Bharat Semi Raises $24 Million: How a Delhi Chip Startup Is Putting AI Inside the Sensor
Posted on 16th Nov 2026 11:03:06 in Artificial Intelligence, Machine Learning
Tagged as: Bharat Semi, AI sensors, Indian semiconductor startup, physical AI, edge AI, chip funding
Bharat Semi, a New Delhi-based semiconductor startup, has raised $24 million in seed funding to build programmable image sensors that carry AI processing on the same piece of silicon that captures the picture. The round was led by Info Edge Group, South Park Commons and Indus Bridge Ventures, with Micromax, Valour Capital, New York-based Gaingels and angel investors Mukul Agarwal and Karthik Sundar Iyer participating.
The deal is unusual for two reasons. First, the company is nearly five years old and still pre-revenue, with its sensors under development and initial commercial production targeted for 2027. Second, it is a rare Indian attempt to build a globally sold semiconductor product rather than a services business or a back-office design centre for a foreign chipmaker.
"Our aim here is to move from sensors to sense-making solutions," co-founder Vrinda Kapoor told TechCrunch. In a separate interview with Mint, she framed the larger ambition plainly: "It's a story of can we build a global technology company out of India."
What Bharat Semi Actually Builds
The product is not a general-purpose processor. It is an image sensor with the image signal processor and a small AI inference engine folded into the same die. In a conventional camera system, three separate chips do three separate jobs: one captures light, one cleans up the image, and one runs the AI model. Data has to move between them, and each hop costs power, latency and board space.
Bharat Semi's approach removes those hops. The sensor captures the frame, processes it and runs inference locally, so a device can decide what it is looking at without shipping raw pixels to a separate processor or the cloud. The company says this architecture can cut system costs by at least 70 percent compared with conventional image sensor, image signal processor and system-on-chip combinations, and reduce data movement between components by up to 99 percent.
Because the sensors are programmable, the same silicon can be reconfigured in software for different applications and different AI models. That matters commercially: a surveillance camera, an air conditioner and a drone have very different power budgets and thermal limits, but they can share one hardware platform instead of three custom designs.
The target markets are broad. Bharat Semi is starting with consumer electronics and surveillance, where cost, power and latency decide who wins, and intends to extend into industrial automation, robotics, drones, automotive and aerospace. Kapoor's example is deliberately mundane: an air conditioner that senses an empty room and adjusts itself, with no expensive companion compute anywhere in the system.
The Money: $24 Million, Three Leads, and a Valuation Reset
The all-equity round closed at $24 million, roughly Rs 215 crore. Kapoor told Mint that the company had initially targeted $10 million to $15 million and raised the target after investor interest came in stronger than expected. The money is expected to fund about two and a half years of operations, with the stated goal of becoming revenue-positive with roughly six products deployed across major customers within that window.
The investor mix says something about how the deal was priced. Info Edge participated through Redstart, its deeptech arm, and its founding partner Vibhore Sharma framed the thesis in hardware terms: "Bharat Semi is tackling a significant limitation in Physical AI at its origin, within the silicon itself. Its ambitions are global, and its technical approach is uniquely its own." Micromax's participation is strategic rather than financial, and Kapoor acknowledged as much, saying it offers a hint of where the company's consumer electronics push could go.
For context on the step-up: Bharat Semi raised $4.63 million in July 2025 in a round led by RedStart Labs and angel investor Ajay Upadhyaya. Tracxn data cited by Mint put the company's post-money valuation at $149 million as of January 2026. That is a rich multiple for a pre-revenue chip startup, and it signals that investors are pricing the category, not the current balance sheet.
Design Wins Before the Product Exists
The most consequential detail in the announcement is not the funding figure but the customer list. Bharat Semi says it has secured design wins with General Atomics, the US defence technology company, and with Kaynes Semicon, the Indian arm of the Kaynes group's semiconductor ambitions. Design wins at this stage are commitments to evaluate and integrate, not purchase orders, and revenue only arrives when production parts ship. Still, they are the difference between a lab project and a supply chain relationship.
The roadmap is staged. First-generation sensors are in prototyping now. Second-generation parts, which add more advanced on-chip AI capability, are expected to reach prototype stage by mid-2027, with customer units targeted by the end of that year. Initial commercial production is planned for 2027.
The team behind that roadmap is small and deliberately senior. The company has about 75 employees across India and the US, including 10 to 12 in Santa Clara doing engineering and business development. It plans to grow the US headcount to around 20 by mid-2027 and has begun hiring engineers in Europe. Co-founder Vinayak Dalmia, who serves as managing director, came from strategy work at McKinsey and private equity, and holds an MPhil from Cambridge. Chief technology officer Boyd Fowler previously held the same role at image-sensor maker OmniVision, which is the kind of résumé that matters when you are asking camera module makers to bet on an unproven supplier.
Why Physical AI Needs a Different Sensor Stack
The category term being used here is Physical AI, which is the industry's shorthand for AI that perceives and acts in the real world rather than generating text on a screen. The economics of that shift are what make a sensor company interesting again after two decades of consolidation.
Sensors today were designed to produce data for humans to look at. A camera module's job was to hand over a faithful image. But when the consumer of the image is a machine learning model, fidelity to human perception stops being the only objective. What matters is whether the sensor delivers clean, relevant, low-latency data to a model that must decide something in milliseconds, at a power budget that a battery-powered drone or a wall-mounted camera can actually afford.
