Episode 3: Mitesh Agrawal CEO, Positron AI

Artificial intelligence is growing faster than almost any technology wave before it. But behind every AI agent, chatbot, image generator, content tool, and enterprise application, there is a deeper infrastructure challenge: chips, data centers, energy, and cost.

In this episode of The Disruptors, Mitesh Agarwal, CEO of Positron, explains how his company is building custom silicon designed to make AI inference faster, cheaper, and more energy efficient.

Mitesh Agarwal is building Positron inside one of the most important markets in technology: AI infrastructure.

While most people experience artificial intelligence through tools like ChatGPT, image generators, AI assistants, or automated workflows, Mitesh focuses on what makes those applications possible behind the scenes. Every AI application needs massive computing power. As adoption grows, the demand for chips, data centers, electricity, and infrastructure continues to rise.

That is where Positron enters the conversation.

Mitesh explains that Positron is building custom silicon, or specialized AI chips, to accelerate AI applications more efficiently. Unlike general-purpose GPUs, which are designed to handle many different types of workloads, Positron is focused on a specific part of the market: inference.

Inference is the stage where trained AI models are actually used. Every time someone asks a chatbot a question, generates an image, uses an AI agent, or runs an AI-powered application, inference is happening. As AI moves from experimentation into daily use, inference could become one of the largest and most important infrastructure markets in the world.

A major theme of the episode is how Positron competes in a market dominated by Nvidia.

Mitesh is clear that Nvidia is one of the smartest and most powerful companies on the planet. But he also explains that Nvidia’s chips are built to solve many different problems. Positron’s approach is different: focus on a narrower set of AI workloads and optimize the chip architecture around those specific needs.

That specialization is what allows Positron to compete on cost, performance, and energy efficiency.

The conversation also explores one of the biggest barriers for any AI chip company: adoption. Customers do not want to rewrite their code, rebuild their workflows, or change the way they already run applications. According to Mitesh, Positron has focused on making its systems integrate into the existing Nvidia-based ecosystem so customers can run the same applications without changing lines of code.

That matters because better hardware alone is not enough. To win customers, Positron needs to show that companies can get more performance for less money and less power without adding operational complexity.

The episode then moves into the larger AI demand curve. Mitesh compares the adoption of AI to the growth of the internet, arguing that AI is scaling much faster. While the internet took years to reach hundreds of millions of users, AI tools reached that level in only a few years. In his view, this creates a massive long-term need for infrastructure.

Bruce and Mitesh also discuss the anxiety around AI hype, bubbles, and market air pockets. Mitesh acknowledges that technology cycles always have push and pull, but argues that the long-term demand for AI compute is still enormous. As AI expands into content generation, agents, robotics, scientific research, and enterprise workflows, the infrastructure layer will need to keep scaling.

Energy becomes another key part of the conversation. Data centers require massive amounts of power, and AI infrastructure is pushing electricity demand higher. Mitesh explains that the future of AI will depend not only on better chips, but also on more efficient systems and expanded energy production through sources like grid power, solar, geothermal, and nuclear.

The episode also touches on DeepSeek and what it revealed about the AI market. For Mitesh, DeepSeek showed that major efficiency gains are possible and that making AI cheaper can actually increase adoption. That connects directly to Positron’s thesis: if intelligence becomes cheaper to run, more people and companies will use it.

Finally, Mitesh shares his own founder journey, from chemical engineering and an early startup focused on water generation and purification to helping build Lambda, a company providing AI compute infrastructure. His path reflects a consistent theme: curiosity, technical depth, failure, revenue discipline, and the desire to build technology that creates real value.

Why this episode matters

This episode explains the part of AI that most people never see.

The public conversation often focuses on applications: chatbots, agents, image tools, content generation, and automation. But the future of AI also depends on the infrastructure underneath those applications. Without faster, cheaper, and more efficient chips, the next wave of AI adoption becomes harder to scale.

Mitesh Agarwal’s perspective is important because Positron is not trying to compete with AI hype at the surface level. It is targeting the physical and technical foundation of the market: inference chips, data center performance, energy efficiency, and deployment costs.

The episode is also relevant because it shows why the AI chip war is not simply about one dominant player. Nvidia may lead the market, but as AI workloads become more specialized, new companies can compete by building hardware optimized for specific use cases.

That is where custom silicon becomes disruptive.

For investors, founders, engineers, enterprise leaders, and anyone trying to understand where AI is going, this conversation offers a clearer view of the infrastructure race. The future of AI will not only be decided by who builds the best models. It will also be shaped by who can run those models efficiently, affordably, and at global scale.

Watch the full episode to understand how Positron is approaching the AI chip market, why inference may become one of the most important layers of AI infrastructure, and how cheaper intelligence could accelerate the next decade of adoption.