Can TakeMe2Space Build The AWS Of Space?

When Indian Space Research Organisation’s (ISRO) PSLV-C61 rocket lifted off from Sriharikota in January this year, Ronak Samantray believed that the hardest part was behind him.

For nearly three years, his team at TakeMe2Space had been building towards that moment. The Hyderabad-based startup had developed a satellite carrying an NVIDIA GPU, hoping to demonstrate that AI models could process satellite imagery in orbit instead of transmitting vast amounts of raw data back to Earth for analysis.

It was an important mission for the startup to convince both customers and investors that computing in space had moved beyond a niche research project to become a commercially viable layer for compute infrastructure.

But the rocket never made it. A technical failure prevented the PSLV from placing its payload into orbit, destroying months of work in a matter of minutes. For Samantray, the setback was a bigger commercial loss than an emotional one. It delayed customer deployments, pushed back fundraising conversations and forced the startup to find another launch opportunity.

TakeMe2Space’s satellite will take off again in October, but the launch itself is only one chapter. Samantray doesn’t see his spacetech startup solving Elon Musk’s vision of making humanity a multi-planetary species or even fulfilling India’s ambitions to lead the new space race. Instead, his focus clearly remains on compute, storage, customer workloads and infrastructure economics.

Founded in 2023, TakeMe2Space is trying to turn satellites into data processing units, rather than just collecting and relaying data.

The idea isn’t novel and has repeatedly attracted scepticism, even on the global stage. Critics question whether it makes economic sense to move computing hardware hundreds of kilometres above Earth when terrestrial data centres continue to become more efficient, making the logistics of launching and repairing in-space even more cumbersome.

Samantray said this conversation is only valid if all of computing is moving into orbit. Instead, the opportunity lies in a much narrower problem: if satellites are already generating terabytes of imagery in space, why spend time and bandwidth transmitting all of it back to Earth before an AI model can begin analysing it?

It is the answer to this question that has formed the bedrock of TakeMe2Space’s three-year-long journey.

From SaaS To Space

The idea for TakeMe2Space began with a problem Samantray encountered while building NowFloats, a SaaS startup that helped small businesses build an online presence, during the peak of the pandemic.

“My (NowFloats’) revenue went down by 70%, but my AWS bill probably came down by just 10%,” he said. The reason? Data centres require land, electricity, cooling systems and thousands of servers that consume power, whether customers fully utilise them or not.

While NowFloats was eventually acquired by Reliance’s Jio Platforms in 2019, the problem stayed in the back of his mind. However, the revelation came quickly, just a year later.

In 2020, the spacetech industry was still more focused on launch services. Companies were competing to build better cameras, map the planet in greater detail or provide internet connectivity from orbit. Investors also viewed satellites merely as machines that captured data.

“Everybody was talking about Earth observation, but nobody was talking about compute and storage,” Samantray said.

With Edge computing pushing processing closer to where data is generated (such as factories and smartphones) to reduce latency and bandwidth requirements, Samantray’s thesis was that space represented the next logical extension of this trend. At the same time, the cost of reaching orbit has been falling steadily for years, driven largely by reusable rockets and higher launch frequencies.

The two serendipitous factors, namely Edge computing and falling space launch costs, paved the way for the inception of TakeMe2Space.

But the idea is not new. US-based Starcloud, formerly Lumen Orbit, has proposed deploying large-scale orbital data centres powered by solar energy, while K2 Space is building high-powered satellite platforms designed to support heavier computing workloads. Google too has been exploring space-based computing through Project Suncatcher.

However, orbital computing continues to be limited by launch costs, radiation, limited opportunities for maintenance and a nascent regulatory environment. Samantray doesn’t argue that every workload belongs in space, noting that large language models like GPT or Gemini will continue to be trained on Earth.

The use case he is chasing instead focuses on data that already originates in orbit. Earth observation satellites, maritime surveillance, disaster management and agricultural monitoring all generate enormous amounts of imagery that is sent back to Earth to identify a handful of insights customers find useful. Processing it at the place of origin would prove to be faster and cheaper, the founder believes.

This distinction also explains why TakeMe2Space has chosen not to build AI models of its own but instead the infrastructure layer beneath them. “I don’t want customers to change a single line of code. Bring your AI model. We’ll provide the infrastructure,” Samantray adds.

