NEW DELHI: Pratik Parmar 指出他的初創公司 Atomo 本月於 Semicon India 2026 會議上展示的一塊電路板。它讓相機和機械能偵測火災、追蹤人數，或於無需持續把資料送往雲端的情況下識別工廠缺陷。
「你不需要互聯網或雲端。發生在你場域內的事情，就留在你場域內，」Parmar 如是說。
所謂的邊緣人工智能（AI）晶片，能在相機、車輛和其他電子裝置中本地運行 AI，並可按特定裝置與任務定制。
Atomo 的電路板目前依賴美國設計的 Amlogic 晶片，因為三年前當他們選擇時，印度本地的替代方案尚未成熟。
但情況正在改變。一個以這些專門的 AI 晶片為核心的新興印度產業已經出現，本土公司現正把首批產品推向客戶測試與商業化。
根據印度電子部門的資料，在政府的晶片設計計畫之下，已有 105 家新創和小型企業獲得使用先進設計工具的機會，並已產出約 35 種可製造的晶片設計。
儘管印度在設計 AI 資料中心晶片的競賽中起步較晚，業界高層告訴 CNA，各方認為該國仍具競爭空間。
“India does not have to build the world’s fastest AI processor. Putting intelligence into cameras, vehicles and everyday electronics could be a more realistic opening,” said Ashok Chandak, president of industry bodies SEMI India and the India Electronics and Semiconductor Association.
“But software, scale and customer adoption remain the real tests.”
INDIA EYES AN AI CHIP OPENING
印度的半導體市場預計由 2026 年的 640 億美元增長至 2035 年的 2000 億美元，其中消費電子、汽車與工業應用將佔需求的 61%。
專家表示，AI 系統日漸需要在本地處理數據以提供更快的回應，而不是依賴遙遠的資料中心，。
例如一輛車需要對前方障礙立即作出反應，而工廠相機可能需要實時發現缺陷。
InCore Semiconductors 的 CEO G S Madhusudan，一家為晶片設計師開發處理器技術的公司，表示這種本地智慧正從汽車、相機延伸至電視與工業設備等產品。
Such chips can continue working when internet connectivity fails, keep sensitive data on-site and reduce cloud costs, Atomo’s Parmar said.
For a factory safety camera, unreliable internet could delay critical alerts, while processing images locally allows it to keep working regardless of connectivity, he said.
印度的成本敏感市場及不穩定的網路連線，讓這些好處尤為相關，他補充道。
Atomo has also sent circuit boards for testing in markets including Japan, South Korea and the Middle East, where Parmar said response time, confidentiality and uninterrupted operation are equally important.
Globally, consumer devices accounted for 65.3 per cent of the edge AI chip market in 2025, driven by growing use in smartphones, laptops and smart speakers, according to a report by market research firm Grand View Research.
India’s market is projected to grow nearly 12 times, from US$984.1 million in 2026 to US$11.7 billion by 2033, the report said.
The size of that opportunity is one reason Indian chip designers are targeting the market, industry experts said.
INDIA AS A TESTING GROUND FOR EDGE AI
Sunil Kumar, whose startup BigEndian Semiconductors is developing an AI-enabled processor for surveillance cameras, said specialised chips can be tailored to individual markets.
A railway camera, industrial camera or vehicle, for example, can require different hardware and software, giving smaller companies more room to customise rather than compete primarily on raw computing power, Kumar said.
India also has a large chip-design workforce, with about 20 per cent of the world’s semiconductor design engineers based in the country, according to Invest India CEO and former Intel India head Nivruti Rai.
Kumar said India’s scale and complexity could make it a testing ground for such technology, pointing to crowd management at the Kumbh Mela, a Hindu pilgrimage that drew more than 660 million people over 45 days in 2025.
“If I’m able to adapt to some place like … Kumbh Mela, solving the same crowd control issue in Finland is much easier,” he said.
Investors are putting more money into the sector. Indian edge AI chip startups raised US$17 million in 2026, nearly four times the US$4.27 million raised in 2022, and up from virtually nothing before 2020, according to data insights firm Tracxn.
The bigger challenge, however, is to persuade a company already using established foreign processors to switch to Indian chips, entrepreneurs building in the space told CNA.
SECURITY CONCERNS GIVE LOCAL CHIPMAKERS A CHANCE
Shashwath Ramkumar, co-founder of chip startup Mindgrove Technologies, said his company looks for moments when customers “want to change anyway” because regulations, technology or product requirements have shifted.
Mindgrove is currently working with Indian surveillance manufacturer Prama to develop its vision chip for security cameras.
India’s new cybersecurity rules for internet-connected CCTV systems, which took effect on Apr 1, have created opportunities for local chip firms as tighter rules affect some suppliers, experts said.
While India has not banned Chinese CCTV cameras outright, tighter security and sourcing rules have restricted the import of some surveillance equipment from the neighbouring country, industry experts added.
Chinese vendors accounted for about a third of India’s CCTV market in 2024, according to market research firm Counterpoint Research.
Similar pressures are appearing in other industries.
Tracxn co-founder Neha Singh said geopolitical issues along with efforts to diversify supply chains are making some Indian automotive, industrial and defence manufacturers more willing to evaluate locally designed chips.
FROM SAMPLES TO SCALE
Despite the market opportunities, turning them into actual orders remains a challenge for Indian chipmakers, experts told CNA.
The founders of Indian edge AI chip firms expect the coming year to be crucial.
Ramkumar said Mindgrove’s work with Prama could take roughly another year to produce a commercial product.
BigEndian is preparing to put its first AI-camera chip into customers’ hands by the end of this year.
Kumar said manufacturers may discuss their needs with chip designers, but are unlikely to place orders until they have a physical chip that has been tested in their products.
Kumar described 2026 as the “year of sampling”, with a meaningful scale potentially emerging in the second half of 2027.
Ramkumar said: “The packaging, testing and logistics are harder than design.”
The challenge is building a supply chain that can reliably deliver chips at scale, he added.
Ramkumar said India’s chip-packaging capacity is beginning to expand, easing one part of the supply-chain challenge.
For instance, CG Semi, a joint venture between India’s CG Power, Japan’s Renesas Electronics and Thailand’s Stars Microelectronics, began commercial production at its chip-packaging and testing facility in Sanand, Gujarat, in July.
The facility can package and test up to 300 million chips a year.
Under the government’s semiconductor mission, advanced chip packaging is a priority, with financial support offered for new assembly, packaging and testing facilities.
“The real test is whether Indian startups can make their chips easy for manufacturers to adopt, reliable at high volumes and competitive beyond India – turning customer trials into repeat orders at home and eventually winning large global customers,” said Tracxn’s Singh.
