Helsing chose Berlin on Nov. 11 to show Clarity, software meant to take the stream of data arriving from drones, satellites and sensors and turn it into something an officer can act on. The company was careful about the boundary: the system sorts and proposes, and a person makes the decision.
That boundary is the product’s design and its sales argument at the same time. European ministries buying software that touches targeting need a record showing a human being chose the target, both for their own rules of engagement and for the legal exposure that follows an incident. A system that recommends and does not decide is easier to procure, easier to export and easier to defend in a parliamentary inquiry.
The comparison Helsing invited is Palantir’s Maven. Maven is the Pentagon program that applied machine learning to battlefield sensor data, and it is the reason Palantir became the reference supplier for military data software in allied capitals. Naming it sets the standard the new product will be measured against.
The European angle is the commercial one. Several governments in Europe have said they want to depend less on American suppliers of defense software, a preference that has hardened as procurement has become a question of sovereignty rather than of price. A European company selling a European product to European ministries is competing on a criterion that has nothing to do with features.
What the software does mechanically is less exotic than the framing. Battlefield data arrives in incompatible formats from systems built by different vendors over different decades. Fusion software ingests those feeds, puts them on a common map, resolves which reports describe the same object, and ranks what it finds by confidence and by relevance to the unit asking. The hard part is not the model. It is the integration, the latency and the willingness of a headquarters to trust an output it cannot reconstruct.
That last constraint shapes the design more than any algorithm. An officer who cannot see why a recommendation was made will not act on it, and a system that cannot be audited will not survive a legal review. Clarity’s pitch, that recommendations go to a person rather than a trigger, is a way of saying the audit trail stays intact.
The training data is where the product gets its claim to accuracy. Helsing says the company built Clarity on data from the war in Ukraine, where drones, electronic sensors and satellite imagery have produced a volume and variety of battlefield information that no exercise can reproduce. Real data from a real war is the most valuable input a defense software company can hold, and it is also the most politically complicated.
Using it raises questions the company will have to answer in every capital it sells to. Data collected in one conflict and used to train a system deployed elsewhere carries export, classification and consent problems that do not appear in a commercial machine-learning pipeline. Customers will ask who owns the data, whether it can leave the country it came from, and what happens when a model trained on one war is applied to another.
Palantir’s advantage is not confined to its software. It is the years of accreditation, the cleared staff, the contracts already running and the fact that a ministry that switches vendors has to rebuild its data pipelines from scratch. Switching costs in defense software are measured in years, and a new entrant has to be enough better to justify the disruption, not merely cheaper.
Helsing’s counter is that it is building from a different starting point. The company sells software first and hardware second, in a market where the primes have historically sold platforms with software attached. That model produces higher margins and faster iteration, and it also produces a company whose revenue depends on winning framework contracts rather than on delivering a program of record.
Procurement is where that ambition meets friction. Defense buying in Europe is national, and each ministry runs its own requirements, certifications and budget cycle. A product that works for one army does not automatically qualify for another, and the sales cycle runs longer than the funding cycle of most startups. Europe has spent years discussing joint procurement. It has spent less time making national systems accept each other’s approvals.
The commercial logic also depends on where the software attaches. Helsing’s drone work gives it an entry point into units that are already buying its hardware, and Clarity can follow the hardware into the same budget line. That is a faster path than selling a data platform to a headquarters that has no relationship with the company, and it explains why the launch was staged as a product event rather than as a procurement announcement.
Skepticism is warranted on one point. A demonstration in Berlin shows a system handling curated data in a controlled room. It does not show what happens when a feed drops, when two sensors disagree, or when the volume of input exceeds what the pipeline was sized for. The harder question is what the boundary means when an officer has seconds to choose. A ranked list carries most of the weight of the decision whatever the doctrine says, and that is the structural problem of every system that compresses a battlefield into options.
For the European governments that want an alternative to American suppliers, Clarity is a test of whether the preference can be converted into contracts. The software has to be good enough to be bought on its merits, and the company durable enough that a ministry can sign a multi-year agreement.
Helsing has now put a product in front of the market and a benchmark next to it. Whether Clarity becomes infrastructure depends on decisions taken in defense ministries over the next several budget cycles, not on the room in Berlin.