This also differentiates it from Pathfinder, an orbital data centre satellite being built by Pixxel in collaboration with indigenous LLM developer Sarvam AI, which will run the latter’s models in space to analyse hyperspectral imagery in real-time.

Taking Compute To Orbit

Over the past three years, the startup has been building capabilities from scratch spanning satellite hardware, radiation protection, power electronics, thermal management, software and orbital deployment.

Samantray said this is because there isn’t yet an ecosystem that can support an orbital computing business. “If we want to build data centres in space, we have to solve problems that conventional satellites were never designed for,” he adds.

Unlike imaging satellites that primarily capture and transmit data, orbital computing platforms need to run commercial-grade processors continuously in one of the harshest environments imaginable. This requires keeping GPUs cool despite the absence of air, protecting them from radiation, ensuring uninterrupted power and making sure they can operate autonomously for years without physical intervention.

To address these challenges, TakeMe2Space has built a portfolio of proprietary technologies to fuel its orbital data centre vision. Among them is RadShield, a radiation shielding material designed to extend the lifespan of data centre-grade GPUs operating in low-Earth orbit. The startup has also developed its own thermal extraction system, high-energy power electronics, lightweight satellite structures and an autonomous satellite architecture designed for large constellations.

Its roadmap further includes building low-cost solar cells and optical inter-satellite links, both targeted to become flight-ready by 2027.

The startup’s flagship satellite, MOI-1A (My Orbital Infrastructure), acts as a flying compute node. It combines a 117 TOPS (trillion operations per second) AI processor, with 16 GB of onboard memory, and a nine-band multispectral imaging payload capable of capturing Earth observation data at a ground resolution of nine metres.

But unlike conventional imaging satellites, MOI-1A is designed to analyse that data while it is still in orbit. It runs AI models directly on the satellite, allowing customers to receive processed outputs, whether crop-health indices, vessel detections or change maps, instead of waiting hours or even days for raw imagery to be downloaded, processed and interpreted on Earth.

The startup says that this approach can reduce transmission costs by as much as 85% while shrinking turnaround times from as long as 24-72 hours to just a few minutes for certain workloads.

However, the satellite is only half the product. On top of this sits its software platform OrbitLab, acting as a web-based interface where developers can deploy AI models to satellites.

Clients can track the satellite’s orbit in real time, select an area of interest, upload their own AI model, validate it in a sandbox environment and then deploy it to the satellite during an available pass. The platform supports multiple deployment formats, allowing developers to use existing machine learning workflows rather than learning space-specific software.

“The moment I build my own model, I’m competing with my customers. Through this approach, I can avoid any conflict of interest between competing customers,” Samantray said.

The startup claims that its customer list spans research institutions, universities, commercial enterprises and AI startups across geographies. For instance, analytics company Tiger Analytics is exploring agricultural applications where AI models can transmit only crop-health insights instead of the raw images themselves.

The Missing Layer Of India’s Space Economy

The journey ahead does not look easy. As per Samantary, the biggest bottleneck right now remains access to space itself as launch opportunities remain limited. A failed mission, as TakeMe2Space learnt earlier this year, could push back a startup’s timeline by months.

While most startups still depend on ISRO missions or overseas launch providers to reach orbit, the situation appears to be changing.

Skyroot Aerospace recently became India’s first private company to reach orbit with Vikram-1, while Agnikul Cosmos is working towards its own orbital-class missions and reusable launch technologies. If those startups can establish a reliable launch cadence over the coming years, players like TakeMe2Space may no longer have to wait for scarce launch windows to test and deploy new hardware.

This evolution matters much to TakeMe2Space. The startup’s next satellite is currently scheduled to fly aboard a SpaceX Falcon 9 mission in October, carrying an upgraded version of its orbital computing payload.

However, the startup’s ambitions stretch beyond a single satellite. TakeMe2Space is now looking to raise $50 Mn to $55 Mn in its upcoming Series A round to launch a six-satellite constellation that is capable of serving customers continuously rather than intermittently. It is also strengthening OrbitLab’s software capabilities and developing higher-power compute platforms.

In Samantray’s view, the satellite launch in October this year will be the startup’s first server in an orbital cloud.

However, the viability of this vision remains to be seen. The economics of orbital computing are still largely unproven, while terrestrial data centres continue to become more efficient and launch costs remain a key bottleneck. Amid this, can TakeMe2Space make computing in space work?

The post Can TakeMe2Space Build The AWS Of Space? appeared first on Inc42 Media.