Image credit: Cogknit — official company visual. A listed defence major is paying up to ₹135.15 crore for a controlling stake in a company whose last reported annual turnover was ₹48 lakh. The company is not a fraud, and the buyer is not reckless: Zen Technologies’ board, which approved the deal on 17 October 2026, is buying technology it would take years to build in-house. The target is Cogknit Semantics, a Bengaluru AI lab founded in 2009 that has spent seventeen years teaching machines to listen — including to Mandarin, at 15,000 feet, with no internet.
The deal: a controlling stake for technology, not revenue. Under the board-approved terms, Zen will acquire a 51.98% controlling stake in Cogknit on a fully diluted basis for an aggregate cash consideration of up to ₹135.15 crore (about $14.2 million), executed through a secondary share purchase and subscription to Series A compulsorily convertible cumulative preference shares priced at ₹27,015.71 per share. The agreement gives Zen the right to nominate a majority of directors, and completion is expected by 13 December 2026. It needs no government or regulatory approvals and is not a related-party transaction. Zen’s filings describe Cogknit as an indigenous developer of AI-based battlefield management and situational-awareness systems for the armed forces — a portfolio extension for a company known for defence training simulators and counter-drone systems. The numbers tell you what kind of deal this is: Cogknit’s turnover was ₹2.24 crore in FY24, ₹0.78 crore in FY25 and just ₹0.48 crore in FY26. Nobody is buying revenue. They are buying capability.
The device: a 600-gram AI translator for the LAC. Cogknit’s best-documented defence product is a handheld Mandarin translation device built for the Indian Army. Weighing about 600 grams, it is an offline, bidirectional system that converts spoken Mandarin to English — a Hindi version was in development — with no internet connection, a deliberate design choice for forward areas where connectivity cannot be assumed. It was one of the 75 AI-enabled products and applications unveiled by Defence Minister Rajnath Singh at the ‘AI in Defence’ symposium in July 2022, where co-founder and managing director Anuroop Iyengar told The Hindu it had been tested in forward areas with Army assistance and that its performance improves as more field data comes in. The problem it solves is real and well-documented: at patrol-level encounters and flag meetings along the Line of Actual Control, the language barrier forces both sides to depend on interpreters. An earlier FICCI compendium presented to Army chief General Bipin Rawat at DefExpo 2018 described Cogknit’s solution — then proposed as a headset-and-tablet setup — as a “100% indigenous solution” deployable for trial at a week’s notice, per the company’s director for defence solutions, Deepak Kumar. The first demonstration was in 2017, followed by several Army trials, with support from the Kalyani Group, and the company was working on shrinking the device to wrist-wearable size and extending its range to 20 feet.
The platform: BLADe and the move to frontline intelligence. The translation work sits under a product family the company calls BLADe. According to Cogknit’s own company timeline, the language-translation solution won a $1-million deal in 2019, commercially launched in 2020 — and in the same year, the company began developing what it terms the “BLADe ISR platform”, ISR being intelligence, surveillance and reconnaissance. The branding on its official site today reads “BLADe — ISR Platform for frontline Intelligence”. Public detail on the platform itself is thin, which is normal for defence systems sold to the armed forces, but the trajectory is clear: from translating what soldiers hear, to the wider sensing, fusion and awareness problem — the battlefield management and situational-awareness systems that Zen’s acquisition filings name as the strategic prize.
The stack underneath: seventeen years of speech AI. Cogknit began, by its own account, as a garage company in 2009, formally incorporated on 25 September 2009. Its core is a speech and language stack: end-to-end and hybrid automatic speech recognition models, plus text and vision agents, developed over what the company describes as more than a decade of research. The speech product, branded AutoVox, processes Indian English, Hindi, Kannada, Tamil and more than eleven Indian regional and border languages — a catalogue that matters because India’s languages include some of the world’s most speech-data-poor ones, and because “border languages” is exactly the vocabulary of the Mandarin translator. The company signed an MoU with Tata Power in 2016 to build automatic speech recognition for Indian languages, works with academic partners including IISc, IIT Bombay, IIIT Bangalore, IIT Hyderabad and IIT Patna, and is an NVIDIA Inception partner. Before defence, it built ICOG, an AI-driven corporate learning assistant commercially launched with L&T Engineering in 2018 — which explains why the lab has always been a product company with a research lab’s CV, rather than the reverse.
The company: an old hand that never went mainstream. Cogknit was founded by Anuroop Iyengar (co-founder and managing director) along with Ullas Sathyanarayana, Deepak Kumar and Ebrahim Mookhtiar. It raised early backing from family businesses, set up a subsidiary in Montreal, Canada, and doubled revenue in FY2019-20 — then largely stayed out of the funding headlines that made Bengaluru’s younger AI startups famous. That is both the charm and the risk of this story. The lab calls itself “India’s first AI product lab” — the company’s own claim, and a debatable superlative — but what is verifiable is more interesting than the slogan: an AI lab that survived seventeen years on speech and language technology, got its hardware tested by the Army in forward areas, had its product unveiled by the Defence Minister, and is now being absorbed by a listed defence prime that has shipped over a thousand training systems and applied for 172+ patents. Zen’s engineers now inherit a team that has spent longer thinking about low-resource-language speech AI than most Indian startups have existed.
More Read
The caveats, stated plainly. The financial trajectory is ugly: turnover has fallen fourfold in two years, and the battlefield-management product that anchors the acquisition’s strategic case has almost no public technical detail — no architecture, no customer confirmation, no published performance data. The Mandarin translator’s Army adoption status beyond trials is also unconfirmed in public reporting; the latest independent verification we have is the 2022 symposium unveiling and the trials reported by The Hindu. None of that diminishes the engineering that is documented — offline neural speech translation in a 600-gram field device, built in Bengaluru for the world’s most linguistically complex frontier — but it does mean the Zen bet is exactly that: a bet that the technology, not the balance sheet, is the asset. On 17 October 2026, one of India’s most successful listed defence companies decided it was worth ₹135.15 crore to find out.
