But there was a catch. Each scan, roughly 100 MB in size, had to be uploaded to the cloud, processed and the results sent back. This cycle was time consuming because of internet connectivity issues and power outages, common to rural areas. It resulted in a long line of patients waiting for results. The delayed scan reports made it harder to isolate a patient or start treatment in time.
This was a reminder of the mismatch between cutting-edge AI and the ground realities it was built for.
“If you’re building healthcare solutions for the Global South, you have to assume the testing conditions are far from ideal. The internet will be patchy. Electricity supply will be irregular. Machines need to have longer battery lives to last at least a day,” said Prashant Warier, co-founder of Qure.ai.
So, the team found a way to re-engineer their system. Instead of directing every scan through the cloud, they embedded the AI model directly onto local laptops attached to X-ray equipment, making the diagnostics system work even in offline mode. These laptops with ‘edge AI’ chips handled the heavy computing needed to analyse the scans on the spot, in under 20 seconds.
Today, Qure.ai partners with more than 15 medical device makers to embed AI directly into edge devices, making advanced diagnostics practical in places where stable internet connectivity is still a luxury. In 2025, the company announced it had crossed five million AI-enabled chest X-rays across 20 countries in Asia, Africa, the Middle East and Latin America.
The shift to ‘offline-first AI’ is gaining momentum in India and finding use cases beyond healthcare, across e-scooters, surveillance systems, smart meters and government services, where latency, power constraints or patchy networks make cloud AI unreliable.
Unlike cloud-based AI, which depends on remote servers to process data, edge AI has task-specific chips to enable real-time decisions—they need not contact distant data centres.
For instance, a surveillance camera in a remote area that needs to detect intruders instantly can’t afford to rely on uploading footage to a distant data centre in order to take quick action. It must process that data locally to trigger real-time alerts. Or in railway safety systems, edge-enabled cameras mounted at crossings or on train engines can instantly detect obstructions on the tracks like stalled vehicles or pedestrians and trigger alerts without waiting for cloud-based processing to take action.
To be clear, edge AI is not about replacing the cloud entirely. It’s about shifting critical intelligence closer to the problem.
Made in India
At the heart of this movement is a new breed of Indian chip designers.
A handful of Indian founders, many of whom have spent years in global semiconductor firms or top research labs, are now building fabless chip startups—where design is done domestically and fabrication outsourced to foundries abroad—from the ground up. The fabless model is the global industry norm, followed even by companies like Apple, which designs its chips in the US but manufactures them abroad. The model allows startups to focus on innovation without the heavy capital costs of running fabrication plants.
BigEndian Semiconductors, founded by IIT alumnus Sunil Kumar, is now building vision-processing chips for surveillance, automotive and industrial use. Shashwath T.R. and Sharan Jagathrakshakan, co-founders of Mindgrove Technologies, and both IIT Madras alumnus, are designing ultra-low-power system-on-chips (SoCs), intended for use in connected appliances, smart meters and other cost-sensitive devices. Netrasemi, founded by former Intel engineers, is developing a homegrown 64-bit RISC-V processor, targeting use cases in consumer electronics and embedded AI.
Together, these initiatives represent a growing pool of vastly experienced engineers betting that India holds a chance in the AI hardware game, through the edge AI route. This is where India’s semiconductor journey might finally find a foothold.
Why it matters
The shift toward edge chips, while global superpowers engage in AI wars, could be a strategic realignment in India’s hardware journey.
While cloud chips, such as Nvidia’s A100s and H100s, are used to train and run massive AI models such as ChatGPT or Gemini, they are expensive and highly dependent on infrastructure. India imports nearly 100% of these high-end chips and despite recent government GPU leasing initiatives, it remains far behind in building foundational AI infrastructure.
Edge chips, by contrast, are narrower in function but more suited to India’s immediate needs. They don’t require massive data centres and can be deployed widely in the field. And while the global chip market is consolidated among giants such as Qualcomm and NVIDIA, the edge AI space leaves room for new entrants solving highly localized problems.
“India is fundamentally a low-bandwidth, power-constrained, cost-sensitive market, so the typical cloud-first AI stack was never going to scale here,” said Vishesh Rajaram, managing partner at the deeptech-focused, early-stage fund Speciale Invest. “India doesn’t just need edge AI,” he added. “We are one of the best sandboxes to build it for the world.”
Last year, Speciale Invest co-led an $8 million funding round in Mindgrove Technologies, alongside Rocketship.vc.
Pranay Kotasthane, deputy director at The Takshashila Institution, a think tank, noted that India’s strength lies not in controlling raw chip supply chains but in its globally embedded talent pool. By harnessing Indian design expertise across the AI stack, particularly in edge applications, India can create dependencies that are geopolitically difficult to ignore.
The Edge AI game
We don’t have the economic muscle to build hyperscale GPU infrastructure like the US or China. So, we started where India has a natural advantage: AI at the edge,” said BigEndian Semiconductors’ Kumar. He had previously worked at Broadcom, Intel and ARM and also co-founded the SaaS unicorn Zenoti.
