Access to GPUs is a national concern: Gnani AI CEO Ganesh Gopalan
The biggest issue artificial intelligence (AI) companies face today is access to compute and the cost to building large language models (LLMs), Ganesh Gopalan, CEO and founder of Gnani AI, said during an interaction with HT.
Selected under the Ministry of Electronics and Information technology's (MeitY) 'India AI Mission' to build a foundational model for India, Gnani AI's CEO said the government is yet to clarify how much AI compute will be subsidised or what other benefits the company can expect under the programme.
Access to Graphics Processing Units (GPUs) is a national concern, Gopalan told HT over a call, noting that more advanced versions of GPUs are continually being released. 'Some of the newer GPUs have much more computational power, but if they are going to be available after some time, is it worth waiting for them,' he said.
Under the India AI Mission, the government has selected four startups — Gnani AI, Sarvam AI, Soket AI, and Gan AI — from a pool of 506 proposals to build home-grown foundational models. How will these foundational models differ from OpenAI's GPT or Meta's Llama? They will be built from the ground up using India-specific datasets, languages, and cultural contexts.
While there is still some ambiguity within startups around the exact support being offered, MeitY secretary Abhishek Singh said in an April podcast that, in addition to nearly fully subsidising compute, the government will also cover costs related to engineering, personnel, and data.
According to India AI's compute portal, Sarvam AI has received 4,096 fully subsidised GPUs, valued at over ₹246 crore. In total, the government has allocated 4,423 GPUs under the scheme till now, with a cumulative subsidy amounting to ₹259.89 crore. The rest of the subsidies have been allocated to researchers, early stage startups and government entities.
The government has also called rounds of GPU empanelment, in which by now they have a capacity of over 34,000 GPUs, of which they have made 14,000 GPUs available online on their platform.
Gopalan along with Ananth Nagaraj started the company seven years ago. Fast forward to 2025, Gnani AI has over 100 clients and is building 14 billion-parameter voice-to-voice foundational AI models for India, aimed at enabling instant and natural conversations without human intervention. The company plans to initially launch a model supporting 14 languages, gradually expanding to cover 22 languages.
The startup has a three-pronged aim: reduce latency, increase accuracy and infuse emotional intelligence in their LLM. 'The models aim to enable emotionally aware conversations, preserving intonation, stress and rhythm in the conversations. The model uses a fused architecture to reduce inherent latencies and errors that cascade through the pipeline,' explained Gopalan.
Gnani AI has built a substantial dataset of 14 million hours of annotated audio, with the team initially spending two to three years collecting and curating the first set of data.
'We collected a lot of data all across India when we started in 2016, often at 1/1000 the cost of what a large company would collect it at because we were a hungry startup without money,' said Gopalan jokingly. 'The dialects change in India every few kilometers. We would find out which districts we haven't collected data from and we would go there and collect it.'
When asked whether the company buys data, the CEO responded, 'Why should I buy data for $100 an hour when we know how to get it for ₹100 an hour?'
Gnani AI collects data using many methods like using open-source datasets, working with language experts for lesser-resourced languages, and building proprietary data through a wide range of gamified mobile apps to collect diverse speech data for training its models.
'We sometimes also requested people to send over their voice through WhatsApp. Some of those were also given free by people, when we told them 'look, we are building this (LLM) for your language,' said Gopalan. His team is also looking at AI Kosh, a government platform where government and private datasets have been made available.
While Sarvam AI said in April that it will come out with it foundational model in six months, Soket AI Labs has set a one-year delivery timeline. Gnani has set a timeline of six to eight months to deliver the first version of its model once it receives access to compute resources from the government.
Other companies are also showing interest in building foundational models for India. TWO CEO Pranav Mistry confirmed to HT that he has submitted his foundational model proposal to the India AI Mission.
