On breaking into venture capital, investing in AI, and what really matters at the early stage
Introduction
As part of our investor interview series at Big Red Ventures, I caught up with Vikram, who is an investor at Cota Capital. What started as a casual reconnect quickly turned into a deep dive into his journey from consulting to venture capital, how Cota Capital operates, and how he thinks about AI, hardware, and early-stage investing.
Anirvan: You spent several years at BCG before pursuing an MBA and eventually moving into venture capital. At what point did you decide VC was the path you wanted to take?
Vikram: I genuinely loved consulting. I enjoyed the dynamism of working on new problems every few weeks, across industries, functions, and sometimes geographies. And I loved the people component: working closely with smart colleagues and clients on every engagement. When I was evaluating career opportunities during my MBA, VC felt like a natural extension that preserved what I liked while addressing a few of the elements that I didn’t.
The biggest of these was having more skin in the game. In consulting, once a project ends, your involvement with the client largely wraps up too. In VC, the work truly begins after the investment is made. You’re working closely with portfolio companies over a long horizon, and the outcomes are deeply tied to the decisions you make together. That was very compelling to me.
That said, I wasn’t 100% certain. I had never actually experienced VC. So the MBA became an opportunity for me to test the hypothesis. I used my time at Booth to try out a range of internships. I started with an early-stage VC internship at Jump Capital through the VC Lab at Booth, which gave me exposure to sourcing, diligence, and sitting in on IC meetings. Then I deliberately pivoted to something very different, a marketing role, to pressure-test my assumptions. While I liked my colleagues and gained a lot from the experience, the work didn’t excite me the way consulting and VC had. I then joined the 49ers Investment Team, which was a multi-stage firm doing both early-stage VC and late-stage private equity. I found the sourcing and diligence work on the early-stage VC side energizing, and also loved the operating work I did with the PE arm. By the time I went full-time, I had actually proved out the thesis I had coming in, while also shedding some of the misconceptions along the way.
Anirvan: You mentioned that a major part of the work begins after you’ve made an investment. Can you walk us through how you split your time across sourcing, portfolio support, and everything else?
Vikram: On average, I’d say about 40 to 45% of my time goes into sourcing and diligence: identifying new opportunities, attending conferences and demo days, responding to inbound companies, and running through the full diligence process, including meetings and data rooms. Another 30 to 35% goes into working with portfolio companies. About 10 to 15% is what we call research: trying to identify new investment theses, speaking with domain experts, and understanding what problems are emerging or going to emerge. The remaining 5 to 10% covers fund administration, LP-facing work, and brand-building activities like speaking at events.
On the portfolio side, Cota tends to be very hands-on. We typically prefer to lead or co-lead rounds, and that comes with a commitment to add real value. After an investment, we usually kick off with a workshop with the portfolio company’s senior leadership to understand their near-term priorities, overlay our own learnings from research and our advisory network, and build a plan for what we’ll focus on together.
Our support broadly falls into three areas. First, sales and marketing. Our team has significant experience helping companies navigate complex enterprise and government sales cycles. We help build the sales strategy, understand customer segments, work on positioning and collateral, and often sit in on sales calls or even run them ourselves, if the founder so desires. On the marketing side, since early-stage startups rarely have dedicated marketing resources, we bring in our Chief Marketing Officer and other experts to work on branding, pricing, and website and collateral strategy. Second is organizational development. As companies scale, one of the hardest challenges is growing the team in a way that stays ahead of growth. We help design the right org structure and recruit from our network or through partners. Third is internationalization, whether it’s expanding sales into new geographies or, for hardware companies, setting up contract manufacturing and building supply chains. Apart from this, we also (as with most VC firms) support future fundraises.
Anirvan: What is Cota’s investment thesis, and how do you think about stage and sector focus? Do you apply any hard revenue thresholds?
Vikram: We invest from pre-seed through Series A, with check sizes ranging from $1 million to $10 million, and we prefer to lead or co-lead. In terms of sectors, we focus on enterprise tech and deep tech. That spans a fairly broad range: everything across the AI stack – models, infrastructure, verticalized and horizontal applications – as well as hardware and hard tech, cybersecurity, DevOps and DevSecOps, and other infrastructure plays.
We don’t apply a rigid revenue bar. It varies significantly depending on what you’re building. For companies that are more distribution-focused – verticalized applications without a strong proprietary data or product moat – we’re often more revenue-sensitive. For deep tech or highly differentiated companies, we’re far less focused on revenue at the time of investment. The bar shifts based on the nature of the defensibility, as well as other factors including the investment round, market size, and more.
In terms of the thesis, we look for what we call “net new”: either a new problem that didn’t exist before, or a new technology that enables solving a problem that was previously unsolvable. For example, as AI models proliferate, you now need to secure not just data and applications, but the models themselves – prompt injection, jailbreaks, and so on. That’s a genuinely new problem. Similarly, as data centers scale and hardware improves, networking requirements around bandwidth, throughput, and latency are categorically different from even five years ago. Solutions built for that world are truly net new.
The second pillar of our thesis is being problem-first and product-first. We want to deeply understand what the core burning problem is, why it couldn’t be solved before, and how urgently the market needs it. On the product side, we spend meaningful time early on understanding the technical architecture, the differentiation, the defensibility, and how data flywheels actually develop and compound over time.
Anirvan: Product-market fit is a phrase that gets used often. How do you assess whether a company has achieved it, or is on a credible path to it?
Vikram: First, a caveat: since we also invest at pre-seed and seed, many of our companies haven’t yet achieved product-market fit at the time we invest. But the question is still relevant because these companies need to get there in the months and years that follow – and most usually do!
