Veerbhan
Kheterpal

A robot in a vineyard, a gap in the silicon

Quadric did not begin as a chip company. It began as a robot meant to roll through vineyards and manage the vines. The machine worked. The problem was underneath it. Veerbhan Kheterpal and his co-founders found that the CPUs and GPUs available to them could not deliver the compute the robot needed inside the power budget it had to live within. The constraint was not a nuisance. It was the whole story.

That realization reframed the company. Instead of building a better robot, Kheterpal and his team went after the thing the robot was missing: a processor architecture designed from dataflow principles, built to run heavy AI workloads without the power draw of a data-center GPU or the rigidity of a fixed-function accelerator. The pivot turned a promising robotics project into a semiconductor IP business.

It was a familiar move for Kheterpal. By the time he started Quadric in 2016 with Nigel Drego and Daniel Firu, he had already founded two companies and carried, as he puts it, "full stack expertise spanning software to silicon." He is an engineer who keeps arriving at the same conclusion from different directions - that the interesting problems live at the seam where code meets hardware.

That sentence sounds almost heretical coming from a chip founder. Semiconductor companies are supposed to obsess over transistors, tape-outs, and process nodes. Kheterpal flipped the priority. His argument is practical: a processor is only as useful as the tools that let developers put it to work. If the software toolchain is an afterthought, the fastest chip in the world sits idle. So Quadric wrote the compiler and the developer experience first, and shaped the hardware to serve it.

Programmable, so it does not go obsolete

The central idea behind Quadric's Chimera processor is a general-purpose neural processor, or GPNPU. It fuses digital signal processing and neural acceleration into a single programmable core, collapsing what used to be three separate blocks - a CPU, a DSP, and a bolted-on NPU - into one. The pitch to chip designers is that a single programmable engine can run any AI model, including models that do not exist yet.

That last part is the whole point. AI models change fast. A fixed-function accelerator tuned for last year's network can be stranded by this year's architecture. Kheterpal frames the alternative bluntly.

His read on the market is that existing options force a bad choice. "Existing solutions are either too power hungry, think GPGPUs, or too restrictive in capability, think AI chips and accelerators," he has said. Quadric's answer is a middle path: programmable enough to survive change, efficient enough to live inside a phone, a car, or an industrial sensor. The Chimera line scales from a single TOPS up to 864, with configurations tuned for automotive safety, and can hold large language models with as many as 30 billion parameters on the device itself.

Three companies, one conviction

Kheterpal's path runs from Kharagpur to Pittsburgh to the San Francisco Bay Area. He earned a B.Tech in electronics and communication engineering at IIT Kharagpur, then a PhD at Carnegie Mellon University. His first venture, Fabbrix, built software that made complex integrated circuits easier to manufacture. It was acquired by PDF Solutions.

Next came 21, Inc., where he served as a technical co-founder focused on power-efficient ASICs for cryptocurrency - a bet on custom silicon squeezing the most work out of every watt. The through-line to Quadric is clear: efficiency, custom architecture, and software that makes the hardware usable.

When he advises other semiconductor founders, Kheterpal returns to a hard question about value and staying power: "Am I creating enough short term value to be able to build my company while having a path to becoming a sizable enterprise?" It is the kind of question asked by someone who has learned that a clever architecture is not a moat on its own. Commoditization, fast-following competitors, and the need for durable advantage are risks he names openly.

The market catches up

For years, on-device AI was a smaller story than the cloud. That is changing. As companies look to cut the cost and latency of running inference in remote data centers, intelligence is moving to the edge - onto phones, cars, PCs, wearables, and robots. Quadric was built for exactly that shift, and the numbers show a market arriving at Kheterpal's door.

In early 2026, Quadric announced a $30 million Series C led by BEENEXT Capital Management, then extended it to $46 million with a second close led by the International Finance Corporation, the private-sector arm of the World Bank. That took total capital raised to roughly $90 million. Licensing revenue, by the CEO's account, grew from about $4 million in 2024 to $15-20 million in 2025, with a target of up to $35 million as the business shifts toward royalties. The company's licensees span automotive, edge LLMs, office automation, and autonomous driving, with incoming interest from humanoid robotics, wearables, and networking.

The IFC's involvement is worth pausing on. A development-finance institution backing an edge-AI chip company signals that on-device inference is not just a Silicon Valley enthusiasm but a piece of infrastructure with reach into emerging markets. Quadric claims it can take a chip designer from first engagement to production-ready silicon in under six months, thanks to a software toolchain built from the ground up rather than bolted on.