Start a trial

Operational notes Partnerships

IHI and Kuva Space: what has actually been signed, and who is “sovereign” over the sensor

6 min read

Triangular glass prism struck by a beam of light that leaves it split into a continuous band, in black and white
A prism does not take a better photograph: it separates light into bands. That is where you learn what a thing is made of.

On 21 July 2026 the Japanese industrial group IHI Corporation and Finland’s Kuva Space announced a memorandum of understanding to explore building and operating Kuva’s Hyperfield-2 hyperspectral satellites in Japan, “including potentially manufacturing them in the country”. They would join the multi-sensor Earth-observation constellation IHI is assembling — around one hundred satellites — presented as a Japanese sovereign capability.

Two readings, useful to any buyer of technology in a public body or a large company: what has actually been signed, and what “sovereign” means when the distinguishing component is made by somebody else.

Who Kuva Space is

Kuva Space Oy was founded in 2016 in Espoo, Finland, at Vaisalantie 2-8. In November 2023 it closed a €16.6 million round led by Voima Ventures and Nordic Foodtech VC, with Earth VC, Springvest and Business Finland. Two verifiable European public contracts: €5 million from ESA in June 2023 as sole provider of hyperspectral data services for the Copernicus Contributing Missions, over five years; €1.8 million in ESA’s Civil Security from Space programme in April 2024, including automatic detection of vessels with AIS switched off. On 1 July 2026 its subsidiary Kuva US was selected by NASA under the CSDA On-Ramp 2 contract. In orbit it has two 12 kg Hyperfield-1 microsatellites; Hyperfield-2 weighs 70 kg and extends the sensors into short-wave infrared. What is sold is not the image: it is the platform that turns it into analysis, for agriculture, aquaculture, environment and carbon, and security — vessel tracking and detection of illicit activity included.

What hyperspectral actually is

It is not “a better photograph”. An optical camera records three broad bands, a multispectral one a few more separate bands. A hyperspectral sensor records tens or hundreds of narrow, contiguous bands: every pixel yields a curve, not a colour. That curve is a signature: it distinguishes materials and states, not shapes — one crop from another, an effluent from clean water, a metal from the paint imitating it. Hyperfield-2’s short-wave infrared widens that repertoire.

Hence the dual nature, not an accusation but a technical fact stated by the parties: the same capability that estimates a crop yield picks out a vessel that has switched off its transponder. IHI says it is assessing exactly that: ships not broadcasting AIS. Atsushi Sato, president of IHI’s aero-engine, space and defence business: “A secure and resilient space capability is increasingly important for Japan’s national security.” Jarkko Antila, chief executive of Kuva Space, describes “a route from demonstration to persistent hyperspectral intelligence”.

What is signed, and what is not

What is signed is a memorandum of understanding: assess priority use cases, define data and system requirements, explore industrial structures to build and operate the capability in Japan, domestic manufacturing included. Around twenty Hyperfield-2 satellites, Kuva estimates, would cover 20 million square kilometres around Japan daily; the global constellation targets 100 satellites by end-2030. Orders are expected in 2027, the year Kuva plans to demonstrate Hyperfield-2 in orbit for the first time: a demonstration that has not yet happened. Neither release states an explicit contractual condition; both make the two dates coincide. Quantities, value, binding deadlines and confirmation of manufacturing in Japan are absent from the public record.

This is not a criticism of the two companies: for this stage a memorandum is the right instrument. It is a lesson in reading. The binding part of an announcement is the part that states a quantity, a date and a consequence for non-performance. The rest is intent. Four questions, reusable on any partnership announcement:

  1. What is signed: an understanding, a framework agreement, or an order with quantities?
  2. What is conditional, and on which verifiable event: who establishes it, on what data, by when?
  3. Who pays if the condition is not met: penalties, termination, refund — or nothing?
  4. What happens to whoever already planned around the announcement: if you are the intermediate buyer, the condition’s risk is effectively yours.

The benchmark sits in the same constellation: from Iceye, IHI has ordered four radar satellites with an option for twenty more, two to be assembled and tested in Japan from September 2026. That is a contract, this is a memorandum — as with AI hardware.

The sovereignty paradox

Japan is building a capability described as sovereign, and the sensor that makes it useful comes from abroad. That is not a contradiction, provided three usually conflated things are kept apart:

  • Data sovereignty: who owns the images, where they are stored, who may copy them.
  • Operational sovereignty: who decides where the sensor points, when, and with what priority against other customers.
  • Industrial sovereignty: who knows how to build, repair and evolve the object.

They can be held separately. Local manufacturing, if confirmed, moves the third; the first two depend on the contracts and on the ground segment, not on the place of assembly.

Here the dependency does have a political umbrella, and it deserves saying: IHI’s statement places it within the Japan–Finland joint statement of February 2026 on cooperation in dual-use and advanced technology, space and satellite data included. That is a real assurance, and it is also its limit: a contractual clause survives a change of government, a joint statement does not.

The second-order consequence concerns the buyer. A capability that is sovereign but dependent on a foreign supplier for its distinguishing component is exposed to three events: export controls, a change of corporate control, and the priority the supplier gives others when capacity is scarce. The first is not theoretical. Regulation (EU) 2021/821 of 20 May 2021 makes the export of dual-use items listed in Annex I subject to authorisation (Article 3(1)); within that list, entry 6A002.b covers “monospectral imaging sensors” and “multispectral imaging sensors” designed for remote sensing applications, with thresholds on instantaneous field of view and wavelengths and an explicit mention of “space-qualified” sensors. Classification is done by the exporter before its own authority, not by the press release.

The buyer writes it down beforehand rather than discovering it later: data licence, place of storage, tasking priority, escrow of documentation, continuity on a change of control, and what happens if an export authorisation is refused. As in the export controls that hit supplies already under way, the clause is negotiated while you still have a contract to sign.

The operational lesson

The case is a space one, the lesson is not. The value does not sit in the sensor but in the chain that carries data to a decision — acquisition, processing, archive, model, the interface of the person deciding. That chain is either under the control of whoever decides, or it is a dependency, whatever you call it. This is why a dedicated AI system must be delivered in two ways, never only one: on-premise in the client’s own infrastructure, or on a dedicated CSIDIA cloud — an environment reserved for the single client, access over a dedicated VPN, data centre in Italy, premises staffed directly by us. Architecture matters more than the brand of the component.

Do you need to assess a partnership announcement, or a tender in which the distinguishing component comes from outside? Let’s talk it through in thirty minutes.

Sources