Welcome to Issue 12, where Hera opens to AI as space defence AI advances
Last week the orbital-AI thesis was hardening from capability into infrastructure. This week that infrastructure got a map: the players building space AI now have national addresses, and the gaps between them are widening.
On the AI4Space side, ESA opened its Hera asteroid mission to outside researchers who want to run onboard-intelligence software 150 million km from Earth, and China launched twin Oriental Smart Eye hyperspectral AI satellites for Indonesia and Uzbekistan. The defence layer moved next: Neuraspace raised €15.6M to scale its dual-use Space Domain Awareness platform for government and military users, and True Anomaly executed a first in-space pursuit of an actively evading target under the Space Force's VICTUS HAZE. The section closed with LiveEO winning a German Space Agency grant for a multimodal vision foundation model fusing optical and radar imagery.
On the Space4AI side, the orbital data-center thesis picked up its first major commercial partnership — SpaceX and Nvidia disclosing a joint compute payload for the Starmind AI1 satellite, timed to SpaceX's first public earnings report. The policy response followed the same axis: ESPI warned that Europe's orbital-compute gap is widening as the U.S. and China accelerate.
Both sides now carry the same geography — the operations and the compute are being claimed nation by nation.
Specifics below.
Did you know?
We have recently launched the Bits & Orbits Insights
Check out the growing set of independent dashboards, trackers, and maps on the
AI × Space frontier:
AI4Space
ESA opens Hera asteroid mission to European researchers to run software in deep space
On 5 August, ESA opened a call for European researchers and companies to run their own software aboard the Hera spacecraft, currently en route to the Didymos–Dimorphos asteroid system some 150 million km from Earth. After Hera completes its primary planetary-defence investigation of Dimorphos — the asteroid NASA's DART spacecraft deliberately struck in 2022 to shift its orbit — the mission will be repurposed as a flying software laboratory in deep space.
Hera's onboard computer runs on a European-developed dual-core LEON3 processor, and ESA is using that architecture to isolate the guest experiments. One core operates the actual spacecraft while the second hosts a sandbox environment optimised for guest software, with facilitated access to Hera's instruments and subsystems. Hosted software runs only two to three hours at a time and is shut off immediately if any problem is detected. ESA's Hera software engineer Jorge Lopez Trescastro said setting this up "in a safe and reliable way has been really challenging, but now we are ready."
ESA is targeting experiments in onboard intelligence, image processing and AI-based autonomy that reduce reliance on ground control. Dietmar Pilz, ESA's Director of Technology, Engineering and Quality, said the agency wants "targeted, innovative experiments that push onboard intelligence beyond today's procedural boundaries." Hera mission manager Ian Carnelli stressed that "the goal is not to hand over control of Hera," but to test ideas safely in a real mission environment.
Ideas are submitted through ESA's Open Space Innovation Platform (OSIP). ESA will select winning ideas in mid-October 2026, with full implementation packages and source code due 31 May 2027, ahead of a one-month operating window in August 2027. The call extends the model of ESA's OPS-SAT CubeSat, which previously invited developers to run software in orbit — including a run of Doom.
China launches twin Oriental Smart Eye hyperspectral AI satellites for Indonesia and Uzbekistan
On August 5, a Smart Dragon-3 (SD-3) carrier rocket lifted off delivering the Oriental Smart Eye hyperspectral satellites 01 and 02 into their planned orbit. Satellite 01, also called Lampung-1, is a cooperation project with Indonesia; Satellite 02, Samarkand-2028, is a project with Uzbekistan.
Both satellites pair hyperspectral imagers with onboard AI computing modules rated at up to 400 trillion operations per second. Rather than downlinking every raw image, the satellites identify features, detect anomalies and extract information in orbit, which can shorten delivery from days to minutes. The hyperspectral imagers record a detailed spectrum for each pixel, producing "spectral fingerprints" that indicate the condition and composition of vegetation, water, soil and minerals.
Lampung-1's applications center on Lampung Province on Sumatra, where agriculture, forestry and fisheries made up more than 28 percent of the provincial economy in the third quarter of 2025; it will monitor crop growth, disaster risks, marine resources and the environment. Lampung governor Rahmat Mirzani Djausal called the launch a "historic moment," and the province is preparing to send over 100 young professionals to China for training in remote sensing and satellite communications.
Samarkand-2028 is designed to track cotton cultivation from sowing to harvest, and Uzbekcosmos has said the hyperspectral data will also produce national cotton and wheat maps. Mukhiddin Ibragimov, deputy director of the Uzbekcosmos Agency, said Uzbekistan sent a technical team to China early in the project and formed a working group with Chinese specialists to develop the onboard AI model, which focuses on crop classification including cotton verification, grain identification and vegetation monitoring. He described the project as one of the largest international space projects in Uzbekistan's history and the first time the country had joined an international space mission of such scale.
Neuraspace raises €15.6M ($18M) to scale defence-grade Space Domain Awareness
On August 5, Portuguese startup Neuraspace announced €15.6 million ($18 million) in new financing to scale its AI-powered Space Domain Awareness platform for defence and dual-use customers and expand its orbital sensing infrastructure. The round combined venture investment from Lince Capital, Explorer Investments, and Armilar Venture Partners with public funds awarded under Portugal's Recovery and Resilience Plan.
Neuraspace's revenue has grown more than 350% over the past year, with customers spanning civil clients such as ESA, defense users including the Portuguese Air Force and NATO, and commercial operators such as NanoAvionics, Spire Global, and Sidus Space. CEO Chiara Manfletti told Payload the company will invest the funds in "autonomous decision-making, explainable AI, sensor fusion, orbit determination, anomaly detection, and defence-grade operational capabilities," and will hire engineers, AI and flight-dynamics experts, and commercial teams to grow its international customer base. Neuraspace also plans to complement its network of optical telescopes with radar to add another data source for the platform.
