Why Fractal Analytics’ IPO is a Stress Test for India’s AI Dreams
By Pallavi Chakravorty ·
As Fractal Analytics launches its initial public offering (IPO) and heads towards becoming India’s first listed pure-play AI company, it does so with a reality check. Persistent GPU shortages have squeezed its IPO and pushed back the launch of its flagship large reasoning model (LRM).
The New York- and Mumbai-based company shrunk the issue size by 42%—from ₹4,900 crore to ₹2,834 crore—just a week before opening for subscription. Investors Apax Partners and TPG Capital also pared their exposure amid valuation concerns. Fractal now aims for a ₹15,400-crore valuation.
The price band was also fixed to ₹857–900 per share. Earlier, the top end of the band was pegged at ₹1,110 per share.
“We were advised to ‘leave money on the table,’” Co-founder and Group Chief Executive Srikanth Velamakanni had told Reuters, arguing that artificial intelligence remains a new and poorly understood category for many investors.
“Investors don’t understand AI right now. We were told to price the IPO attractively enough for investors who are new to the category to come in, learn from it and grow with it,” he said.
The deeper issue, however, is that Fractal is attempting to build India’s foundation-model playbook in real-time, under public-market scrutiny and amid acute hardware scarcity.
Its biggest vulnerability is the dependence on compute. Fractal’s reasoning LLM relies on sustained access to Nvidia’s H100 GPUs, which are globally rationed. They are deployed under the IndiaAI Mission by data centre operators, including Yotta Data Services, NxtGen, and CtrlS. This makes execution timelines unpredictable—a serious red flag for public-market investors who prize certainty.
Fractal is also caught between two worlds. Its legacy business is analytics-led, services-heavy, and people-dependent, while its future bet is capital-intensive, model-driven, and compute-hungry. That complicates both its operating model and valuation narrative.
Delayed Model, Increased Scrutiny
The 26-year-old company was selected under the IndiaAI Mission to help build India’s first LRM—a next-generation AI system designed to move beyond pattern recognition into structured reasoning and decision-making. Progress on the project, however, has been dragged by delays in securing compute.
“We don’t have the compute yet. We are waiting for the Indian government to give a green light. We need to train a large model and don’t have the number of chips required,” Velamakanni told Moneycontrol.
As a result, the model’s launch has been pushed back to August.
Fractal is entering the public markets before its most ambitious product is ready. At a time when enterprise AI firms are increasingly expected to own proprietary models to protect margins and differentiate themselves, the implications of Fractal delaying the launch of its LRM could be significant.
“In a market where proprietary models are expected to differentiate enterprise AI firms, any delay can impact customer confidence, investor expectations and competitive positioning. Fractal’s decision to go public before the model is live puts it in a visible spot,” says Sanchit Vir Gogia, Chief Analyst, Founder and CEO of Greyhound Research.
He believes the move is nevertheless calculated. “By going public now at a moderate valuation, it can fund other parts of its business while continuing to build the model in tandem. This preserves optionality. It also shifts the narrative from ‘we are betting everything on this model’ to ‘we are a diversified AI company with an LLM coming soon,’” he adds.
Going public, however, opens the company to wider scrutiny. As a private company, delays were easier to contextualise. Now, investors will relentlessly track model timelines, GPU access, and IndiaAI support.
India’s GPU Problem
The GPU shortage has emerged as a major constraint as India attempts to position itself as a global AI hub. Velamakanni highlighted the financial strain on companies building foundation models.
“The government pays only about 40% of the LLM cost. Overall, just 25% of the project is subsidised,” he told Moneycontrol.
Fractal has received ₹120 crore in government support, including access to Nvidia H100 chips, with a cap of 1,400 GPUs per month and an average allocation of around 700 GPUs.
Acknowledging the global supply crunch, Nvidia chief Jensen Huang said at the World Economic Forum that GPUs are “effectively sold out across the cloud”, with even older-generation chips proving difficult to rent.
Yotta CEO and Managing Partner Sunil Gupta has also called for a sharp increase in funding for the IndiaAI Mission to expand compute capacity. The Union Budget 2026 allocated ₹1,000 crore for the initiative, down from ₹2,000 crore last year, within a five-year budget of ₹10,000 crore. Out of last year’s allocation, only ₹800 crore was utilised.
“Two years ago, there was no demand for hundreds of GPUs. Today, we are rationing 10 or 20 GPUs. All our 8,000 H100 GPUs are fully utilised, and another 8,000 B200 GPUs will go live in March,” he told BusinessLine. “IndiaAI’s budget needs to rise from ₹10,000 crore to at least ₹50,000 crore.”
Hari Balaji, Partner in the Technology Consulting Practice at EY India, says the shortage is structural and likely to persist.
“Given the global chip shortage, Indian start-ups will face constrained access for years. This can be mitigated through government support, private investment, joint ventures with hyperscalers and innovation—as models like DeepSeek demonstrated last year,” he tells AIM.
The GPU crunch is increasingly acting as a stress test for startups. Founders who built roadmaps around assured GPU access and speculative infrastructure scaling are now being forced into difficult pivots. Many are discovering that technical expertise alone is insufficient without guaranteed access to compute capacity.
At the heart of this shift lies a stark reality: training large models, particularly those with tens of billions of parameters, is expensive and operationally fragile.
“Compute costs are now consuming up to 80% of startup funding in some cases,” Gogia explains. “Many firms that raised capital during the 2023–24 AI hype cycle are now burning cash on idle GPUs or spending weeks queuing for capacity. Worse, utilisation levels in some companies have dropped below 30%.”
Where are the GPUs?
GPUs form the backbone of LLMs, and surging global demand—combined with shortages of high-bandwidth memory and storage chips—has driven costs sharply higher.
The Economic Survey 2025–26 noted that India’s ability to build large AI data centres is constrained less by demand or funding than by access to GPUs and supporting infrastructure.
The issue has also entered the political arena. AAP MP Raghav Chadha raised the matter in Parliament, calling GPU scarcity the single biggest constraint on India’s AI ambitions.
“India’s compute pool has around 34,000 GPUs, a very small number compared to what is required globally to train advanced models,” he said in his speech, questioning the government on targets, timelines and geopolitical efforts to secure predictable access to advanced compute.
Analysts argue that a shift in architecture and shared infrastructure is essential. “Enterprises need to adopt hybrid compute models that can scale across GPU generations, vendors, and even architecture types,” says Dipankar Ghosh of Thinkthrough Consulting. “That means designing AI workloads that can run on a mix of H100s, A100s, AMD accelerators, and even CPU-GPU combinations.”
He adds that India must invest in national or regional compute hubs that allow startups and smaller enterprises to access GPU capacity without building from scratch. “While the IndiaAI Mission has taken a step in this direction, utilisation must be tracked judiciously.”
Stressing self-reliance, Balaji argues that India should incentivise domestic chipmakers to reduce long-term dependence on Nvidia, while negotiating higher GPU caps for startups in the near term.
With the IndiaAI Impact Summit just around the corner, the credibility of India’s AI ambitions will hinge on how quickly it can secure more GPUs and support companies under the IndiaAI Mission.
In many ways, Fractal’s journey is a test case for India’s AI ecosystem. It illustrates what happens when ambition collides with the realities of hardware bottlenecks and market scepticism, and shows that strategic recalibration, if communicated clearly and executed well, can be a strength rather than a liability.