For the past six months, the IndiaAI Mission has been working on building speech and voice APIs. Progress, however, has been slow. GPU allotments were delayed, timelines were tight, and it was only last week that E2E Networks bagged a ₹177 crore contract to supply GPU resources to Bengaluru-based Gnani.ai to build its goal of voice and speech AI models. 

Yet, despite the activity, nothing foundational has shipped from IndiaAI so far. 

OpenAI, meanwhile, has been moving at a very different pace. It recently rolled out gpt-realtime, billed as its “most advanced” speech-to-speech model. Developers and enterprises can now build production-ready voice agents straight out of the box. 

The launch comes after the company set up an office in India, started local hiring and unveiled a tailored ChatGPT subscription plan for ₹399 per month. Moreover, the story has taken a new turn. The company is in talks with Reliance and other data centre companies like Yotta, Sify Technologies, E2E and CtrlS to potentially collaborate on a $500 billion ‘Stargate’ for India, as per a report by The Economic Times.

All signs point to what might be next. It seems only a matter of time before OpenAI brings speech APIs in Indian languages into the mix. 

What Happens to IndiaAI Then?

Though both IndiaAI and OpenAI have repeatedly stressed that there is no rivalry and only collaboration in building India’s sovereign AI, if OpenAI ships its Indian voice stack first, IndiaAI risks sliding into irrelevance. 

Indian companies have been waiting and tinkering with Sarvam AI, AI4Bharat, and CoRover.ai’s speech and voice models for a very long time. And they seem reasonably satisfied, for the time being. 

What they really want, though, are stable APIs and quick availability. With the recent allocation of GPUs to Gnani through E2E, it is safe to assume that it would take a few more months to release the model.

Gnani.ai is currently training a 14-billion-parameter voice AI model. At the same time, Sarvam has developed the Samvaad platform to enable companies to create conversational voice agents in Indic languages for their platforms, which include WhatsApp and on-call features.

AIM reached out to E2E Networks, which declined to comment about the IndiaAI Mission, while Gnani remained unresponsive at the time of publishing this article. 

Once OpenAI’s voice stack is integrated into apps and enterprise systems in India, there is no reason for companies to switch. Unless IndiaAI can offer something radically better or cheaper, it risks becoming just another bureaucratic arm that funds projects, not the entity that defines the market.

This is a real cause for concern. The IndiaAI Mission was pitched as India’s answer to sovereign AI, to make sure Indian companies and users are not forced to depend on American tech giants for core infrastructure. Furthermore, even Meta has begun ramping up, reportedly recruiting contractors in India to build a Hindi-language AI chatbot.

Read: IndiaAI, Where Are the GPUs?

But if Indian speech and voice ultimately end up running on OpenAI, the mission will have lost its most critical battle before it even starts. 

Ankush Sabharwal, founder and CEO of CoRover.ai, is optimistic. “Any model can be deployed to support streaming for real-time voice conversations like AI telephony,” he told AIM. For him, the bottleneck isn’t just about speed, but also accuracy, dialect coverage, tone, pitch and emotion. 

“Delay does not matter as long as we know what we are doing and doing it rightly. No FOMO,” Sabharwal said. In his view, the winners will be the ones with vision, unique data and real use cases. “Better adoption can give us everything—model, tech and scale.”

This is the reality. OpenAI’s Whisper, once hyped as a breakthrough ASR model, still performs poorly on many Indian languages. So far, local startups with data-trained Indic models are winning on accuracy—the ultimate make-or-break factor.

But if OpenAI fine-tunes gpt-realtime with proper Indian language coverage, most startups will switch overnight. Their priority remains reliability, latency and cost.

IndiaAI’s Stalled Speech Push

On paper, the IndiaAI Mission has all the right ingredients. The government’s Bhashini platform already supports 35 Indian languages and powers voice access to public services. Apart from IndiaAI, BharatGen—launched last year—was also pitched as a framework to combine speech, language and vision.

But execution has been slow. Proposals were called, applications submitted and deadlines missed. Six months later, the much-touted speech APIs, which were supposed to be a flagship offering, remain unreleased. The only options available today are those already live on the websites of the selected startups. 

For example, Gnani has entered the fray with Inya AI, a plug-and-play tool for creating voice and chat agents.

The delays suggest that the lag is not just about engineering complexity. IndiaAI is tied down by scale, compliance and paperwork. And that leaves room for OpenAI to move in. 

Sam Altman has openly said that GPT-5 “understands Indian languages way better”. An OpenAI Academy with the IT ministry is in the works, and the company is also planning to set up a data centre in the country.

Read: OpenAI’s Stargate Is a Threat to IndiaAI Mission

Vivekananda Pani, co-founder of Reverie Language Technologies, pointed out that companies like Gnani have been working on Indian speech tech for years “against all odds”. 

Unlike global startups, Indian firms had little external support when it came to building. “In the age of LLMs and GenAI, the need for resources is enormous. India has now been recognised as a massive market. Without timely support for these Indian endeavours, the race is almost a giveaway to the West,” Pani told AIM.

Worth the Wait Maybe

Indian startups are not waiting around—not for the government and not for Indian startups either. Most companies building voice-first apps are integrating Google Cloud’s speech-to-text and text-to-speech APIs. Enterprises, for their part, are ready to adopt. 

For example, in India, Meesho runs a voice bot in six languages that resolves 90% of customer queries at one-fifth of the cost. It handles 60,000 calls daily. For companies like these, whoever offers a more reliable API first becomes the default choice.

That is why OpenAI’s speed matters. Enterprises are not going to wait another year for IndiaAI to finish building.

Sudarshan Kamath, founder of smallest.ai, is blunt about why IndiaAI hasn’t shipped. According to him, building a speech-to-speech model in even one language isn’t rocket science. Any company with more than $10 million in funding should be able to do it in a year. 

“Then why wasn’t it done in the last two years? The problem is not capital. The problem is talent, focus and intent,” he told AIM, adding that most of the best researchers he knows have stayed away from IndiaAI. “So OpenAI might win the race, Google too, and we’ll be caught doing politics while it happens.”

That leaves IndiaAI Mission with two options. First, ship something now, even if the APIs are basic. Getting them in the hands of developers is better than being late to the market. Second, use policy. IndiaAI can subsidise usage of its APIs, or mandate them in government services. 

Without either, OpenAI will take over speech in India by default. And the clock is ticking.

The post IndiaAI Risks Losing Its Voice to OpenAI appeared first on Analytics India Magazine.