By the end of 2023, the generative AI (GenAI) fever was deemed to be more than just a fad. And 2024 perhaps established the same with many calling it bigger than the internet itself. But what’s in store for GenAI in 2025?
Despite oscillating between moments of awe and scepticism, GenAI started embedding itself in the Indian tech narrative, mirroring global shifts. What began as a flurry of creative experimentation, quickly gave way to a deeper question: what next?
In 2024, enterprises and innovators alike focused on analysing specific use cases, exploring niche applications across domains, and addressing fundamental challenges in implementation.
Though many industry leaders believe that even today, the lack of AI literacy is a major challenge in the country, and Indian enterprises, startups, and the government are still just scratching the surface of GenAI, it goes without saying that the adoption rate has increased significantly in the last one year.
Indian GenAI startups have risen to the occasion, driving innovation and pivoting to solve pressing problems across sectors. As predicted, Indian language models have gained prominence, discussions around AI regulations have intensified, and besides the model and application builders, more startups have emerged as enablers of GenAI deployment and infrastructure for enterprises.
What’s Next: The GenAI Predictions For 2025
As the ecosystem continues to mature, key questions about ethics, regulation, guardrails, and maximising efficiency loom larger than ever. These challenges signal that the technology is transitioning into a phase of pragmatic evolution.
Industry experts believe that 2025 will address these issues to a large extent, bringing further innovation and reshaping key sectors – finance, healthcare, travel and tourism, and media and entertainment.
While it might be difficult to predict the exact ways in which a fast-changing technology like GenAI will take shape a year from now, there are a few key things in focus – hyper-personalisation, conversational AI taking a step forward with voice-based interactions becoming more commonplace, the emergence of sector-specific LLMs, ethical AI steadily taking centre stage, among a few others.
Meanwhile, the competition on the ground is getting stronger than ever.
“As new opportunities and use cases emerge in the next 12-18 months, only some of us will be successful. All may not be commercially successful. I think it’s kind of a lottery, and no one knows today which use cases are going to be successful,” said Ankush Sabharwal, founder and CEO of CoRover, the startup which is building models to make conversational AI more human-centric.
Conversational AI Advancing With Voice In Focus
Have you noticed how bank customer support lines have changed these days?
It is rare to see a system these days which requires you to input the details on a keypad. One can simply respond by speaking and the bot on the other side understands. This is just one instance of how voice-based solutions are getting more prominent over text.
In fact, towards the end of 2024, this trend started picking up. Gnani.ai launched a speech-to-speech large language model (LLM) that can handle over 10 Mn voice interactions daily. NPCI, IRCTC, and CoRover together introduced “Conversational Voice Payments” for UPI payments on the train booking platform.
Gaurav Kachhawa, chief product officer at Gupshup, told Inc42 that businesses are increasingly integrating speech-to-speech or speech-to-text capabilities into their customer service platforms.
Gupshup, which helps companies enhance customer experience by deploying AI agents, recently helped a global beauty and haircare brand use GenAI-based voice assistant on WhatsApp.
The process begins with a voice command from the user, for example, “I have dry hair and dandruff, which shampoo would suit me best?” Combining the power of voice inputs and GenAI, the bot gives the user a seamless shopping experience.
In another instance, Gupshup helped a logistics brand use regional language-based chats on WhatsApp to generate interest from delivery partners.
“The focus is increasingly on solutions that can assist people across languages and dialects, making services more accessible to diverse user bases. Real-time translation capabilities are becoming crucial, allowing businesses to serve global audiences effectively,” Kachhawa said.
CoRover’s Sabharwal sees the shift towards more realistic conversational bots happening in two stages – first voice and then video. He said that voice will be the first big transformation in conversational AI, which is still in its infancy, and next would be video. For instance, imagine asking a question on a telehealth platform about basic symptoms of fever. In contrast to generating text or only an audio speed, there would be models or real-life doctors answering it in real-time with lip-synced.
However, it is pertinent to mention that on the back end, CoRover is doing speech-to-text and then text-to-speech conversions to deploy the voice-based models. Sabharwal said that direct speech-to-speech models are not common yet.
Investors To Explore Domain-Specific LLMs
With too many large language models already in the market and the space of LLM builders getting saturated, the next big opportunity for startups lies in building specialised LLMs for particular sectors that can solve specific challenges in certain domains and provide better accuracy than general models.
These LLMs will excel given their ability to learn and adapt to well-defined contexts instead of trying to be the jack of all trades.
Microsoft BioGPT is one such example in the field of biomedical. Last year, Bloomberg launched its BloombergGPT, an LLM that was purpose-built from scratch only for finance. In India, too, investors are exploring opportunities here.