The capital is directed largely at NeuraspaceDEF, its dual-use SDA platform for government and defense users, which fuses data from multiple sensors and uses AI to surface autonomous risk and threat assessments so operators can detect and respond to physical and cyber threats on orbit.
True Anomaly executes first defence pursuit of an evading target under Space Force VICTUS HAZE
On August 3, True Anomaly said it had become the first company to execute an in-space pursuit of an evading target, a defence demonstration under the U.S. Space Force's tactical responsive space mission VICTUS HAZE. Its Jackal spacecraft was tasked with tracking and imaging Puma, a Rocket Lab-built satellite acting as the target, while Puma actively maneuvered to shake Jackal off its tail.
The sortie escalated True Anomaly's first VICTUS HAZE objective in July, in which Jackal found and characterized a non-evasive Puma in an unknown orbit. On July 14, Space Systems Command tasked the team with pursuit; as Jackal approached, Puma broke away, forcing Jackal to track the movement, replan, and re-engage. "The mission objective was different because in this case, the target was not going to cooperate," True Anomaly COO Sarah Walter told Payload.
According to Walter, the team produced a validated mission plan within 30 minutes of tasking, then executed a 23-minute ingress and burn sequence to close in on Puma. Jackal used its "Multi-Object Detection and Tracking" algorithm to generate target orbit estimates from its own onboard measurements, closing to roughly 10 km of Puma against target guidance of 10 to 20 km. It then spiraled and circled back around Puma, giving Jackal two opportunities to capture multi-aspect imagery.
The mission carried surprises, including higher-than-normal thermal temperatures at the outset, which Walter said the team adapted to through the operators' relationship with Mosaic, True Anomaly's mission command-and-control software. Walter noted Jackal has "barely made a scratch in the fuel and the capability that we have left."
LiveEO wins German Space Agency (DLR/BMFTR) grant for multimodal Vision Foundation Model for EO
On 5 August, LiveEO announced it had won a high six-figure grant from the German Space Agency at the German Aerospace Center (DLR), funded by the Federal Ministry of Research, Technology and Space (BMFTR) under the National Program for Space and Innovation. The award backs an 18-month feasibility study, already running since 1 June 2026 and scheduled to finish 30 November 2027, for a Vision Foundation Model that detects change across multimodal remote-sensing data by fusing optical and radar (SAR) imagery in a single architecture.
The study targets a shift away from the task-specific models that dominate current Earth observation, where each new data source or use case typically requires a model trained from scratch. LiveEO's aim is a generalist model that learns how the Earth's surface looks and changes across many data types, then adapts to a range of monitoring tasks with far less specialized training. The multimodal angle addresses a long-standing split: optical imagery resolves detail but needs daylight and clear skies, while SAR penetrates clouds and works day or night yet encodes signals that are harder to interpret. Each modality has historically been processed separately.
The feasibility study will test whether a single model can reason across both modalities at once and detect change reliably even when one data source is obscured or unavailable. Change detection across optical and SAR underpins LiveEO's products, which monitor vegetation, surface movement, and third-party activity along power lines, railways, and pipelines.
Space4AI
SpaceX and Nvidia unveil Starmind AI1 orbital compute satellite partnership
On August 4, SpaceX and Nvidia announced a partnership to design the compute payload for Starmind AI1, an orbital data-center satellite meant to run AI workloads in low Earth orbit. SpaceX disclosed the deal alongside its first public quarterly earnings report.
Each Starmind unit will carry Nvidia Rubin GPUs and Vera CPUs to handle data-center-class AI processing in orbit, and SpaceX committed exclusively to Nvidia hardware for the payload. The companies have set prototype testing for early 2027. SpaceX has filed with the FCC to orbit up to one million solar-powered Starmind AI satellites. The planned constellation would link the satellites with laser inter-satellite links.
The announcement came with the financial backdrop of SpaceX's Q2 2026 results, which showed the shift toward selling compute. Starmind extends that compute buildout into orbit, though it needs still to be proven whether data center in space can work at commercial scale.
ESPI warns Europe's orbital compute gap is widening as U.S. and China accelerate space-based data centers
On August 5, the European Space Policy Institute (ESPI) warned that Europe risks becoming dependent on foreign orbital computing infrastructure as U.S. companies and China's government-led push accelerate the development of space-based data centres.
The warning follows a previous report of the institute about the topic in November 2025. It highlights a growing gap: the EU funded the ASCEND feasibility study in 2023 to assess orbital data centres, but has not yet moved to commercialise the concept. ASCEND was a Horizon Europe study led by Thales Alenia Space. It defined a minimum viable product of 10 MW and an ambition of 1 GW deployed by 2050, with preliminary conclusions describing a paradigm change involving several thousand tons of infrastructure and hundreds of launches per year.
The report specifically focused on Chinese activities, including ADAspace which deployed the first 12 satellites of a planned 2,800-satellite AI computing constellation in May 2025, with in-orbit tests demonstrating core capabilities by February 2026, while multiple Chinese ventures including China Mobile, Comospace, and Shanghai Xingshu pursue orbital data-centre projects. In the U.S., Starcloud and Cowboy Space have each reached unicorn valuations exceeding $1 billion. ESPI called on the EU to recognise orbital data centres as an emerging pillar of technological sovereignty, recommending a flagship European ODC initiative under the 2028–2034 Horizon Europe programme, early-customer contracts from ESA and the EU, and a startup-focused "European ODC Challenge."
Till next time,
Meta-beat Column of this week
Read also about the AI Pipeline that sits at the core, producing this Newsletter, including its ups and downs of this week: