India aims for the stars with orbital datacentres

Why put datacentres in space?

For Narendra Sen, founder and CEO of NeevCloud, the case for orbital datacentres boils down to three irreplaceable advantages: energy, latency and sovereignty.

“In space, you have unlimited access to solar power, and you can simply radiate excess heat into space,” said Sen. “None of the water and energy problems that plague terrestrial facilities exist up there. That is the clearest contrast with land-based datacentres, which are today shifting from carrier-neutral to power-neutral and carbon-neutral models because power has become the primary bottleneck.”

Under the new partnership, Agnikul will provide the launch vehicle and orbital hosting platform using its extendable upper-stage rocket architecture, alleviating the need to build a dedicated satellite for orbital payloads.

In space, you have unlimited access to solar power, and you can simply radiate excess heat into space. None of the water and energy problems that plague terrestrial facilities exist up there

NeevCloud’s compute modules will leverage Agnikul’s platform, generating solar power in orbit and integrating AI chips directly within the architecture. Initial deployments are expected to consume around 10kW (kilowatts) to 15kW of power, with the potential to scale to 50kW, 200kW, or higher in future iterations.

Moin SPM, Agnikul’s co-founder and chief operating officer, explained that terrestrial datacentres are fundamentally constrained because they must be built around population clusters and power grids. “They struggle to serve remote regions, moving assets, or applications needing global low-latency coverage. Underwater facilities solve cooling efficiently but remain geography-bound and maintenance-heavy,” he said.

Priya Krishnamurthy, director at Altos India, a subsidiary of Acer active in the localisation of AI servers for the Indian market, noted that orbital datacentres open up compelling possibilities, particularly in terms of access to near-limitless solar energy, reduced dependence on terrestrial infrastructure, and the ability to operate beyond geopolitical boundaries.

“They can also significantly lower latency for certain satellite-based services and reduce land and water usage compared to traditional or underwater facilities,” she added.

According to Sen, orbital datacentre nodes can deliver sub-15ms (millisecond) response times by reducing the physical distance between compute and users, compared to the general round-trip latency of 100ms to 200ms for terrestrial datacentres. “No underwater facility can match that for globally distributed, real-time AI inference,” he said.

There’s also a broader strategic element at play. Punit Badeka, co-founder of EON Space Labs, noted: “The next war is going to be a space war. Space is the place where the next strategic significance lies, and countries will need to match infrastructure and strengths up there.”

Redefining PUE

If orbital datacentres are to compete with their on-ground counterparts, traditional metrics like power usage effectiveness (PUE) must be completely rethought.

“PUE was designed for a world where power, cooling and land were relatively predictable inputs,” said Sen. “In the AI era, we are already seeing its limitations: rack densities have moved from 8-10kW to 100-150kW. A space datacentre, powered entirely by solar and with no mechanical cooling overhead, would structurally achieve PUE figures closer to 1.0 – essentially all power goes to compute. That is theoretically superior to even the best terrestrial facilities.”

Operating from orbit could also reduce the need for multiple replicated terrestrial edge datacentres, cutting graphics processing unit (GPU) duplication, capital expenditure, land use and cooling infrastructure.

However, the environmental footprint of launching datacentre systems into space remains a major consideration. “We are clear-eyed about the launch footprint,” Sen admitted. “The environmental impact of launches needs to be addressed, and deployment consumes Earth resources that cannot be recovered. Agnikul’s convertible upper-stage model partially addresses this by repurposing hardware that would otherwise become debris.”

Engineering and maintenance challenges

The challenges facing space-based compute are not trivial. Building an orbiting datacentre means tackling radiation, component degradation, thermal management in a vacuum, limited maintenance, and a connectivity chain reliant on space links and ground stations, said Sen.

There’s also the problem of cooling. On Earth, heat dissipates mostly through convection – via air or water – a process that simply does not exist in the vacuum of space. That said, Krishnamurthy believes the absence of atmospheric heat in space and the availability of continuous solar energy reduce the need for traditional cooling infrastructure.

“However, thermal management doesn’t disappear; it simply becomes a design challenge involving radiative cooling and heat dissipation in a vacuum, which requires highly specialised engineering,” she added.

Also, unlike terrestrial or underwater facilities, human intervention is near-impossible, making maintenance of orbital datacentre systems and satellites difficult. “Repairs or upgrades would be infrequent, complex and expensive, often relying on robotics or future in-orbit servicing capabilities,” noted Krishnamurthy.

As such, operators are looking at hardware refreshes rather than field servicing. “Repairing a satellite in space would likely cost more than launching a replacement,” Sen admitted. “Instead, the plan is to de-orbit any failed or outdated module and replace it. The satellite will be intentionally withdrawn and safely de-orbited at the end of its lifecycle. It’s closer to how we think about consumer electronics than traditional datacentre operations.”

Agnikul’s Moin SPM is confident that the operational challenges of space datacentres can be overcome. “Radiation hardening, autonomous operation and debris mitigation are established design disciplines. Building for space demands rigour from day one, and that rigour is precisely what separates credible orbital infrastructure from conceptual proposals,” he said.

The use case for orbital inference

When space datacentres come online, their primary function will be AI inferencing, not training.

“Inference accounts for 10 times the demand once training is done,” Sen noted. “The planned infrastructure supports latency-sensitive defence systems, unmanned vehicles, border surveillance and remote healthcare use cases.”

Over 80% of the world’s population lives more than 200ms away from the nearest AI datacentre, rendering real-time applications like autonomous driving, remote robotic surgery, and industrial automation unreliable in vast regions of the Global South. To truly democratise AI, we must decouple it from terrestrial limitations

Processing data directly in space makes sense for Earth observation platforms, which generate huge volumes of data that are more efficiently analysed in orbit than downlinked to Earth in full. “Real-time AI inference for remote environments, vessels, aircraft and disaster zones needs compute access that ground infrastructure cannot deliver without meaningful delay,” added Moin SPM.

For governments and defence organisations, the appeal of highly secure, sovereign data processing isolated from geopolitical disruptions is compelling. Badeka said domestically controlled space infrastructure will reduce India’s reliance on foreign satellite imagery for sectors like agriculture, empowering users with real-time, actionable advice.

Timeline and market potential

The pilot launch for NeevCloud’s project is scheduled before the end of 2026, with an ambitious aim to scale to more than 600 orbital edge datacentres over the next three years.

Moin SPM described the current phase as a critical proof-of-concept. “We are validating the architecture, power systems and data pipeline before scaling. The longer-term goal is a distributed fleet of modular platforms in orbit.”

For Sen, the project is much more than a one-off experiment. “Over 80% of the world’s population lives more than 200ms away from the nearest AI datacentre, rendering real-time applications like autonomous driving, remote robotic surgery, and industrial automation unreliable in vast regions of the Global South. To truly democratise AI, we must decouple it from terrestrial limitations.”

Sen estimates the space datacentre sector could be worth $3bn to $5bn over the next two to three years, eventually swelling to a $50bn global market by 2040.

“India’s positioning is intentional,” said Sen. “This is not just about building infrastructure; it is about ensuring that the next decade of global AI intelligence – from village health posts to border surveillance – does not depend on foreign compute that can simply be switched off.”