Google Built the Chip. SpaceX Built the Railroad.

A small research satellite is now in orbit with Google's AI chips on board. Planet built it with Google. A SpaceX rocket carried it up on a shared ride. After it reached orbit, Google confirmed the team had made contact.

On October 1, Google said its Project Suncatcher prototype had launched aboard SpaceX's Transporter 18 mission. The first job is narrow: learn how Google's AI chips handle the conditions of space.

It is easy to look at this and see only a small research satellite. Farzad thinks that would be the mistake. In his view, the chip gets the headline and the ride is the story.

What this proves, and what it does not

Start with the limits, because Farzad does. A working commercial data center in orbit would be a much bigger achievement, and this mission does not establish that yet.

What it does show is a division of labor. Google brings the computing research. Planet brings spacecraft experience. SpaceX supplies the launch. Different companies, different legs of the same trip.

So when you ask who benefits if AI starts moving into orbit, the answer can include more than the company whose name is on the chip. Somebody has to carry the hardware up, carry the replacement hardware up, and make the next generation affordable enough to send.

The rideshare setup matters too. An experiment can travel aboard a shared SpaceX mission, which means a team can buy launch capacity without first creating a rocket company. You can be extremely ambitious about the application and still buy the ride from a specialist. A new entrant does not need to rebuild every capability underneath its idea.

The railroad test

Farzad's frame is a railroad. Make a place much easier to reach and you change what people build there. That is why launch cost is the first thing he looks at. It sets the size of the opportunity.

His illustration: picture a useful machine that costs almost as much to deliver as it costs to buy. You would be very selective about where it goes. Maybe one small installation. Maybe you kill the project. Now cut the delivery bill enough that you can send several machines, learn from them, and replace the least efficient design. The destination is the same. The economics of working there have changed.

He is clear this is an illustration and not a price forecast for Suncatcher. The point is the connection. Cheaper access lets more teams investigate orbital computing, and more attempts can produce designs nobody would fund when every try was prohibitively expensive. A launch provider gets paid for that activity before anyone knows which computer design wins.

SpaceX is not the only company that can reach orbit. Blue Origin's New Glenn reached orbit on its first mission in January 2025. Farzad's question is what a customer can reliably get from each provider for the job it actually needs done: the cost, the capacity, and the ability to keep delivering.

Cheap kilograms change the machine

Farzad thinks one benefit of cheaper launch gets missed: design freedom. When every kilogram is expensive, engineers fight every kilogram. When kilograms get affordable, they can spend that freedom making the machine more useful.

Picture two versions of the same experimental computer. One has a smaller cooling system and must run more cautiously. The other has a bigger cooling system and weighs more. Neither wins automatically; it depends on the real engineering and the real costs. But changing the price of carrying extra mass changes which trade-offs the team is even allowed to consider.

The same logic covers protection, spare capacity, and parts that are easier to manufacture. If a pricey lightweight component saves less on transport than it adds in manufacturing, a heavier one might eventually make more sense. So launch reaches well past the rocket line on a spreadsheet. It shapes the whole machine that goes on top.

Free sunlight, expensive heat

The attraction is solar energy and room to build more computing. Google's research describes satellites in an orbit chosen to keep them in sunlight for much of their operating time. Panels turn that light into electricity, and the electricity runs the chips.

Farzad's caution: sunlight hitting a panel and useful computation reaching a customer are different things. In between sit the panels, the electronics, the chips, and a way to manage heat. A workshop with free fuel still has to buy machinery, keep it running, and get the product to the buyer.

People hear that space is cold and assume the heat problem goes away. It does not. The chip still makes heat. On Earth, a fan moves warm air off a hot part. Outside a spacecraft there is essentially no air to do that, so heat has to move through the thermal system and leave as radiation. NASA's spacecraft guidance describes that distinction.

That means cooling has a size and a mass, and you pay to launch it. If the computer makes more heat than the system can reject, you cannot promise to run it harder forever. You change the design or how it operates. A heavier thermal design might be worth flying if the extra computation pays for the whole package. That is why Farzad will not judge the rocket and the computer separately.

