A new chapter in the global competition for space-based intelligence is taking shape. Planet Labs, the San Francisco-based Earth observation company, has announced plans to deploy two prototype satellites under its Project SunCatcher initiative in early 2027, each carrying Google Tensor Processing Units (TPUs) designed to process artificial intelligence workloads directly in orbit — rather than relaying raw imagery to ground stations for analysis.
The announcement, made during Planet Labs' Q3 2026 earnings call on 10 December 2025, signals a fundamental rethinking of satellite architecture at a moment when governments, militaries, and humanitarian organisations across the world are demanding faster, more reliable intelligence from space.
The Latency Problem — and Why It Matters Globally
For decades, Earth observation satellites have operated on a simple principle: capture data from orbit, transmit it to the ground, and process it there. That pipeline introduces latency — sometimes measured in hours — between an event occurring and useful information reaching the people who need it. In contexts where seconds and minutes matter, this gap has real consequences.
Consider a flood spreading across Bangladesh's delta region, a wildfire igniting in Australia's interior, a suspicious vessel entering contested waters in the South China Sea, or an armed convoy moving through a conflict zone in sub-Saharan Africa. In each case, the speed at which satellite imagery can be interpreted and acted upon is not merely a technical detail — it is a matter of lives, sovereignty, and strategic advantage.
By embedding TPUs directly into the satellite bus, SunCatcher is designed to run inference models while still in orbit. A satellite passing over a disaster zone could autonomously identify affected areas and transmit only the most relevant image tiles, rather than flooding downlink bandwidth with unprocessed data. The same logic applies to detecting vessels of interest, monitoring deforestation across the Amazon basin, or tracking agricultural stress patterns across the Sahel.
Edge AI in Space: A Global Race Taking Shape
SunCatcher enters a landscape where multiple powers are investing heavily in the intersection of artificial intelligence and satellite infrastructure. China's commercial Earth observation sector — anchored by companies such as CGSTL and backed by state priorities — has expanded rapidly, with Beijing integrating AI-driven image analysis into its broader geospatial intelligence apparatus. The European Space Agency has pursued its own AI-on-board programmes, including the Phi-Sat series, which demonstrated on-orbit cloud filtering as early as 2020. India's ISRO and private players in the country's growing space sector are also exploring intelligent satellite architectures as part of New Delhi's ambitions to become a leading space economy by 2040.
Against this backdrop, the Google-Planet Labs collaboration represents a significant escalation. Google is not merely supplying hardware — the company is funding the SunCatcher R&D programme directly and has embedded its Gemini team in the effort, suggesting that the TPU deployment is tied to broader research on running large-scale AI models in severely resource-constrained environments. For Google, SunCatcher functions as a testbed for TPU performance beyond the controlled conditions of a data centre — a validation step with implications that extend well beyond space.
A Multi-Vendor Approach to Orbital Compute
Planet Labs is pursuing a parallel track with its separate Owl constellation, a 1-metre class monitoring fleet that will deploy NVIDIA GPUs, with the first technology demonstration targeted for late 2026. The simultaneous use of Google TPUs and NVIDIA GPUs across distinct satellite programmes reflects a pragmatic, multi-vendor strategy — and mirrors a broader pattern visible across the technology industry, where organisations hedge against supply chain concentration risks by diversifying their compute partnerships.
That diversification has geopolitical dimensions as well. Export controls on advanced semiconductors, which the United States has tightened significantly in recent years to restrict access by China and other competitors, make the choice of chip supplier in aerospace programmes a matter of policy as much as engineering. Satellites carrying cutting-edge AI accelerators from American manufacturers are subject to export licensing regimes that shape which international customers can access the resulting data and services.
Implications for Global Customers and Partners
Planet Labs operates one of the world's largest commercial Earth observation constellations, providing imagery and analytics to customers spanning government agencies, defence ministries, agricultural companies, humanitarian organisations, and financial institutions across dozens of countries. On-orbit AI processing, if it performs as intended, would meaningfully increase the value of that data for time-sensitive applications — particularly in regions where ground infrastructure is limited and downlink access is constrained.
For international customers in particular, the ability to receive pre-processed, analytically enriched outputs rather than raw imagery could reduce the technical burden of working with satellite data and expand access to geospatial intelligence in contexts where local processing capacity is scarce. Development agencies, climate monitoring bodies, and border security organisations operating in the Global South stand to benefit disproportionately from reduced latency and more targeted data delivery.
If SunCatcher's 2027 prototypes validate the technology, a broader commercial rollout would mark a genuine inflection point in the global satellite data industry — one where the competitive frontier shifts from who can image the Earth most frequently, to who can understand what they are seeing, fastest.