“As the transformer-led LLMs reach their performance asymptotes, new investments in that layer are going to slow down unless a remarkably new architecture is invented, or a specialised LLM (for example, in bio and material science) is developed,” said Hemant Mohapatra, partner at Lightspeed.
For contact, as per data on Hugging Face, currently, there are 1.5 lakh transformer-based text-generation models in the market.
Vikram Ramasubramanian, partner at Inflection Point Ventures, also believes that there is potential for LLMs specialised in the pharma and healthcare space because the sector has its own nuances.
“We’ve got LLMs for people interaction and customer engagement. We will get LLMs for manufacturing, planning, scheduling, for data analytics or the management part of it. We will also get LLMs for the finance projections, forecasting, and more for the healthcare data science and the life sciences,” he added.
The Age Of Composite AI Begins?
With the increasing dependence on GenAI to solve more business complex problems, GenAI companies are slowly integrating multiple AI methodologies to create AI solutions. Some industry leaders have dubbed this advanced AI as composite AI, which largely uses classic natural language processing (NLP), GenAI, and predictive AI to provide more customised and accurate output.
In fact, CoRover is also using this approach for its products, its founder said.
Gartner said in a research report this year that composite AI represents the next phase in AI evolution. Rather than focusing solely on GenAI, AI leaders must look to composite AI techniques as value will be largely derived from projects based on familiar AI techniques, either stand-alone or in combination with GenAI.
“For example, integrating rule-based systems with machine learning allows enterprises to handle unstructured data better, thereby enhancing their ability to derive insights from diverse datasets. By embracing composite AI, organisations can solve problems that were previously too complex for single-technique AI models to address,” the research organisation noted.
And the trend that will further accelerate the popularity of composite AI is the end of the “one-size-fits-all” model.
Personalisation Is Taking Centre Stage
It goes without saying that GenAI has so far found its biggest use cases in customer experience and going forward this is expected to remain a major trend. However, personalisation will increasingly play a key role in enhancing it further and improving customer retention.
While many industry experts believe that GenAI is still a few years away from driving hyper-personalisation, the shift towards bringing in personalised models has begun.
For instance, Gupshup’s Kachhawa said that enterprises are no longer satisfied with simple, generic replies when it comes to customer engagement. They are looking for ways to make the conversations feel more personal, meaningful, and relevant to cater to specific needs.
Gupshup claims that using its WhatsApp Business API, which sends highly personalised rich media messages, Arha Media saw a 5% conversion rate for payment messages and a 10% increase in repeat subscriptions by sending 3-4 Mn messages per month, showcasing the power of targeted and timely communication.
Anindya Das, cofounder and CTO of Neysa, also confirmed that in media and entertainment, it is working with clients and partners to help them deploy GenAI for content creation, language localisation, and hyper-personalised recommendation systems for streaming platforms.
While the focus on personalisation is increasing to meet individual and business needs, the shift has just begun. Industry experts see the need right now for the adoption of this approach in more departments in a company beyond just the marketing teams.
Agentic AI: The Next Big Leap From Traditional GenAI
With personalisation becoming the key and a growing need to improve the way LLMs currently solve problems, agentic AI is coming up as the next big wave of change in 2025. It’s also the latest buzz in Silicon Valley.
Deloitte describes autonomous GenAI agents or agentic AI as software solutions that can complete complex tasks and meet objectives with little or no human supervision.
The consultancy firm predicts that 25% of companies that use GenAI will launch agentic AI pilots or proofs of concept in 2025, which will grow to 50% in 2027. India has also started embracing this leap.
If we look at the global AI landscape, Apple recently launched ‘Apple Intelligence’, a “personal intelligence system” that combines the power of generative models with personal context to deliver intelligence that is said to be more useful and relevant.
In December 2024, Google launched its latest AI model, Gemini 2.0, which CEO Sundar Pichai dubbed as the company’s entry to ‘a new agentic era’.
In India, Hyderabad-based startup Pulse offers an agentic AI platform aimed at SaaS product teams. It aggregates and analyses multi-source data to provide actionable insights, predictive analytics, and strategic recommendations. The startup raised $1.4 Mn in its seed funding round from Endiya Partners and the founders of Zluri and Yellow.ai in November.
Gupshup told Inc42 that more than 50% of its clients are actively exploring ways to use agentic AI to streamline their existing workflows. This shift indicates that agentic AI, with its deeper capabilities, is becoming the preferred solution for forward-looking AI projects.
“While GenAI primarily augmented workflows and chatbot functionalities, agentic AI takes this further by enabling virtual agents that can engage in complex, context-driven conversations and offer personalised recommendations or services,” said Gupshup’s Kachhawa.