Space also exposes electronics to radiation that designers must account for. A real satellite tests how the actual equipment behaves, along with the heat questions and the stress of getting to orbit. That is the part Farzad likes. The hardware has to teach you something, and the next design starts from measurements instead of an attractive idea. A bad result still pays. A misbehaving component found early becomes a specific problem to solve.

One satellite is not a data center

Say the tests go well. One satellite still is not a giant computer. Once several machines work together, the links between them become part of the machine.

Farzad compares it to a group project split across people in different rooms. Faster laptops help, but not if everyone spends the day waiting on each other's files. Some work splits into independent pieces. Some needs constant coordination. So adding chips does not automatically give you the performance of one larger, well-connected system.

Google's proposed architecture uses optical links between spacecraft, sending information with light. Planet has described a planned two-satellite mission targeting 2027 to test coordinated operation and high-bandwidth communication. Farzad keeps the questions separate: can the hardware run, and can many pieces cooperate? A yes on the first makes the second worth chasing. It does not answer it in advance.

The loop, if it closes

Once launch gets easier, the next constraint becomes worth solving. In Farzad's view that is how the railroad turns into a feedback loop. A team that sees a plausible path to deploying more hardware has more reason to improve it. A useful service creates demand for more deployments. More deployments support more launch and manufacturing work. Every step depends on the one before it producing something customers value. Drawing a circle between company logos does not get you the loop.

He uses shipping the same way: make moving things cheaper and people find more things worth moving. A customer who can afford one test at today's price might afford several at a lower one, compare them, and find a design worth scaling. Another customer might enter for the first time. Lower prices and a bigger business can coexist if volume grows enough, though he says plainly that a bigger response is not guaranteed for every product or every price change.

Who gets paid, and for what

Eventually the customer wants a result. Farzad's test is what it costs to deliver the same useful job from orbit compared with doing it on Earth. A business that needs a big batch of analysis done overnight compares price, completion time, and whether the output meets its requirements. It gets no extra value because a supplier used a satellite. If an orbital system can use its energy advantage and efficient hardware to offset deployment costs, it might compete for suitable workloads. If not, the attractive diagram does not pay the bills. Nobody needs to believe all computing moves to space for the research to make sense. One valuable class of work could be enough.

This is where Farzad gets excited about SpaceX, and where he draws a line. SpaceX could participate through transportation even when another company owns the computing. Competitive computing services of its own would be a separate, additional opportunity. Selling a launch, operating a customer's computer, selling computing time, and owning the application are different businesses with different costs and responsibilities.

The prototype establishes SpaceX's transportation role on this mission. It does not establish that SpaceX owns Google's AI, receives a share of Google's future computing revenue, or has a direct Google launch contract. Farzad's read is that the strategic point still holds: a company that can deploy hardware has one capability any future orbital service would need, and it can sell that capability to others or use it for its own projects.

Cheaper launch cuts both ways, and he says so. It could squeeze the money made per delivery while raising the number of deliveries. The question becomes whether the provider keeps its own costs competitive while serving a larger market. No business line gets a guarantee of enormous margins forever.

The bill, not the address

Picture ordering a computing job from a service that picks the most suitable hardware available. If some of it sits in orbit, you would probably care far more about the result and the bill than the address. Farzad is explicit that this is a possible future service and not something this prototype delivers. But it gives the engineering a destination: make the unusual location disappear from the customer's list of problems.

Farzad's bet is that cheaper access to space will allow far more useful activity than anyone can see today. Google is testing one piece of that future. SpaceX helped get it there. The real payoff, in his view, would be engineers designing machines for orbit because the numbers finally make sense, testing an approach, learning where it falls short, and sending a better version without treating every delivery as an extraordinary undertaking.

Back to that small satellite. The computer belongs to a research program with its own technical questions. The ride is an existing service. Put them together and a huge industrial idea becomes a sequence of things people can build and test. That is the railroad. Now watch what people start shipping on it.