NeuralSat's €2.9M Bet: 72 Satellites on a Seed Budget

€2.9 million. Seventy-two satellites.

Those two numbers sat next to each other in NeuralSat's announcement on 12 November, and the distance between them is the story. The Würzburg startup raised a €2.9 million ($3.2 million) pre-seed led by Capnamic Ventures and Bayern Kapital, with ZOHO.VC, Robin Capital, BAB, Mission E and KIEMIS Holding participating.

The money is meant to fund one demonstration node in October 2027. The 72 satellites are meant to follow from 2028.

NeuralSat wants to sell relay and shared compute. A satellite sends its traffic to a NeuralSat node, the node processes it, and downlinks what matters to a ground station. The company quotes up to 50 Gbit/s over optical and high-frequency radio links, plus hosted payload slots for customers who would rather ride than build.

Our read is blunt. A pre-seed of this size buys a proof point, not a constellation. The only figure that has to be true by the end of 2027 is the demo node.

What €2.9 million buys

Pre-seed is doing real work in that sentence. NeuralSat was founded in 2026 by Mohd Bilal, Manohar Karnal and Rishi Jaiswal, with Aurélie Bressollette listed as principal advisor. No customer has been disclosed.

The stated use of funds is narrow: an in-orbit demonstration to "validate the core elements" of the relay and processing architecture, then first operational satellites in 2028. That is a sensible sequence for hardware. It is also a schedule with two hard gates, a launch slot and a working optical link, and neither is fully under the company's control.

The investor mix explains part of the pitch. Bayern Kapital is a public venture investor in Bavaria with about €1.3 billion under management; Capnamic is a private fund. A sovereign-infrastructure story travels well with that combination. NeuralSat also sits in ESA's Business Incubation Center and NVIDIA's Inception program, two affiliations that cost little and signal a lot.

The founding team maps onto the hard parts. Bilal comes from embedded software, Jaiswal from mechatronics, Karnal from thermal engineering. Thermal is not a side issue in orbit; it is the constraint that decides how much compute a node can carry.

The downlink is the bottleneck

The case for relay starts with arithmetic. ESA projects roughly 100,000 satellites in orbit by 2030, up from 10,200 in 2024. Each generation of Earth-observation and signals payloads produces more data than the last. The path to the ground has not scaled at the same rate.

NeuralSat's answer is to separate the payload from the plumbing. Instead of every spacecraft carrying a full communications and computing stack, operators buy time on shared nodes. The satellite keeps its power and volume for the instrument. The node handles the traffic and hands back the parts that matter.

Just as the internet transformed how computers interact on Earth, we intend to transform how intelligence is connected and delivered in space.— Mohd Bilal, CEO, NeuralSat

The framing is the strongest part of the pitch and the easiest to overstate. Bilal says most satellites launch without onboard networking capability. That is largely true for cost-targeted smallsats. It is less true for the constellations that already run relay layers, and those are exactly the customers NeuralSat needs to win.

The latency argument is thinner than the capacity argument. SiliconANGLE put a ground round trip at 4 to more than 30 milliseconds, and a satellite-to-satellite hop at 3 to 30 milliseconds. Over long distances, inter-satellite links can be 30% to 50% faster than routing through the ground. Real, measurable, and not a step change for most business models.

Capacity is the better sell. A shared relay node can absorb bursty traffic that a single satellite cannot justify carrying. It can also move data to a ground station that is actually available, which matters more than raw speed when ground-station capacity is the binding constraint.

That is an economics argument, not a physics miracle. NeuralSat is betting that operators would rather rent plumbing than own it, the same trade that moved terrestrial compute to clouds.

Europe's relay gap, and the queue around it

NeuralSat is not first. Kepler sells a commercial optical relay service. Skyloom, acquired by IonQ, has deployed 84 optical terminals and a relay architecture in low Earth orbit. In November, NASA committed up to $1.07 billion to Viasat, SES and Astranis for commercial relay capacity.

Then there is the orbital-data-center crowd. Starcloud raised $250 million in August at a $2.3 billion valuation. Cowboy Space and Orbital Matter are building hardware for compute above the atmosphere. As we wrote in November, that layer is already crowded; Satlyt's orbital compute pitch is one example of how fast the pitch deck multiplies.

NeuralSat's positioning is deliberately different. It is not replacing terrestrial data centers. It is moving data that is already in space, and selling the link plus a shared processor as a service. That is a narrower claim, and a more defensible one.

Company announcements and reporting, November 2026.

The European angle adds a second buyer. ESA's incubation, Bavaria's public capital and the language of sovereignty all point at institutional customers who want relay capacity that is not entirely American. That is a real procurement trend. It is also a slow one.

The 50 Gbit/s question

One detail deserves more attention than it received. Payload, EU-Startups and Bayern Kapital say the network targets up to 50 Gbit/s. SiliconANGLE reported 50 gigabytes per second. The difference is eightfold, and it separates a strong relay link from a different class of system.

The consistent figure across the company's own announcement and two trade write-ups is 50 Gbit/s. Treat the gigabyte number as a transcription error until NeuralSat publishes a link budget.

The harder constraint is thermal. Processors in orbit have no air to carry heat away, so compute capacity is bounded by radiator area and power budget. NeuralSat's distributed-compute pitch spreads that load across multiple spacecraft, which helps. It does not make heat disappear.

Hosted payloads are the quiet commercial idea. If a customer can bolt an instrument onto a NeuralSat bus, the company earns revenue before its own constellation is complete. That is the line item to watch in 2027, because it is the only path to revenue that does not require 72 satellites.

Then there is the capital math. €2.9 million across 72 satellites is about €40,000 per spacecraft if the seed were the only money. It will not be. The gap between a seed round and a constellation is the gap NeuralSat has to close with a demo, a customer and a follow-on round, in that order.

What would make it real

Hardware companies fail on integration, not on slideware. For NeuralSat, four things would turn the plan into a business.

A demo node that actually reaches orbit in October 2027. A named hosted-payload customer before the first operational launch. A disclosed optical-terminal and ground-station supply chain. And a follow-on round large enough to buy satellites rather than prototypes.

None of those is exotic. All of them are scheduling problems, and scheduling is where pre-seed space companies usually slip.

The 72-satellite map will be redrawn several times before 2028. Constellation plans usually are. The October 2027 node is the one number that has to be true.