Firefly Aerospace Wants to Put Artificial Intelligence in Space

Artificial intelligence is all the rage these days.

Demand for computer chips tailored to run AI software was the No. 1 reason shares of Nvidia stock tripled last year. AI helped Alphabet grow its revenues 15% last quarter. AI powered 27% growth at Meta Platforms. But what is AI really good for, aside from helping college kids write term papers and making internet search a bit faster (at the cost of having to double check every search for AI hallucinations)?

Is AI just a fad, or can it actually do some useful stuff?

One company that aims to find out is Swiss edge computing company Klepsydra, which, with a little help from Firefly Aerospace, aims to test whether AI can be used to fly spaceships later this year. In April, I had the opportunity to discuss this Klepsydra mission with Firefly CEO Bill Weber, who explained why AI may be uniquely suited for use in space.

Image source: Getty Images.

Who needs space robots?

I admit that I went into this conversation skeptical. For years I’ve seen press releases about various companies wanting to put advanced computers (and occasionally even quantum computers) on spacecraft. And my response (and perhaps yours as well) was always: “Wait. Don’t spacecraft already have computers? What’s actually new about this?”

So here’s what’s new. First, yes, spacecraft do have computers on board already, and those computers can handle routine tasks such as collecting data, transmitting data to ground control, receiving instructions, and making simple “if A, then B” decisions. To date, what they haven’t had is AI computing capability that can respond to surprising, unexpected situations and decide how to react to it in real time, without having to “phone home” to NASA for instructions.

Think of it this way (and a hat tip to Weber for coming up with the analogy). Say you’re on an airplane and the pilot has a heart attack. A passenger steps into the cockpit and radios flight traffic control for instructions on how to land the plane. Now, the controllers can do this. They can tell the passenger to pull this lever, press that button, and read back the numbers on that gauge over there. Step by step, turn by turn, the controller can (hopefully) talk the passenger down — but it’s a slow process, with the potential for error increasing at each step. A better solution is to have a copilot on the plane who can fill in when the pilot is unavailable.

And that’s what AI on a spacecraft basically is — an artificial pilot, who can fly the spacecraft in the absence of a live pilot.

Klepsydra plus Elytra

At least in theory. Sometime later this year, Firefly will launch its Elytra-1 mission, aka flight FLTA006. For the first time ever, Firefly will put its new Elytra Dawn space utility vehicle in orbit, and it will carry an edge computing platform running Klepsydra AI navigation software. Klepsydra AI models will process sensor data that Elytra collects, relay the processed data to Earth, and independently make real-time guidance, navigation, and control (GNC) decisions — without requiring specific instructions from Earth.

And Elytra will keep on testing Klepsydra’s AI software for an entire year in orbit.

The future of spaceflight

Why is this important? As Weber explains, there are two big issues with using computers on spacecraft that need to communicate with Earth-based servers. One is bandwidth — the more instructions you need to send them, the more bandwidth you need. And two is distance — the farther away from Earth the spacecraft moves, the longer it takes to send instructions over your limited bandwidth.

Putting AI on the spacecraft solves both these problems because: (1) you don’t have to send as much raw data back and forth if the AI can process the data itself, and (2) you often don’t have to wait between transmissions at all if the AI can think for itself. In a second hypothetical, Weber looks to a future where a rocket fuel depot floating between Earth and the moon might be scheduling dockings for multiple spacecraft looking to refuel at it simultaneously. The bandwidth demands for coordinating the movements of all these spacecraft turn-by-turn from ground control on Earth would be enormous, and the need to send instructions, wait for responses, and then send more instructions would be both tedious and problematic. Much better to put an AI on the fuel depot and let it make these decisions itself.

The Elytra-1 Klepsydra mission will mark a first step in figuring out whether this solution will work.

What it means for investors

Firefly hasn’t announced an IPO yet, but has dropped a few hints. But investors looking forward to a Firefly Aerospace initial public offering need to understand: Elytra is Firefly’s space tug. Klepsydra is just the AI that Elytra will be carrying, and it belongs to another company entirely.

So no, this news does not mean that Firefly has become an artificial intelligence stock.

What it does illustrate is Firefly’s long-term strategy of building spacecraft like Elytra as platforms and carriers of other companies’ advanced technology. Going forward, Elytra has lined up dozens of customers interested in using Elytra for missions including spacecraft refueling, satellite repair, and satellite repositioning, and utilizing such tech as docking mechanisms, robotic arms, and so on to enable these missions. Much like Firefly will use Klepsydra’s AI to navigate its Elytra, the company will also identify and install other companies’ tech to accomplish these other missions for its customers.

This, in a nutshell, is the investment thesis for buying into a future Firefly IPO.