Semiconductor Chips Have An Elemental Problem And

What happened

India’s chip build-out continues in full swing. 6 Lakh Cr to meet the rising chip demand. 0) has expanded the government’s focus to chip equipment, design and domestic intellectual property (IP). Taken together, the two initiatives reflect a wider shift in India’s semiconductor strategy: building more fabs is important, but having indigenous technologies will determine whether these facilities can produce advanced chips at scale.

This shift is creating a fresh need for materials that can enable the next generation of semiconductor devices. Among the most pressing challenges for these new materials is heat. As GPUs become increasingly powerful, they consume more electricity and generate more heat. Each layer within the chip adds resistance to the flow of heat, making thermal management increasingly difficult. For Discovered Materials, this makes semiconductors a particularly compelling starting point.

The founders believe the industry is entering a period where new materials will be needed not just to make chips smaller, but to make increasingly powerful chips physically viable. With traditional materials discovery taking years of research and experimentation, the startup is betting that AI can help accelerate that search.

At the heart of Discovered Materials’ proposition is an AI-driven system designed to compress the long and repetitive process of materials discovery. Its AI agents can propose new material structures, assess their properties and refine the candidates based on what they learn. The agents have access to a coding sandbox, web search and a database of existing materials.

They can create candidates through code, draw on published research and compare proposed materials with known structures and performance benchmarks. The startup has also built verification tools to determine whether candidates meet specific requirements. These tools can estimate properties such as thermal conductivity and static dielectric constant, among other parameters.

Some of these research runs can consume around 100 Mn tokens as the system repeatedly searches, simulates and verifies potential materials, said Ramdas. This approach allows Discovered Materials to screen far more candidates than a researcher working manually. Ramdas estimates that an individual researcher can generate around 20 candidate structures a day, while the startup’s system can screen more than 2,000 candidates daily.

Even a successful laboratory synthesis does not mean commercial viability. The startup also needs to determine whether it can be produced at a fab unit where materials need to meet stringent manufacturing requirements.

Why it matters

But building fabs and increasing chip production addresses only one part of the challenge. The materials that go into a chip, and their ability to manage heat, power and other physical constraints, could increasingly determine how far semiconductor performance can be pushed. This is the gap that Discovered Materials is looking to address.

Founded in 2026 by materials scientist Akash Ramdas and AI researcher Advaith Sridhar, the startup is using AI agents to accelerate the discovery of new materials for semiconductor applications. As the semiconductor industry pushes against the physical limits of existing materials, the startup is betting that AI can help researchers identify and test a far larger pool of potential alternatives.

The startup recently raised $9 Mn in its maiden funding round from marquee names such as Lightspeed India, Y Combinator and Peak XV Partners. With plans to now scale its experiments and compute, how did the two founders go from being batchmates at IIT Madras to solving one of the biggest bottlenecks that will define the next generation of computing?

For Ramdas, the problem of finding better semiconductor materials is not a new one. A graduate of IIT Madras, he went on to pursue his master’s and PhD at Stanford, where his research focused on the computational discovery of materials for nanoelectronic applications. His cofounder Sridhar came from the world of AI.

After graduating in electrical engineering at IIT Madras, Sridhar pursued a master’s degree in AI from Carnegie Mellon University and went on to work at a Bay Area startup, Persona AI, before joining Luma Labs. The two founders have known each other since their graduation days, and their conversations eventually led them to explore whether the advances in AI could be applied to the complex process of materials discovery.

This is especially critical as much of the semiconductor industry’s progress has been driven by one central idea for decades: make transistors smaller. But this approach is increasingly running into the limits of physics. As components shrink towards the scale of individual atoms, packing more transistors into a smaller footprint is becoming harder. This is pushing the industry to explore new architectures and ways of improving chip performance.

What to watch

However, producing promising candidates is only the first step. The harder question is whether those materials can actually be made. Current AI models remain weak at identifying practical synthesis routes for many of the materials they propose. A candidate that looks attractive in a simulation may be uneconomical, unstable or too difficult to produce in a laboratory. Ramdas described this as the startup’s biggest technical limitation.