We think about product-market fit across three dimensions. The first is the level of pull versus push. Is there meaningful inbound interest? Are customers spreading the word without being prompted? That organic demand signal is far more meaningful than a highly efficient sales team going outbound. Second, depth of engagement. Once the product is deployed, how much are customers actually using it? Are they using it frequently, for hours a day, building workflows around it? That sustained engagement is a strong indicator. Third – and this one is fuzzier – price sensitivity. If customers are genuinely solving a burning problem, they tend not to nickel-and-dime on price. If every deal involves someone trying to shave a small amount off the invoice, that’s usually a sign that the product is nice-to-have rather than must-have.
Anirvan: Can you walk us through the steps of your investment process, from first meeting to final decision? Are there any hard criteria that can disqualify a company outright?
Vikram: The process typically takes two to three weeks, though it can move faster for competitive deals. We’ll usually have two to three meetings with different members of the Cota team, and we generally want the CEO present for most, if not all, of them, along with the CTO or engineering and product leads for at least some.
The first call is about getting to know each other, covering the broad contours of the company, and building an initial view. Subsequent conversations go deeper into the areas we find most exciting or most uncertain, typically starting with the problem and product, and then depending on stage, the go-to-market strategy, financials, and unit economics.
In parallel, we conduct our own market research: competitive landscape, existing alternatives, and the broader tailwinds and headwinds in the space. If the product requires deep domain knowledge that we don’t have on the investment team, we bring in our advisory bench to weigh in on the technical architecture or go-to-market approach. Toward the end of the process, we try to do a few customer interviews where relevant.
As for hard checks, there aren’t many absolute gates. The closest thing to a hard constraint is the “net new” criterion. We’re generally not excited by incremental improvements on existing solutions. We’re looking for something genuinely disruptive, whether that’s solving a brand-new problem or applying a new technology to a problem that was previously intractable. That’s the filter we hold most firmly, even if it’s somewhat subjective.
Anirvan: Cota invests across the AI stack. How do you evaluate companies in this space, from hardware and networking to infrastructure and applications? And are there any layers you deliberately avoid, such as hardware?
Vikram: We’re genuinely broad across the AI stack: hardware, software, infrastructure, verticalized and horizontal applications, as well as the ecosystem that plays around AI like networking and security. The one area we do least of is foundation models, primarily because the capital requirements are usually so significant that they would need to raise several hundreds of millions in investment before the product is even market-ready.
On hardware specifically, we actually lean into it. We’re particularly excited by solutions that blend hardware and software, or software that’s deeply integrated at the hardware level. That’s precisely because those companies tend to have more durable technical or product advantages. There can also be meaningful supply chain moats. And frankly, it’s an area where we can add more value. We have the expertise and the network to help portfolio companies navigate contract manufacturing, build supply chains effectively, and expand internationally. We want to be the right fit for the companies we partner with, not just a passive check.
Anirvan: You’ve written an article about physical AI. What are the most significant problems you believe physical AI will solve, and where do you expect to see the fastest adoption?
Vikram: Physical AI is interesting to us precisely because it fits our “net new” lens. It enables things that simply couldn’t be done before. Let me give a few examples.
In industrial settings, you’ve had telemetry, sensors, and monitors that can tell you something is wrong for several years now. What they can’t tell you is what is wrong, why it happened, and what to do about it. Physical AI bridges that gap. By combining multimodal data – visual, audio, time-series, telemetry – you can build systems that have true context about a failure or anomaly and can generate meaningful, actionable responses.
A related shift is from point-in-time sampling and reactive responses to continuous, proactive monitoring. Instead of sampling a manufacturing line and catching defects after the fact, you can now track every product and every process end-to-end in real time.
The third piece is autonomy, and this is where it becomes particularly impactful from a labor perspective. The US faces a significant shortage of truck drivers, for instance. Autonomous logistics doesn’t displace workers who are there, it fills roles that couldn’t be filled. The same logic applies in healthcare, where there are massive shortages of nurses and other workers, and in manufacturing. Physical AI can begin plugging those gaps.
Perhaps most importantly, physical AI opens up settings that were previously inaccessible to software. In healthcare, you couldn’t deploy a reactive, point-in-time system in an operating room. Now, with full contextual awareness, you can support surgical visualization in real time. One of our portfolio companies, Activ Surgical, is doing exactly that. In defense, you can’t afford sampling or gaps in visibility. Atomathic’s radar work is another example of what’s now possible. The near-term adoption leaders are warehousing, logistics, and manufacturing, but the trajectory into healthcare, defense, and beyond is already underway.
Anirvan: Finally, how is Cota using AI in its own day-to-day operations? And do you think AI will eventually be able to influence high-stakes decisions such as follow-on investments, fund structuring, and the like?
Vikram: We’re not yet at the stage where AI is making or significantly influencing critical investment decisions, whether that’s a new investment, a follow-on, or how we think about fund structure. But it’s delivering real productivity gains across several areas.
The most obvious is research. Rather than combing through dozens of reports and search results, AI tools help summarize and contextualize information quickly. You still need to go back to sources, but they help you get oriented faster and be more targeted in what you read in depth. That applies equally to thematic research, pipeline evaluation, and portfolio work.
On sourcing, some firms, Rocketship, my previous employer, for example, have built proprietary data-driven sourcing algorithms that help surface high-potential companies from a large universe. Over time, as these tools accumulate more context, seeing what we’ve invested in, what’s worked, what hasn’t, how we’ve evaluated similar situations, I do think AI could become useful for more complex tasks and potentially some decisions. But I think there’s still meaningful ground to cover before that’s the case.