Pathfinder Satellite 2026: India’s First Orbital Data Centre by Pixxel & Sarvam AI

Pixxel was founded in 2019 by Awais Ahmed and Kshitij Khandelwal — both alumni of BITS Pilani. It builds hyperspectral imaging satellites that capture data across over 250 spectral bands in the Visible and Near-Infrared (VNIR) and Short-Wave Infrared (SWIR) spectrum — capturing up to 50 times more data than conventional multispectral satellites. Key milestones:

• Named one of TIME magazine’s 100 Best Inventions (Sustainability, 2023)
• Recognised as a Technology Pioneer by the World Economic Forum (2024)
• Google led a ₹300 crore (~$36 million) funding round in 2023
• Launched three demonstration hyperspectral satellites and a commercial constellation called Fireflies
• Operates MegaPixxel — a 30,000+ sq ft Spacecraft Assembly, Integration & Testing (AIT) facility in Bengaluru; developing Gigapixxel, designed to scale to 100 satellites

Sarvam AI is an Indian full-stack AI company building sovereign language models trained entirely within India. Its Sarvam 105B model places India sixth on the Artificial Analysis Intelligence Index — currently the only credible Indian entry on the global AI benchmark leaderboard. The Pathfinder partnership extends Sarvam’s sovereign AI infrastructure from terrestrial to orbital systems.

An orbital data centre is a satellite equipped with high-performance computing hardware (GPUs) designed to process data directly in space — a form of edge computing. Instead of transmitting raw data to Earth for analysis, the satellite processes it in orbit and returns only the actionable intelligence.

The core problem with traditional Earth observation: high-resolution hyperspectral images are enormous files that are slow and expensive to downlink. An orbital data centre bypasses this — as the Sarvam partnership explains: “Instead of waiting hours for raw imagery to reach a ground station, the satellite can flag a wildfire, track a crop disease, or monitor a pipeline leak in the same pass it observes them.”

The bigger driver is the AI energy crisis on Earth:

• Terrestrial data centre electricity use: 415 TWh in 2024 (~1.5% of global electricity)
• IEA projection: 650–1,050 TWh by 2026; nearly doubling by 2030
• AI-related power consumption growing 50% annually through 2030
• Top 4 global hyperscalers: projected $715 billion AI capex in 2026
• Grid interconnection queues in prime data centre markets: 7–12 years (Northern Virginia, California, Germany)

In orbit, solar panels receive near-continuous, unfiltered sunlight — up to 8× more energy per panel than ground installations — with zero grid dependency.

Pathfinder is designed as a demonstrator mission to validate whether data-centre-grade hardware can operate reliably in space. It will test:

• Real-time AI inference and model training in orbit using frontier-class GPUs (same generation as ground-based AI training hardware)
• Power management during eclipse periods (~35–40 min per ~90 min LEO orbit)
• Thermal management in vacuum — where convection cooling is impossible; heat must be removed through radiative cooling panels
• End-to-end data pipelines from orbital capture to actionable intelligence without ground relay

The satellite will carry Pixxel’s flagship hyperspectral imaging camera, making it one of the first in the world capable of capturing high-fidelity hyperspectral data AND simultaneously analysing it in orbit using foundation AI models. Sarvam’s language models and inference platform will run directly on the satellite’s GPU compute layer — creating a fully sovereign pipeline: India-built models on an India-built satellite, with no foreign cloud or ground infrastructure dependence.

Orbital data centres face a distinctive set of engineering challenges:

• Thermal Management: Space vacuum eliminates convection (fans, liquid cooling). GPU heat must be removed through radiative cooling panels that emit infrared radiation into space. Engineering this at data-centre-grade GPU scale within 200 kg is the primary technical challenge.
• Radiation Exposure: Cosmic radiation causes “bit flips” — random changes in memory — and long-term semiconductor degradation. No commercial GPU is inherently radiation-hardened.
• Eclipse Periods: LEO satellites pass through Earth’s shadow ~35–40 minutes per 90-minute orbit. Power storage and AI workload buffering during eclipse must be designed in.
• Miniaturisation: Fitting data-centre hardware into a 200 kg spacecraft requires simultaneous miniaturisation of power, cooling, and compute. An estimated 100–500 satellites may be needed to match the compute of a single terrestrial data centre.
• Non-repairability: A satellite cannot be physically repaired once in orbit — all redundancy must be built in before launch.

Pathfinder enters a rapidly developing global competitive landscape:

• Starcloud (USA): Founded January 2024 by SpaceX Starlink and Airbus alumni. In November 2025, launched Starcloud-1 — a 60 kg satellite carrying an NVIDIA H100 GPU — becoming the first company to train an LLM in space and run a version of Google’s Gemini in orbit. Raised $170 million in Series A (March 2026), reaching a $1.1 billion unicorn valuation — fastest YC company to unicorn status. Starcloud-2 will carry the NVIDIA Blackwell B200 chip.
• Axiom Space (USA): Launched the first two Orbital Data Centre (ODC) nodes to LEO on 11 January 2026, operating on Kepler Communications’ optical relay network.
• China — “Three-Body Computing Constellation”: First cluster of 12 satellites launched on 14 May 2025. Ultimately envisions 2,800 satellites.
• ESA: On 5 May 2026, awarded a contract to Edge Aerospace under its Space Cloud programme to study orbital data centre architectures.
• Google’s Project Suncatcher and SpaceX (FCC application for up to 1 million data centre satellites) are also active in this space.

India, through Pixxel and Sarvam, enters this competition as a late but sovereign participant — building both the satellite and the AI models domestically, with an explicit technological sovereignty framing that distinguishes Pathfinder from purely commercial ventures.

The convergence of hyperspectral imaging and on-orbit AI inference has immediate applications across India’s priority sectors:

• Agriculture: Real-time detection of crop disease, pest infestations, soil moisture stress, and yield prediction using spectral signatures invisible to conventional satellites
• Disaster Management: Immediate identification of flood extents, fire fronts, landslide-affected areas, and cyclone damage — in a single orbital pass, without waiting for data to reach a ground station
• Environmental Monitoring: Detection of pipeline leaks, industrial pollution plumes, illegal mining, and deforestation — all with distinctive spectral signatures
• Urban Planning: Monitoring urban heat islands, water body health, and construction activity
• Defence and Strategic Intelligence: Near-real-time analysis of sensitive sites without dependence on foreign processing infrastructure