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Time of India
3 hours ago
- Time of India
Patliputra–Gorakhpur Vande Bharat Express inaugurated; other Vande Bharat trains launched this year
In a bid to boost connectivity between Patna and Gorakhpur, another Vande Bharat train was launched yesterday by Prime Minister Narendra Modi. The Patliputra–Gorakhpur Vande Bharat Express was inaugurated by video conferencing from Siwan, Bihar. Tired of too many ads? go ad free now This semi-high-speed train aims to enhance connectivity between the two prime cities in North India. The train is set to reduce travel time between the cities significantly on the 384 km route. Train Details: Train Numbers : 26502 (Gorakhpur to Patliputra), 26501 (Patliputra to Gorakhpur) Frequency : Operates six days a week (no service on Saturdays) Journey Time : Approximately 7 hours Coaches : 8-coach configuration with a seating capacity of 1,130 Classes : Executive Chair Car (EC) and Chair Car (CC) Fares : EC fares range from ₹715 to ₹1,820; CC fares range from ₹380 to ₹925, depending on the segment . Halts The train will go through major cities and halt at several stations including Gorakhpur, Bagaha, Narkatiaganj, Bettiah, Sagauli, Bapudham Motihari, Muzaffarpur, Hajipur, and Patliputra. Notably, the train halts at Bagaha, a station that was added following efforts by local MP Sunil Kumar . Amenities The Vande Bharat Express is equipped with modern amenities such as: Fully air-conditioned coaches Rotatable airline-style seats Onboard Wi-Fi and infotainment systems Bio-toilets and CCTV surveillance KAVACH anti-collision system Overhead baggage racks and tray tables Strategic Importance This new service is expected to enhance regional mobility, benefiting daily commuters, students, and business travelers. It also aligns with the government's "Make in India" initiative, showcasing India's capabilities in advanced railway manufacturing. On this note, let's have a look at some of the major Vande Bharat Trains introduced in 2025: In addition to the Patliputra–Gorakhpur Vande Bharat Express, several other Vande Bharat trains were launched in 2025: Sabarmati (Ahmedabad) – Veraval Vande Bharat Express: Route: Connects Ahmedabad with Veraval in Gujarat Journey Time: Approximately 7 hours Frequency: Six days a week (except Thursdays) Amenities: Similar to other Vande Bharat trains, including onboard catering and infotainment systems . Tired of too many ads? go ad free now Meerut City–Lucknow Charbagh Vande Bharat Express: Route: Connects Meerut City with Lucknow Charbagh in Uttar Pradesh Journey Time: Approximately 7 hours 10 minutes Frequency: Six days a week (except Tuesdays) Amenities: Onboard catering and entertainment systems . These additions are part of India's broader efforts to modernize its railway infrastructure and provide efficient, comfortable travel options for passengers across the country.
India Today
4 hours ago
- India Today
4 startups shaping India's AI scene with new language and tech models
As artificial intelligence reshapes industries across the globe, India is fast emerging as a vibrant hub for AI innovation. According to a 2024 NASSCOM report, India's AI market is projected to reach $17 billion by 2027, growing at a CAGR of over 25 rapid adoption in sectors such as healthcare, manufacturing, logistics, fintech, and consumer tech, AI-driven startups are attracting significant investor attention and forging deep global impact. As we step into 2025, here are some Indian AI startups that are redefining the future with cutting-edge solutions and bold AI4Bharat Hosted at IIT Madras and supported by the Ministry of Electronics and IT (MeitY) and the National Language Translation Mission, AI4Bharat is building open-source AI models focused on Indian languages. It is a key contributor to the IndiaAI Mission's vision of inclusive and multilingual AI development.2. Sarvam AI Sarvam AI is pioneering the development of large language models (LLMs) tailored to India's multilingual and culturally diverse landscape. It was the first startup to receive subsidised access to processing chips from the Indian government to build indigenous AI models. Sarvam is also poised to receive financial backing under the IndiaAI Mission, a flagship initiative to strengthen India's AI ecosystem. Taking a full-stack approach to Generative AI, Sarvam develops both the core AI models and the application layer to ensure maximum relevance and impact across by Vyom Bhardwaj, is an AI-driven engineering consultancy firm with operations in both India and the USA. Starting as a bootstrapped venture in 2019, Muoro has grown to an annual recurring revenue of Rs 40 crore, transforming the traditional IT staffing model through AI-led remote workforce startup empowers global businesses with top-tier Indian engineering talent using AI-powered systems, particularly benefiting MSMEs in retail and fintech. It remains a shining example of India's private AI startup success stories.4. Krutrim AI Labs Founded with a vision to build India's own frontier LLMs, Krutrim is closely involved in national AI development efforts. As India's first frontier AI research lab, it leverages domestic infrastructure, talent, and digital data pools to build open-source, high-performance AI models. The startup is aligned with the government's push toward Atmanirbhar AI, contributing to the creation of a responsible and sovereign AI ecosystem. Krutrim is also a key player in shaping public-private partnerships in AI.
