Space Lasers Are Building the Data Highway for Orbital AI

Every serious data network on Earth is a set of cables under the ocean. The fastest network now under construction is a mesh of light beams in orbit, and venture capital is betting it will carry AI traffic between continents before the decade is out.

The orbital compute race has a plumbing problem. SpaceX has filed for a constellation of up to a million satellites to host orbital data centers, with Anthropic named as a partner, and Google is exploring TPU clusters in space. As we wrote in August, Starcloud extended its raise by $250 million to put Nvidia silicon on the orbital compute map. The harder question is how the data moves.

The network is the real bottleneck

A satellite is a radio relay by default. It catches a signal from the ground and beams one back down. To move data between two spacecraft, traffic used to fall to a ground station and climb back up. That model stops working once compute is distributed across hundreds of nodes.

Radio has hard limits. The usable spectrum is crowded, international coordination is slow, and a wide beam spreads energy into open space. An optical inter-satellite link (OISL) changes the physics: a narrow infrared beam near 1550 nanometers carries far more data, packed into a beam so tight it is almost a thread.

Jordan Vannitsen, CEO of the Luxembourg-based laser firm Odysseus Space, states the logic bluntly. The binding constraint in orbital compute is transport, not processing. A satellite that cannot exchange data at high speed remains an island, no matter how fast its onboard chip is.

EON's target throughput of 2.4 terabits per second is roughly 960 times the rate of demonstrated space-to-ground links today, which run at about 2.5 Gbps. The mesh does not eliminate ground stations. It makes them optional where they are scarce: over oceans, the poles, and regions without landing points.

NASA proved the underlying technology on the Artemis II mission. Private companies including York, Kepler, and Cailabs have linked orbit and ground. What is left is scale and reliability.

Scale is the hard part. Two satellites can be racing along at more than 27,000 kilometers per hour in different directions, and the beam connecting them is narrower than a pencil at that range. The engineering that makes this work, known as pointing, acquisition, and tracking, is the story that took laser links from a handful of exotic demonstrations to tens of thousands of terminals in orbit.

EON's bet: a laser highway over the oceans

Endeavor Optical Networks was founded in May and emerged from stealth in August. The plan is about 20 satellites, each able to provide a dedicated link between two continents, with a demonstration spacecraft targeted for the end of 2027. The first fleet aims for 24-hour coverage of early customers, using redundant ground stations and weather data to keep the beam intact through cloud.

The routes are the point. Long hauls such as France to Australia, and crossings without existing infrastructure such as Africa to South America, are where undersea fiber is expensive, slow, or absent. EON intends to sell dedicated capacity to hyperscalers and AI labs that want control over their data transit.

Charlie Horowitz, EON's CEO and a former Apex Space executive, said the demonstration satellite should offer the highest optical downlink throughput yet seen: at least 800 Gbps and perhaps a terabit. His co-founder and chief technology officer, Tyler Presser, is a NASA-trained astronautical engineer.

The competition is bigger and slower. Blue Origin's TeraWave plans 5,048 satellites and up to 6 Tbps, but deployment will take years.

Data centers have high standards for quality and redundancy. Satellite internet is just now progressing from a technology of last resort to dependable, high-bandwidth infrastructure. That's not to say it will be impossible to make satellites optimized for data center connectivity, just that it will be harder and take longer than most entrepreneurs suggest.— Caleb Henry, director of research, Quilty Space

The terminal supply chain gets consolidated

The scarce asset in this build-out is the optical terminal: the telescope-and-laser assembly that holds a link while two spacecraft race past each other. Rocket Lab's CEO, Peter Beck, described the constraint directly. High-performing, cost-effective terminals have not been available at the volumes constellation operators need.

Rocket Lab closed the $155.3 million acquisition of Mynaric in April, folding in the CONDOR Mk3 terminal line and Mynaric's position inside Rocket Lab's $1.3 billion Space Development Agency contracts for 36 satellites. The deal gives Rocket Lab its first European footprint and, more importantly, production capacity for the one component every mesh constellation depends on.

Mynaric's CONDOR Mk3 terminal supports links beyond 6,500 kilometers and is built to the Space Development Agency's optical terminal standard, which exists precisely so satellites from different vendors can interoperate. That standard is the quiet unlock for the whole category.

Spire demonstrated a two-way optical link between satellites 5,000 kilometers apart. Kepler operates an optical data relay service in low Earth orbit. QOSMIC, backed by Accel and Prosus, raised $3.33 million to build optical ground stations for what SpaceNews calls the orbital data economy.

The capital is moving into the plumbing

Observable Space closed a $90 million Series A in May, led by Lux Capital with RTX Ventures co-leading. The Michigan-based company builds laser and optical ground systems for high-bandwidth downlinks from satellites and for precision tracking of space objects. The same week it announced a $94 million contract to expand its ground-based optical telescopes.

The pattern is consistent across the last two quarters. Early-stage capital is flowing into the transport layer, not just the compute nodes. RTX Ventures co-leading a ground-hardware round is a signal that defence primes would rather back the plumbing than build it in-house.

The mesh is the market.

The physics counterweight

IEEE Spectrum spent the year pushing back on the orbital data center narrative. Its editors counted about 14,500 active satellites, two-thirds of them Starlink, and worked the launch math: a million-satellite constellation would require roughly 16,666 dedicated launches, which at current cadence means decades. Starcloud's single Nvidia H100 in orbit ran with a radiator too weak to let the chip operate at full power.

The cooling arithmetic is unforgiving. A 700-watt H100 needs about 1.4 square meters of radiator at 60 degrees Celsius. A 100-megawatt orbital facility would need 2,500 such radiators, each a large radiative wing. That is why the million-satellite vision is so far from the near-term balance sheet.

None of that invalidates the laser mesh. The practical market is inference at the edge, relay capacity, and dedicated high-value routes, the exact segments EON and Observable Space are aimed at.

Does the orbital transport layer beat ocean fiber before 2030?

Signals to track

Development scenarios

Sources