India’s AI push faces a familiar hurdle: the chip shortage.
The country pitched for building sovereign AI models at the India AI Impact Summit earlier this year. However, since last year, India increased its GPU arsenal by a mere 4,000—that’s a drop in the compute ocean, with AI giants like OpenAI training frontier models on up to 25,000 GPUs.
A US-based firm now wants to change how those chips are financed and deployed.
Compute Labs, founded in 2024 by Albert Zhang (CEO), Nikolay Filichkin (CBO), and Xingfan Xia (CTO), is positioning GPUs as an investable asset class.
The company, which last raised $3 million upon incorporating, connects investors with infrastructure demand from neocloud operators and wants to democratise access to AI infrastructure through GPU tokenisation. It structures deals so that investors own the hardware and earn returns based on usage.
Filichkin says the idea is to make computing accessible beyond large technology firms. “What we’ve developed is a way for them to invest in the fundamental infrastructure of AI, the asset class itself, which is compute,” he tells AIM.
Compute Labs structures deals through special-purpose vehicles (SPVs). These entities own the GPUs, while investors hold stakes in them. Neocloud operators then use the hardware and share revenue.
This model allows smaller infrastructure players to scale without upfront capital. It also offers investors a way to participate in AI growth without taking on equity risk.
However, investor apprehension is real. Many still treat GPUs as an unfamiliar asset, fearing they may become obsolete. “They don’t know how long this market will last,” Filichkin says, adding that questions around utilisation and residual value continue to shape investor decisions.
The question still stands that if GPUs are such high-demand, revenue-generating assets, why hasn’t traditional finance solved this already? Warren Hosseinion, Head of Capital Markets at Compute Labs, says interest has only recently picked up. “It was only in the last two years that they started getting involved in this market.”
India’s Demand, Capital’s Hesitation
IndiaAI Mission aims to make computing more accessible. Despite policy support, funding remains a bottleneck, and neocloud operators must raise capital before they can deploy GPUs.
Filichkin said Compute Labs works with such operators to bridge that gap. “They tell us we need 2,500 GPUs or 10,000 GPUs. We will go ahead and make that into an investment opportunity,” he explains.
The company then raises funds from institutional investors and family offices. It uses that capital to procure and deploy GPUs in India. These partners include operators bidding for state-backed compute contracts. Hosseinion says the firm is “in very early talks” with larger operators such as Yotta.
Compute Labs is already engaging with Indian players. “We’ve been engaged with three different local Indian neoclouds,” Filichkin notes, hinting that one of them is a larger operator involved in the AI mission.
He observes that while distributing capital into India is relatively easy, attracting investors remains a challenge due to structural constraints. These include legal enforcement issues and lower investor confidence compared to Western markets.
At the same time, India holds a unique position.
Filichkin notes that the country has better access to compute hardware than some restricted markets, like China, even as it struggles to mobilise capital at scale.
He points out that while China benefits from strong domestic reinvestment and state-backed funding, it faces restrictions on accessing certain advanced chips due to restrictions imposed by the US on NVIDIA and AMD GPUs. But the global demand is massive. Compute Labs has around 30 customers seeking $10 billion worth of compute.
In mature markets, lenders have clearer mechanisms to recover assets in case of defaults. In India, these systems are still evolving, which makes investors cautious.
To address this, Compute Labs SPVs tether escrow mechanisms to ensure cash flows remain protected and distributed transparently among investors.
What are the Remaining Risks?
Hosseinion describes the broader trend. “Humanity has decreased the cost of intelligence. But how do we drive that reduction? It’s through the use of compute.”
This shift has also changed how GPUs are valued. Older chips still retain demand due to inference workloads. “We’re seeing that for a chip like the A100, almost six years old, they’re retaining up to 65% of their value,” Hosseinion notes.
Despite strong demand, the model carries risks. Supply chain disruptions, energy costs, and geopolitical factors all affect pricing and deployment.
There are also operational challenges. Investors often want guaranteed usage before committing capital, while customers want infrastructure ready before signing contracts.
Filichkin compares the current moment to early transport systems. He refers to the chicken-and-egg problem, where demand drives infrastructure, which then drives further demand. “Everybody’s getting cars, and we need more infrastructure to make this available,” he illustrates.
For India, the question is not just about building models. It is about who owns the compute that powers them.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.