Instead of importing and assembling parts like many others in the ecosystem, Kumar and his team chose to go deeper by designing their own chips tailored to India’s most pressing and high-volume use cases in public infrastructure, including surveillance systems, and defence and automotive applications, where real-time video or image processing is essential.
India’s video surveillance market is projected to reach $7.12 billion by 2030.
View Full Image
“The chips powering surveillance cameras in India are almost entirely imported from a single Chinese supplier today. We saw an opportunity to not only replace them but make them smarter, secure and India-specific,” he said.
BigEndian’s camera-focused chips are currently being tested by original equipment manufacturers (OEMs) such as Honeywell for use in consumer security systems, traffic surveillance and industrial monitoring.
While the chips are manufactured at a fabrication facility in Taiwan, the architecture and design are developed in-house by BigEndian in India. Unlike chips that are exported or rebranded, BigEndian’s semiconductors are shipped back to India as finished and tangible components for local deployment. Kumar noted that the designs are specifically built for India’s conditions to account for power fluctuations, limited connectivity and security flaws commonly found in foreign firmware.
“These aren’t generic processors. They’re tightly integrated stacks of security firmware, AI engines, and signal processors shrink-wrapped into a 5mm die,” he said.
Mindgrove’s local chips
A similar principle guides Shashwath T.R., founder of Mindgrove Technologies, which is building India’s first commercial-grade high-performance microcontroller (MCU).
“India has millions of smart meters, washing machines and mobility devices. These aren’t ‘smart’ unless they can run AI locally. That’s what we’re solving for,” said Shashwath.
View Full Image
Mindgrove’s chips can be locally programmed, audited and secured—an important consideration in the environment of geopolitical chip vulnerabilities. The company’s first product, a general-purpose microcontroller designed entirely in-house in Chennai, is now moving into volume production after completing its prototyping phases last year.
Shashwath says Mindgrove has already signed on a few early OEM partners and the chip is expected to show up in biometric authentication devices, smart meters, robotics and even thermal printers over the next year.
These chips aren’t just for India, though. They’re designed around both Indian and international specifications, allowing them to serve similar markets with shared conditions like patchy power, limited compute and regulation-heavy device sectors.
In an era of rising concern over data privacy and digital sovereignty, the ability to process sensitive data locally from a smart meter, biometric scanner or factory sensor, can reduce technical and regulatory risks. That’s a feature India’s public infrastructure increasingly wants.
Challenges within and outside
Despite the rise of these startups, India’s hardware realities are rooted in systemic challenges. Hardware startups are up against a hostile environment, plagued by challenges such as low investor interest, long timelines and supply chain gaps.
Last September, BigEndian Semiconductors raised $3 million in a funding round led by Vertex Ventures SEA & India.
“We first mapped the entire value chain,” Kumar said. “Who builds what; where the components are sourced; what local OEMs want. Only then did we start designing.”
Kumar explains that unlike software, the hardware industry leaves little room for trial and error. Mistakes or pivots cost much more and development is very capital-intensive. This makes it a daunting space for both founders and investors, unless there’s a clear path to revenue. As a result, many startups begin with commercially viable applications such as industrial surveillance before venturing into more complex and long-cycle domains such as defence.
“Anyone trying to actually do real building is breaking their heads against an ecosystem that doesn’t exist,” said Shashwath. Indian edge chip makers can make a real difference if they sell in the Global South, where they are currently up against larger players.
But unlike their global peers, in Taiwan or China, where tight integration between chipmakers, foundries and design houses allows new iterations in as little as two weeks, Indian teams can take up to six months to ship the same update.
The specialized software and licenses required to design, test and verify chips are also prohibitively expensive for Indian startups. Startups must carefully ration tool usage among engineers (because they can afford only a limited number of licenses), even with support from the Design Linked Incentive (DLI) Scheme, an initiative by the ministry of electronics and information technology. The scheme offers financial and infrastructural aid to encourage domestic semiconductor design.
“We’re always rationing tools,” Shashwath said, “and it shows in how long we take to ship.”
We’re always rationing tools, and it shows in how long we take to ship.
—Shashwath T.R.
Both Kumar and Shashwath believe that while schemes like DLI are helpful, they are not enough. What India’s semiconductor ecosystem really needs is patient capital, easier access to tools and IP, more domestic buyers and better alignment between R&D and manufacturing.
The local electronics market in India is only beginning to warm up to local chip makers. Despite a strong base of electronics manufacturing in India, most OEMs currently rely on global chip suppliers.
“Many Indian OEMs still hesitate to switch from global incumbents unless there’s a proven delta on cost, performance or availability,” said Speciale Invest’s Rajaram. “Startups that offer a drop-in alternative with local support, tighter integration and a roadmap win faster.”
Kotasthane of The Takshashila Institution argues that India should double down on fabless incentives, niche areas like compound semiconductors and even fund open-source AI accelerator projects under the National AI Mission.
The upside is real. If India gets edge AI right, it can build a sovereign technology stack for itself, and for the world’s underserved regions.