OPINION | The Great Compute Crunch: Why India’s AI destiny depends on backend infrastructure 

There is a strong link between AI infrastructure and semiconductor chips

We are currently standing at a definitive inflection point in the global technology ecosystem; one where the initial euphoria surrounding generative artificial intelligence applications is beginning to cede ground to the harsh economic realities of physical implementation.

While the public imagination remains captivated by large language models and consumer-facing chatbots, a nuanced analysis of the value chain reveals a significant divergence in long-term investability.

Reminiscent of early internet era

I have been observing this space closely, and it reminds me of the early internet era where the "application layer" eventually consolidated into a state of oligopoly. In search, Google emerged as the singular victor; in social media, Meta consolidated the market. A similar dynamic is currently unfolding within the AI application landscape. The cost of training frontier models is growing exponentially, creating a barrier to entry that is insurmountable for most startups; this leaves the front-end companies in a precarious "winner takes all" situation where the second or third player faces diminishing returns.

A defensible IP moat is hard to create

This is exactly why I fundamentally believe that for an Indian venture capital ecosystem, betting on "front-end" AI applications presents a precarious risk-reward profile. These applications often function as "thin wrappers" around foundational models provided by US hyperscalers.

Consequently, they lack defensible intellectual property moats. A feature release by a foundational model provider can render a standalone application obsolete overnight.

The backend is what will count

Instead, the strategic pivot must be toward the "infrastructure layer"; comprising sovereign cloud architectures, orbital computing arrays, and domain-specific edge silicon. This is where the durable, high-yield assets are being built.

It is of utmost importance for India to bet on this backend infrastructure because when the sector becomes very big, these backend companies are what will thrive immensely. It is akin to the "picks and shovels" philosophy; during a gold rush, the most reliable path to wealth generation is not mining for gold but supplying the essential tools required for mining.

The data center and compute deficit

When we look at the Indian macro-context, we find ourselves in a paradoxical position; we are the world's largest producer of data per capita yet remain heavily dependent on foreign infrastructure for processing and storing this data. The vast majority of India’s enterprise AI workloads reside on servers owned by foreign hyperscalers, which can lead to strategic vulnerability. Digital sovereignty requires that the entire stack, from the silicon to the software orchestration layer, be subjected to national jurisdiction and control.

This is not just about national pride; it is about economic resilience. The "cloud tax" comprising egress fees and premium support costs results in significant capital flight. We need to repatriate this capital by building indigenous infrastructure that offers better unit economics.

This thesis is what drove our conviction to back companies that are solving these fundamental problems. Take Kluisz.ai, for instance, an AI-powered cloud data and intelligence platform.

The public cloud model, while being largely used, has also shown certain inefficiencies for large scale enterprises. With AI touching every aspect of the modern day businesses, and organisations finding it complicated for managing both AI and cloud resources, it leads to overprovisioning. This means companies are paying for unused capacity in some cases. Kluisz is aiming to address this by an autonomous platform that can automate complex tasks like deployment, scaling, and security management. Developers can now use an AI native platform to deploy modular applications with ease.

For a market like India, where the developer base is massive, building a product that offers ease of public cloud with the control and cost efficiency of a private one is a game changer. Specially for the Indian SMEs and enterprises, who can now deploy sophisticated AI workloads without worrying about operational overheads.

The orbital compute revolution

However, software orchestration is only one part of the equation. We also face a looming resource crisis on the physical side. Terrestrial data centers are voracious consumers of energy and water.

In a country like India (tropical climate), the energy required for cooling down the servers is much higher than cooler countries both in terms of energy consumption and cost. This is a thermodynamic tax on our emerging AI industry.

TakeMe2Space is solving this problem not just for India but potentially for the world. Their offering is simple, beat the thermodynamic tax by moving data centres to Low Earth Orbit (LEO).

Given the physics of space, this solution, apart from being radical, offers distinct advantages like radioactive cooling and unlimited solar energy that terrestrial facilities can’t. Giants like Nvidia are also exploring and actively working on concepts of data centers in space. Their interest in this further validates the idea and its growing global appeal.

Innovating companies like TakeMe2Space are solving the compute deficiency of India from data centre lens by building the world's first "Orbital Data Centre" infrastructure that is open to all. Researchers and developers will be able to access satellite compute for $2 per minute.

By offloading this heavy energy burden from Earth’s grid to space, they are turning a thermodynamic problem into a strategic advantage. It is a bold bet; at times, we face difficulties in life when some stairs appear too steep to climb. This shatters the confidence from within. But instead of looking at the slope, if we give our maximum effort into climbing one stair at a time, we accomplish what we set out to achieve.

The silicon link

While orbital compute solves the energy problem for training and the sovereign cloud solves the management problem, we still need to address the "last mile" of AI: inference at the edge. In a world increasingly populated by smart devices, transmitting every byte of data to a central cloud is inefficient. This is why we invested in Netrasemi. They are designing high-performance Edge AI System-on-Chips (SoCs) based in Thiruvananthapuram. Unlike general-purpose GPUs which are power-hungry, Netrasemi employs a Domain-Specific Architecture (DSA) optimized specifically for vision and video analytics. Their patented "Graph Stream" execution pipeline allows data to flow between hardware kernels with minimal CPU involvement, resulting in higher performance per watt and lower latency.

There is a strong link between AI infrastructure and semiconductor chips; you cannot have true digital sovereignty without domestic silicon capability. Netrasemi is pioneering the use of "Chiplet" technology, allowing them to stitch together multiple smaller dies to create powerful processors. This modularity improves yield rates and allows for rapid customization.

Once a customer integrates a Netrasemi chip into their product line, or deploys their core banking system on Kluisz, or builds a satellite workflow around TakeMe2Space, the switching costs are high. This creates "stickiness" and predictable recurring revenue, which is a far more durable asset class than a consumer app that fights for attention every day.

The light at the end of the tunnel

The synthesis of these investments reveals a cohesive strategy. We are witnessing the emergence of a new stack where Kluisz provides the software layer for sovereign data processing, TakeMe2Space provides the energy-efficient hardware layer for large-scale training, and Netrasemi provides the efficient silicon layer for real-time inference. Together, these companies form a self-reinforcing ecosystem. A satellite built by TakeMe2Space could theoretically run on Netrasemi chips to process images in orbit, coordinated by Kluisz’s autonomous cloud and compute platform.

The backend infrastructure companies are not just support functions; in the AI era, they are the main event. By betting on them, we are betting on the digitization of the physical world and the resolve of Indian engineers to solve thermodynamic problems in space and lithographic problems on silicon.

This is a long-term play. It requires patience and a willingness to understand the macros and evolving trends to develop a foresight of what could happen in space in the coming decade. In hindsight, when we look back at this period, I am confident we will realize that these infrastructure bets were the foundation of India's digital future. Once we catch that light at the end of the tunnel, we become virtually unstoppable.

(Bhaskar Majumdar, Co-founder and Managing Partner, Unicorn India Ventures.)

Views are personal, and do not represent the stand of this publication.