significantly improve outcomes, and that is where the opportunity lies.

The same as consumer tech, Users will pay for usage, engagement, outcomes, or subscriptions. Strong products can monetize effectively, as seen with Pocket FM and AstroTalk. Across content, education, and astrology, companies have demonstrated that Indian consumers are willing to pay for the right product.

This is what makes India interesting: a large population combined with a proven willingness to pay for high-quality offerings. In AI-driven learning, especially language learning, there are already strong breakouts like SpeakX and Supernova. Monetization is not the core challenge; it is already validated.

The real challenge lies in the infrastructure layer. This is why strong technical teams matter. AI enables more creativity, products can be multimodal, explore new formats, and deliver experiences that were not possible earlier. However, technical depth is now essential. In the past, consumer companies could rely on strong product sense, psychology, and marketing while outsourcing technical work. That is no longer viable. Without capabilities like memory and context, even a strong product will struggle. The product must continuously improve, and that depends on technical excellence.

In consumer, there are three key buckets: distribution, retention, and monetization. Distribution is about how many people know you exist. Retention covers stickiness and habit formation, how often users return, how much they rely on your product, and how strongly they associate with it. Monetization focuses on willingness to pay and how the business scales.

The core challenge is building stickiness. With the explosion of content and apps, the real competition is for user attention. You are not just competing with another app; you are competing for a person’s limited time. The goal is to earn mindshare, getting users to depend on your product.

That is what we look for. Across our companies, whether it is Shram, Glide, or our edtech investments, the focus is on capturing user mindshare and building strong user affinity. The approach varies by sector. What works in travel differs from education, and both differ from proactive agent-based products like those Shram is building.

Structurally, if you break down funding rounds, it is the opposite. Growth rounds, by nature of funding size and quantum, are larger, so more money goes into them. However, if you look at the number of rounds rather than the amount of funding, there are far more rounds at the seed stage. This is true anywhere in the world. More people start up, and there is always a funnel determining how many graduates to follow-on funding.

Sometimes founders decide that a certain number of rounds is enough; not every start-up needs to go to Series D, E, or F, depending on the business model. In other cases, they do not progress because outcomes do not materialize. It is a tough funnel globally, and the bar for raising Series A, B, C, and beyond is only getting higher. As more companies enter, competition to advance increases, and at every stage, the bar for raising money rises.

At the same time, business models are changing due to AI, so not all companies need to raise capital. That is one level of the answer. The second is that there are many more seed funds than growth funds, which is simply a function of how the market operates, similar to the rest of the world.

No, it's exactly the opposite. The companies are back. The better the models become, the better it is for application-layer companies. The core thesis is that models will continue to improve, compute will get cheaper, and intelligence will become commoditized. We already have access to extremely powerful models, and that capability will only expand.

The question then is how companies at the application or physical AI layer bring that intelligence into the real world in a trustworthy, reliable, personalized, and sticky way. Foundational model companies are unlikely to solve highly specific, vertical problems. For example, a company like OpenAI or Anthropic is unlikely to focus on niche use cases such as imaging the earth’s oceans like Piersight, building specialized robotic systems like Armatrix, or creating highly tailored AI mentorship platforms for specific exam preparation.

Even in areas like travel, where models may understand user preferences and context, vertical companies differentiate through proprietary data, specialized workflows, and deeper integration. These companies access or generate unique datasets that foundational model providers either cannot access or do not prioritize. They also build the pipelines and infrastructure required to continuously improve their systems, capabilities that are outside the scope of foundational model companies.

The overarching thesis is that as foundational models improve, specialized companies benefit. They build verticalized solutions, leverage unique data, and create differentiated products on top of increasingly powerful underlying intelligence.