He said enterprises are increasingly adopting agentic AI over traditional GenAI applications due to its deeper context understanding, reasoning, and ability to automate entire business processes.
However, brokerage Bernstein in a research note titled “Apple’s AI Rollout and Agentic AI” recently noted that agentic AI remains an open research problem with no proven solutions in the space, and with many leading startups releasing initial solutions that simply don’t achieve sufficient reliability for real-world use cases.
Ethical AI & Deepfake Detection To Dominate Narratives
At the start of 2024, we had predicted that with two major elections – the US and India – around the corner, deepfake cases would emerge as a challenge. And along with law enforcement, the government and legislators, there is also a parallel effort by tech startups to create what is being called “Ethical AI” or eAI.
Though there were not any significant steps taken to ensure it and many instances of deepfake photos and videos kept shaping election narratives in the country, the conversation around global AI regulations started taking more real shape in 2024.
The Indian government’s INR 10,372 Cr IndiaAI Mission has mentioned ethical AI as one of the key objectives out of seven initiatives it wants to focus on over the next five years.
Meanwhile, moral principles like ethics cannot be implemented solely with government regulations rather it has to be embedded as the core value of organisations. Gupshup’s Kachhawa confirmed that one big thing companies are focused on right now is making their AI systems transparent.
He said that the concept of “responsible AI” has evolved beyond basic compliance to become a competitive advantage and businesses are seeing real benefits such as increased user trust and higher engagement if they make AI transparency and data privacy a priority.
It is also important to remember that sectors like BFSI have to prioritise deepfake detection as the misuse of AI might also put their customers’ data and transactions at risk. The emergence of startups like Clarity, Sentinel, and India-based Kroop AI makes this space particularly interesting.
However, IPV’s Ramasubramanian has a slightly different approach here. He believes that GenAI still is nascent and there are challenges with hallucinations. Hence, startups are emerging that can test and fine-tune the data to ensure compliance. In India, AiEnsured is one such example of a startup doing it.
Portkey is another such example whose platform enforces reliable LLM behaviour with its AI guardrails.
With that, it is interesting to note that the infrastructure layer of GenAI and LLM Ops is emerging as a key focus for investors to infuse money into.
GPU-As-A-Service To Become More Prominent
GenAI is expensive and a major cost of building foundational models is driven by the computational expense – the graphics processing units (GPUs) or high-end control processing units (CPUs).
To make it more cost-efficient, flexible, and scalable, GPU-as-a-Service (GaaS) is emerging into prominence.
Highlighting this upcoming trend, Neysa’s Das said that with a significant emphasis on cost-efficiency and scalability, many of its clients were actively looking for scalable GaaS solutions. And Neysa launched its Velocis AI Cloud, which now provides them with an end-to-end AI cloud platform that enables clients to train and fine-tune models while choosing the GPU that best aligns with their use cases.
In 2024 NVIDIA also released its Nvidia AI Enterprise product that includes services which cost $4,500 per graphics processor used per year.
CoRover’s Sabharwal added, “Whether software companies make money or not, the infrastructure providers would definitely make money. There is a growing requirement for high computing – it could be GPUs or more advanced CPUs or some alternative to GPUs emerging.”
As part of its IndiaAI MIssion, the Centre is also working to deploy over 10,000 GPUs through strategic public-private collaborations.
There are a few other developments expected in the GenAI sector, such as the increasing use of technology in data analytics, the emergence of multi-modal AI, the rising penetration of small language models, more focus on Sovereign AI, and the collaborations of Blockchain and GenAI.
In fact, commenting on this interesting trend of decentralised AI, Bruce Keith, CEO and cofounder of InvestorAi, said that blockchain technology and AI together have great potential to enhance data security.
“Blockchain gives you authentication and AI gives you automation and more consistent output, and the convergence of the two together will start to answer a lot of data security questions that organisations have. However, since blockchain and crypto are perennially interlinked and different countries have different views on crypto, it is kind of holding back the development of Blockchain and AI together but it is going to change,” said Keith.
However, there are other key aspects that need to be addressed immediately. One of these is the huge power consumption by the large data centres that are required to drive developments in AI. The regulations are also weak and ask for more government support
Jaspreet Bindra, CEO of AI&Beyond, said that INR 10,000 Cr funding by the Centre is still insufficient compared to other countries, and there is a lot of hype with less actual support. “My vision involves bringing GenAI to the masses, much like Aadhaar and UPI. This would require the government to treat GenAI with the seriousness of a national mission,” he said.
With that, it now remains to be seen if 2025 can usher in some of the biggest shifts in the GenAI ecosystem globally and if India can keep up with these shifts.
[Edited By Nikhil Subramaniam]