"In the near future, growth in the sensor market will be driven by capturing data for machines," Dalmia said. "This data may never be seen by humans, but it will drive AI at the edge. This requires a different type of system, where sensing, processing and software are designed together."
The demand-side case rests on forecasts that are large and uncertain in equal measure. Business Standard cited Morgan Stanley's projection of around a billion humanoid robots in use by 2050, and a McKinsey finding that no widely adopted intelligent sensor suite exists today. Treat the billion-robot figure as a directional bet rather than a plan. The second finding is the one that matters for Bharat Semi: if the intelligent sensor layer genuinely does not exist yet, the market is not being contested by an incumbent with a decade of head start, even though Sony, OmniVision, onsemi and Samsung all sell excellent conventional image sensors.
India's Semiconductor Moment, and the Sobering Part
The funding lands in a policy environment that has changed dramatically. India's Semicon 2.0 programme, approved in July 2026, carries an outlay of Rs 1,27,500 crore and covers chip design, manufacturing equipment, materials, fabrication, packaging, research and talent development. In November 2026 the government reported that five of the 12 approved semiconductor projects had begun commercial production.
That is real progress, but the gap is still wide. India has committed billions to become an alternative manufacturing hub as US companies diversify supply chains away from China, and it has not yet established itself as a major chip manufacturing centre. Bharat Semi's own approach reflects that reality: it intends to remain fabless, outsourcing manufacturing and concentrating on design and software. A separate fabrication project the company announced in 2024, involving a technology partnership with the US Space Force, remains in development and is still looking for domestic partners to build and operate the facility. The new $24 million will not go toward it.
The broader deeptech funding base is also thin but warming. Bengaluru chipmaker C2i Semiconductors raised $15 million in February 2026, sensor and navigation startup Yaanendriya raised Rs 15 crore in July, and construction robotics company Flo Mobility raised $2.5 million in May. Those are small cheques by global semiconductor standards, where a single tape-out at an advanced node can cost tens of millions. A design company that stays fabless and targets mature nodes can operate at Indian scale; a company that wants to own the leading edge cannot.
What This Means for Indian Businesses Building With AI
For most Indian SMEs, the near-term relevance is not the chip itself but the cost curve it implies. Machine vision has been expensive because every camera deployment needs a sensor, a processor and the engineering to connect them. When sensing and inference collapse into one component, the bill of materials for a smart camera, a quality-inspection station or an occupancy-aware HVAC system falls. Kapoor's 70 percent system cost claim, if it holds in volume production, is the kind of change that moves computer vision from pilot projects to standard fitouts.
There is a practical implication for software teams as well. Edge inference produces structured events rather than video streams: a count, a classification, an alert. Applications built around those events need far less bandwidth and storage than applications built around video, and they avoid sending raw footage across a network, which simplifies data protection compliance. Businesses currently paying for cloud inference on continuous camera feeds should expect device-side processing to become the cheaper default for high-volume, low-complexity visual tasks.
The timeline deserves emphasis. Production is targeted for 2027, second-generation parts reach customers at the end of 2027, and the company is pre-revenue today. Any procurement plan built on this specific supplier should be treated as a 2028 conversation. What has changed this week is that the capital exists to finish the design, not that the silicon is available.
The Risks Worth Watching
- Execution on an unproven process. Combining a pixel array, an image signal processor and an inference engine on one die is a yield and thermal challenge as much as a design challenge. Sensors that work in a lab often struggle when they must run continuously for years in a sealed camera housing.
- Incumbent response. Sony, OmniVision, onsemi and Samsung already ship stacked sensors with on-chip processing. They can add inference acceleration to a mature, high-volume product line faster than a startup can qualify a new one.
- Design wins are not revenue. General Atomics and Kaynes Semicon are credible names, but qualification cycles in defence and industrial markets are long, and a design win that does not convert to volume leaves the company dependent on the next round.
- Fabless dependency. Outsourcing manufacturing keeps capital requirements down, but it also means foundry allocation, pricing and process availability are outside the company's control, and small customers sit at the back of the queue when capacity tightens.
- Valuation pressure. A $149 million post-money mark on a pre-revenue company sets a high bar for the next round. The company has roughly two and a half years of runway to reach revenue-positive status with six products deployed, and that is an aggressive schedule for hardware.
The honest read is that Bharat Semi has raised enough money to find out whether its thesis works, not enough to guarantee it does. What is genuinely new is the shape of the bet: an Indian company designing a differentiated chip for a global market, backed by Indian institutional capital and a strategic investor with a reason to care about the consumer device end. Whether or not the sensors ship on schedule, that template is worth tracking.
Sources
- TechCrunch - Indian chip startup Bharat Semi lands $24M seed round as it eyes global AI sensor market
- Mint - Bharat Semi raises $24 million in seed round led by Info Edge, South Park Commons
- Business Standard - Bharat Semi raises $24 million to develop intelligent sensors for AI
- YourStory - Bharat Semi raises $24M to develop AI-enabled sensors
- Entrackr - Bharat Semi raises $24 Mn in seed round from Info Edge, South Park Common, others
- ET EnterpriseAI - Bharat Semi raises $24 million in seed funding to scale AI sensor technology
- Indian Startup News - Bharat Semi raises $24 million to build AI sensors for robots, cameras and drones