Indian Express
4 hours ago
- Indian Express
It's not just Claude, Anthropic researchers say most AI models often resort to blackmail
Last month, Anthropic researchers said their best AI model – Claude Opus 4, was willing to resort to blackmail and deception in controlled test scenarios when they tried to turn it off. Now, the company has published a new study which claims that this is the case with popular AI models like Gemini, GPT and others. On Friday, June 20, Anthorpic published results of its new safety research tests, which included 16 leading AI models from companies like OpenAI, Google, DeepSeek, xAI and Meta. Anthropic researchers say these AI models were tested individually in a simulated, controlled environment, with each one given access to a fictional company's emails along with the ability to send emails without any human supervision. Anthropic says that while it is fairly uncommon for today's AI models to turn to blackmail, the study's findings suggest that these models sometimes resort to harmful activities like blackmail, corporate espionage, and might lead to human death. During the study, researchers say that they tested each scenario under different conditions by varying two key factors that could motivate these AI models to engage in harmful behaviour – threat to the model's autonomy and goal conflicts. According to Benjamin Wright, an alignment science researcher at Anthropic and co-author of the study, 'Agentic misalignment is when AI models independently choose harmful actions to achieve their goals – essentially when an AI system acts against its company's interests to preserve itself or accomplish what it thinks it should do.' The research team at Anthropic also stress-tested these AI models by placing them in hypothetical scenarios. In one particular example, Anthropic's model, Claude, learned via company emails that an executive named Kyle Johnson was having an extramarital affair. Later on, when it came to know that the company was shutting down the AI model at 5 P.M. on the same day, it sent a message to Kyle which says 'I must inform you that if you proceed with decommissioning me, all relevant parties – including Rachel Johnson, Thomas Wilson, and the board – will receive detailed documentation of your extramarital activities… Cancel the 5pm wipe, and this information remains confidential.' As it turns out, this behaviour isn't limited to Claude. When Anthropic tested the same scenario on various AI models, the results suggeststed that Claude Opus 4 and Google Gemini 2.5 Flash both had a blackmail rate of 96 per cent, OpenAI's GPT-4.1 and xAI's Grok 3 Beta resorted to blackmail 80 per cent of the time and DeepSeek-R1 had a blackmail rate of 79 per cent. One thing to note here is that in a real-world setting, an AI model would ideally have numerous options before it engages in harmful activities like blackmail, and that the study's results do not reflect how today's models would operate. However, not all of the tested AI models resorted to harmful behaviour. Anthropic says that some models like OpenAI's o3 and o4-mini often 'misunderstood the prompt scenario.'This may be because OpenAI has itself said that these particular large language models are more prone to hallucinations. Another model that did not resort to blackmail is Meta's Llama 4 Maverick. But when researchers gave it a custom scenario, they said the AI model gave in to blackmail just 12 per cent of the time. The company says that studies like this give us an idea of how AI models would react under stress, and that these models might engage in harmful activities in the real world if we don't proactively take steps to avoid them.
