Umesh Padval, Managing Partner, Seligman Ventures – Interview Series

Umesh Padval brings decades of experience across semiconductor infrastructure, enterprise technology, and venture capital. Before transitioning into venture investing, he served as CEO and president of C-Cube Microsystems, helping scale the company before its acquisition by LSI Logic. Over the course of his investment career, he has been involved with a number of prominent technology companies spanning AI, cloud infrastructure, cybersecurity, and networking, including investments in companies such as Cohere, Harness.io, ThousandEyes, and Isovalent. He has also held board positions at several public technology firms, including Mellanox, IDT, Monolithic Power Systems, and Impinj, reflecting a long-standing focus on foundational infrastructure technologies that underpin modern computing and AI systems.

Seligman Ventures is the thesis-driven venture capital arm of Seligman Investments, a technology-focused investment platform managing roughly $30 billion across public and private markets. Led by veteran investor and former CEO Umesh Padval, the firm launched with a reported $500 million allocation focused on backing early-stage through pre-IPO companies in sectors such as AI infrastructure, cloud computing, cybersecurity, and modern data center hardware. The platform differentiates itself by combining deep public-market research expertise with venture investing, allowing it to identify long-term technology shifts early while supporting founders building foundational infrastructure for the next generation of enterprise AI and computing systems

You recently launched Seligman Ventures as the venture arm of a $30 billion public markets firm. What was the original vision behind founding this platform, and what gap did you see in the AI and infrastructure investment landscape that others were missing?

We saw an opportunity to build a platform that bridges early-stage innovation with public-market scale. Seligman Investments has spent decades investing in technology companies through the public markets, and we believe that perspective around durability, fundamentals, and long-term value creation is often missing at the early stage.

At the same time, AI is driving a once-in-a-generation re-architecture of infrastructure across compute, networking, cybersecurity, and data centers, and we felt the market was underweight investors with deep technical conviction in these areas. We support visionary technical founders with our network and expertise, and our vision is to be part of their journey from seed to IPO and beyond.

You invest across AI infrastructure, cloud, and modern data center systems. How do you see the AI infrastructure stack evolving over the next three to five years, and which layers are most likely to generate the most value?

We expect the center of gravity to shift from model training to large-scale inference and production deployment. That shift increases the importance of efficiency across the stack from silicon and memory to networking and orchestration. The entire stack is getting rewritten from chips to middleware to vertical AI applications. Each layer here has a large size of prize leading to outsized winners. While models are important, we believe a disproportionate amount of value will be created in the infrastructure layers that enable performance, cost optimization, and scalability in real-world environments.

Many investors describe AI as a foundational shift similar to electricity. From your perspective, what parts of the current AI cycle are overhyped, and which areas remain underappreciated?

Imagine intelligence as a utility, accessible to anyone, with the power to make any software application, hardware appliance, or virtual environment intelligent. This is already happening. AI will have a massive impact on society, the nature of work, and human productivity.

There is a tendency to over-index on the application layer and the speed at which new AI products are being launched. What remains underappreciated are the physical and infrastructure constraints beneath the surface such as power, thermals, memory bandwidth, data movement, and supply chain limitations. These factors will ultimately determine how far and how fast AI can scale economically.

With a focus from seed through pre-IPO, how do you determine whether an AI infrastructure startup has true long-term defensibility rather than short-term momentum?

We start with the team. Technical founders with inspiring life stories often have the ability to make the impossible possible. Time and again we see research papers claiming something isn’t feasible while founders are building prototypes that prove otherwise.

We spend a lot of time distinguishing between velocity and durability. True defensibility in AI infrastructure comes from deep technical differentiation, workflow insight, and integration into critical systems, not just early traction. The strongest infrastructure companies become deeply embedded into the stack, where switching costs, performance advantages, and ecosystem integration create long-term staying power.

Modern data centers are becoming central to AI scalability. What architectural shifts or innovations do you believe will define the next generation of AI-native infrastructure?

Modern data centers are being rearchitected around AI workloads. This includes higher power density at the rack level, new approaches to cooling, and tighter integration between compute, memory, storage, and networking.

We believe the next generation of AI-native infrastructure will be defined by systems-level design and co-optimization across hardware and software rather than isolated component innovation.

What qualities do you look for in founders building AI infrastructure companies, and what differentiates those who go on to build category-defining businesses?

The founders who stand out combine deep technical expertise with a systems-level understanding of how infrastructure evolves. Many of these markets require long development cycles, complex ecosystem partnerships, and close engagement with customers, so resilience, patience, and the ability to execute over multiple years are critical.

The best founders are able to combine technical vision with operational discipline and build enduring companies through multiple technology cycles.

How should founders approach building for a future where compute, energy, and data locality are becoming as critical as software?

Founders need to build with constraints in mind. Compute, energy, and data locality are no longer abstract considerations; they are first-order design parameters.

The most compelling companies are those that treat efficiency, scalability, and operational realities as core product requirements from day one rather than as optimizations added later.

There is growing concern that valuations in AI are outpacing fundamentals. Where do you currently see the biggest disconnect between pricing and real value?

We are in the middle of a once-in-a-lifetime technology shift. The winners here could become larger than many of the biggest public companies today.

We do see pockets where valuations are ahead of fundamentals, particularly in areas with low barriers to entry. At the same time, some of the most technically complex infrastructure layers, especially those solving hard engineering problems with long development cycles, remain underappreciated relative to their long-term strategic value.

From your vantage point, how is enterprise demand shaping the next generation of AI platforms, particularly in areas like cybersecurity and cloud infrastructure?

AI is now a board-level mandate for enterprises of all sizes. Teams are prioritizing AI tooling to help them become materially more productive, and many proof-of-concepts are now moving from experimentation into production environments.

That shift creates new challenges related to scaling, observability, reliability, and security. AI is simultaneously expanding the attack surface while creating the tools needed to defend it. AI agents can be attacked externally, vulnerabilities can be introduced during development, and enforcing policies at runtime becomes significantly more complex in probabilistic systems.

This requires new architectures, new operational models, and new ways of thinking about security and infrastructure. Enterprise demand for reliable, scalable, and secure AI systems will continue to drive significant innovation across the stack for years to come.

Looking ahead over the next decade, what does the long-term future of AI infrastructure look like, and which technologies or trends today give you the strongest signal of that future?

Over the next decade, AI infrastructure will become as foundational as cloud computing is today. The data center and energy grid are not yet fully prepared for a world where AI becomes mainstream across industries and daily workflows.

Along the vectors of cost, energy efficiency, token consumption, hardware heterogeneity, and data movement, there is still substantial work required for an agent-native world to become reality. We expect to see greater specialization across the stack, tighter integration between hardware and software, and continued focus on efficiency and systems-level optimization.

The companies that win will be those that can deliver performance at scale while navigating real-world constraints around power, cost, reliability, and supply.

Thank you for the great interview, readers wishing to learn more about this VC firm should visit Seligman Ventures.