Pocket FM is training its AI model to extend storytelling. Is the investment worth it?
“We are currently testing a very primitive model, which will take some time. We are also working on getting the graphics processing unit,” Prateek Dixit, co-founder and CTO of Pocket FM, told . Mint. The company’s team is already working on strengthening learning for LLM. Pocket FM expects LLM to be ready five to six months after this.
The startup plans to purchase between 30 and 50 of an NVIDIA A100 or H100 GPU in a staggered manner. These units cost between $8,000 and $25,000. Despite the high cost, Dixit believes the investment is strategic.
“You can understand that it’s not just a cost decision. It’s more of a strategic asset for how we can expand storytelling through AI,” Dixit added.
In addition to hardware, LLM drives include infrastructure upgrades, hiring skilled AI engineers, and increased R&D investment. Pocket FM currently spends 8-13% of its revenue (about $26 million) on R&D, with 40% allocating it to AI programs. This is expected to grow by 1% to 2%.
The Pocket FM model was built on top of open source basic models such as Meta’s Llama 3, which is specifically designed for storytelling. The company currently uses open source models to fine-tune the needs of a specific genre, but over time it has reached a level of smooth content quality. “With our own model training, we can improve quality more and more.”
Popular content on the platform includes drama, fantasy and thrillers. The company’s writers have created thousands of stories in these categories. The plan is to use LLM for everything from story creation and comic creation to developing stories, character arcs and even AI-based videos.
“The idea is to take these basic models and fine-tune them in different writing styles. It’s also a use case for comics,” Dixit said.
Earlier this year, Pocket FM launched Pocket Toons, its webcomic platform. To this end, the company created an AI-powered studio called Blaze making comics, “at one third of the cost, automated processes such as background rendering, scene composition and color are 20 times faster while maintaining artistic creativity,” the company said.
Pocket FM has been using Natural Language Processing (NLP) (a subset of artificial intelligence (AI)) for 10 languages that produce audio content across it. Other use cases include text summary, metadata generation, and genre labels.
“LLM forms the backbone of a powerful IP engine that not only drives our audio formats, but will also power future innovations in multiple storytelling media,” Dixit said.
Is the investment worth it?
Experts can help in the long run on whether building a domain-specific language model (DSLM) is worth the price.
Considering the speed of development of the AI industry, it is hard to say whether DSLM can withstand the test of time. “I don’t think building proprietary models is a good idea, and it’s so fast in innovation in the industry,” said Anushree Verma, senior director of Gartner.
According to Gartner, business spending for this type of number is expected to reach $838 million in 2025 and grow to $11.3 billion by 2028. The market is expected to grow at a CAGR of 233%.
Verma added that open source generative AI models are becoming a viable source for domain-specific models, rapidly closing performance and reliability gaps through proprietary models and providing cost-effective and flexible alternatives to model training and specialization.
However, given that Pocket FM knows the use cases it wants to build, custom LLMs can also benefit them in the long run. More importantly, building a method that doesn’t require them to rent a GPU for training means they don’t have to worry about data security issues and protect their intellectual property rights.
“Operating your own optimization model over time, especially using an open source foundation can cut inference costs by up to 80% over time,” said Sameer Jain, managing director of Primus Partners, a global management consulting firm.
Inference is the process by which a trained AI model uses its existing knowledge to draw its own conclusions on data it has never seen before.
Ultimately, the company expects that by having its own GPU and LLM, it will be able to significantly reduce its AI costs. “The unit cost of each generation of content on scale is reduced. We are not talking about one-time use cases. We want to constantly generate inferences from the model,” Dixit said.
Deeper AI drive
In addition to LLM, the Pocket FM also has a co-pilot, which is used to create content internally in German, English and Hindi. The company is still fine-tuning it to work with other indicator languages such as Tamil, Telegu, Kannada, Marathi and Bengali. “We will release this tool publicly within a few months,” Dixit said.
The company is also building AI agents that can participate in every step of the story creation process, from the beginning of the story to where the cliff hanger might be suitable for adding.
“We are building them in a way that a single module can act and trigger separately. A story might have a very good cliff hanger, but the pace is not good. I should be able to ask the model to solve these specific queries,” said Pocket FM co-founder.
Meanwhile, Pocket FM is considering the acquisition for the first time, considering two strategies: Lean AI company builds LLM for stories or AI-based voice and video and video, followed by companies with a large community of writers.
“We are building an AI entertainment suite, so it’s great to have companies that can bake into our systems,” Dixit said. Although the company hasn’t actively set aside funds for inorganic growth, they say they will have opportunism about the acquisition.
Pocket FM is knocking on the door to private equity players around the world as it appears to raise another round of funding. The company hopes to raise $100 million to $200 million, this time at a unicorn valuation.vccircle In March.
The company last raised a $103 million Series D funding round in March 2024, led by Lightspeed India Partners, with a valuation of $750 million. So far, the company has raised a total of $170 million in rounds and has products like brand capital, Tencent, Stepstone Group, and the like on its cap table.
Pocket FM competitor Kuku FM also leverages AI for similar use cases. Kuku FM uses AI on its platform to create scripts such as “Secret Billionaire”, “Prison Women” and “Bloodstone Fortune”.
Throughout the industry, companies now choose to build their own models as they hope to take advantage of the large amount of user data they have collected over the years. Hygiene and Health Startups build their own small language model that runs on Openai and Anthropic’s LLM. ED-Tech Startup PhysicsWallah is building smaller models to solve problems related to physics, chemistry, mathematics and biology.
This year’s strategy
Although the United States has always been the main source of revenue for FM, accounting for 70-75% of total revenue, the company expects European markets to take off this year.
Last year, the company expanded to Europe and Latin America. Currently, Pocket FM is available in Germany and the UK, and the company entered the company only six months ago and claimed that both markets have contributed 5% of its revenue.
Instead of choosing to live throughout Europe at the same time, they spread their way into different countries. They will live in France around October and finally in the Netherlands around January 2026.
Dixit expects Europe to contribute up to 30% in two years. As a result, the company said their revenue will “grow multiple”. India currently contributes 10-15% to the revenue of pocket FM. Pocket FM expects revenue percentage contribution to remain unchanged, while “it is expected to grow significantly, possibly 2-3 times”.
The company claims that its revenue exceeded $200 million in fiscal 25 years and annual recurring revenue was $250 million.
In FY24, Pocket FM’s revenue station ₹261 million, compared to ₹According to regulatory documents accessed by business intelligence platform Tofler, it was 1.3 million in fiscal year 23. The company cuts losses to ₹160 million in fiscal year 24 ₹750 million in fiscal year 23.
Founded in 2018 by Rohan Nayak, Prateek Dixit and Nishanth KS Pocket FM, it was originally an audio series platform. The company has since changed its name to Pocket Entertainment. Now it runs three vertical, pocket FM, pocket novel and pocket pinky.
