Arastu Sharma and the Bet on Physical AI: A Profile of Shodh AI’s Founder
Most AI founders in 2026 are building products on top of someone else’s model. Arastu Sharma chose a harder path. His company, Shodh AI, is building its own foundation model, and not for language. It is designed to reason about molecules, materials and manufacturing.
That choice puts Sharma and Shodh in one of the most ambitious and least crowded corners of the AI industry. It also makes the company a useful case study for any founder weighing whether to build on existing models or build something new.
The Path to Shodh
Sharma’s background combines academic research, corporate research and defense work. According to Shodh’s launch announcement, he founded the company after research at the University of Cambridge and Microsoft Research, followed by work with India’s DRDO defence AI ecosystem.
His speaker profile for AIM’s Cypher 2026 conference describes him as CEO of Shodh AI and of Aestr by GyanVihar, and a board member at Suresh Gyan Vihar University, and notes his association with the University of Cambridge as a Microsoft researcher since 2019. The same profile frames his work as sitting at the intersection of AI research, entrepreneurship and education.
That mix matters. Building a foundation model for science requires deep research credibility, access to talent and the ability to work with large institutions. Sharma’s background touches all three.
What Shodh Is Building
Shodh’s flagship is LUCAN, which the company calls a physical AI foundation model designed to reason across the physical world, from molecular behavior through materials and process conditions to industrial equipment and manufacturing.
The practical goal is to help industries move discoveries from the lab to the factory faster. In chemicals, pharmaceuticals and materials, a process that works in a small flask often behaves very differently at production scale. Engineers traditionally close that gap with repeated experiments and pilot plants. Shodh wants to replace part of that trial and error with computation.
The company reports early industrial results. In one specialty chemicals project, it says its modelling identified process changes linked to 17% more product from the same raw material. These are company figures from early deployments and have not been independently verified.
Government Backing
Shodh is one of the companies selected under the Government of India’s IndiaAI Mission to build foundation model capabilities for science. According to the company, that support funded the first generation of its model, and the government is now backing a larger scale up of training and scientific computing.
That puts Shodh inside a wider global trend. Governments around the world are investing in domestic AI capacity, and physical and scientific AI is one of the areas where a country can build something distinctive rather than compete head on with the largest American labs.
Why His Bet Is Different
Three things set Shodh’s approach apart from the typical AI startup.
It builds a model, not a wrapper
Many AI startups add features on top of a provider’s model. Shodh is building its own. That is slower and more expensive, but if it works, the result is far harder for competitors to copy.
It targets the physical world
Language models learned from the enormous amount of text on the internet. There is no equivalent public source of industrial process data. That scarcity makes physical AI harder to build, and also harder to replicate once someone has the data and the model.
It sells outcomes industries already pay for
Manufacturers spend heavily on process development and optimization. A tool that increases yield or shortens scale up is easy to explain in financial terms: more product, less waste, faster launches.
The Challenges Ahead
Sharma’s bet also faces real obstacles:
- Proof at scale. Early results need to be repeated across many customers and independently validated.
- Data access. Industrial companies guard process data closely. Building trust with them takes time.
- Long sales cycles. Manufacturing buyers are cautious, and adopting new process tools involves safety and quality approvals.
- Competition. As physical AI draws attention, larger technology companies will likely enter the space.
What Founders Can Learn From Shodh
Pick a hard problem with a clear payoff. Shodh’s target, making manufacturing scale up faster and cheaper, is difficult, but the value is easy to measure.
Use your background as a moat. Sharma’s research and institutional experience give Shodh credibility that a generic startup could not easily match.
Align with where public money is going. National AI programs can provide funding and compute to founders working on strategically important problems.
Be careful with early numbers. Striking results attract attention, but lasting credibility comes from independent validation.
The FounderFeat Take
Arastu Sharma is building in a part of AI that most founders avoid because it is slow, expensive and technically difficult. That is precisely why it could matter. If Shodh proves that a foundation model can meaningfully speed up how physical products are made, it will have built something far more defensible than most AI startups of this era.
FounderFeat will be following Shodh’s progress, especially as independent results emerge.