From Earth to Orbit: How Lumen Orbit is Transforming AI Training with $10M Funding
Lumen Orbit has successfully secured over $10 million in funding to advance its ambitious project of establishing data centres in space, as reported by TechCrunch. This innovative startup, part of Y Combinator, aims to revolutionise artificial intelligence training by creating data centres beyond Earth’s atmosphere, with a vision that includes achieving gigawatt capacity. To kick off this venture, the company is set to launch a demonstration satellite in 2025, collaborating with Nvidia’s Inception Program.
The appeal of space-based data centres lies in their ability to harness the powerful and continuous solar energy available in orbit. Unlike terrestrial facilities, they would not be hindered by the cyclical nature of day and night or the energy losses caused by the atmosphere. Lumen Orbit asserts that this advantage could enable them to obtain energy at a staggering cost that is 22 times lower than current prices.
To effectively manage the heat generated by their operations, these space-based data centres will require substantial radiators, which will be approximately half the size of the accompanying solar arrays. Inside the compute modules, advanced cooling techniques will be essential; they are considering either direct-to-chip liquid cooling or two-phase immersion cooling. These methods are critical for achieving the necessary high power densities while maintaining a space-efficient rack configuration for optimal performance.
Heat will be effectively transferred from the modules to the radiators through the implementation of several cooling loops, utilising two-phase systems where feasible. The aim is to achieve a Power Usage Effectiveness (PUE) comparable to that of cutting-edge hyperscale terrestrial data centres.
Conversely, the company estimates that deploying its space solar arrays would require an investment of $2 million while launching the compute module would add another $5 million to the expenses. Additionally, $1.2 million would be necessary for radiation shielding, bringing the total projected cost to $8.2 million. In the harsh environment of space, the least expensive solar modules may not be the most effective option. Therefore, the company could face the necessity of overprovisioning its resources or planning for additional launches to compensate for the degradation these modules will endure after years of continuous operation and exposure to radiation.
A significant uncertainty lies in the area of launch costs. The company’s calculations are based on a projected launch cost of $30 per kilogram, leading to their $8.2 million total estimate. However, current market prices sit around $1,520 per kilogram when utilising Falcon Heavy. The anticipated advancements with SpaceX’s Starship could potentially revolutionise this landscape, especially since the company has recently succeeded in catching its booster rocket during ongoing test flights. If these innovations prove fruitful, launch costs might plummet to around $150 per kilogram for single-use rockets. Moreover, if both the booster and upper stage can achieve reusability, along with a reduction in Starship manufacturing expenses, launch costs could eventually fall within the $10 to $30 per kilogram range. However, this optimistic scenario remains uncertain and carries no guarantees.
In the unforgiving environment of space, the likelihood of system failures may far exceed that experienced on Earth, necessitating an increased number of launches dedicated to replacements rather than upgrades. This situation poses a significant risk: even a minor percentage of failures could jeopardise the integrity of the entire module, ultimately requiring its complete replacement.
Visually, the proposed system resembles a spine-like tower, with modules neatly connecting to it. These modules are organised in four distinct rows encircling the central structure. Notably, the container is engineered to connect to the main framework through a single mechanical docking port. This design facilitates seamless integration of network connections, power supply, and cooling systems with the overall data centre.
However, several uncertainties linger regarding the durability and replaceability of this central spine. It is likely to house its networking components, which may deteriorate over time. Furthermore, it remains ambiguous whether the costs associated with maintaining or replacing this spine have been factored into the overall budget calculations.
SHARE