Philip
Johnston

A graphics card phoned home and said "Greetings, Earthlings."

In November 2025 a refrigerator-sized satellite called Starcloud-1 reached low Earth orbit carrying a single, very expensive passenger: an Nvidia H100 GPU, roughly a hundred times more powerful than any chip that had ever computed in space. A month later it did something no machine off the planet had done before. It trained an AI model. Then, in the dry register that has become its founder's signature, the spacecraft beamed back a message: "Greetings, Earthlings, or as I prefer to think of you, a fascinating collection of blue and green."

The man who wrote that line is Philip Johnston, and he would like to move your data center off the planet. Not metaphorically. Not someday. He has a launch manifest.

Johnston is co-founder and CEO of Starcloud, a company that began life in early 2024 under the less cinematic name Lumen Orbit. The premise sounds like a pitch you would politely decline: build the world's AI infrastructure in orbit. Then you look at his arithmetic, and the polite decline gets harder to deliver.

The biggest constraint for data centers on Earth today is energy - specifically the ability to build new energy infrastructure fast enough.Philip Johnston, on why he looks up

Here is the argument, stripped to the studs. AI's appetite for electricity is growing faster than any grid can be built. Data centers wait years for substations, transmission lines, and water rights. In orbit, none of those bottlenecks exist. The sun never sets on a satellite in the right path, so solar power is close to continuous. The deep cold of space is a free, infinite heat sink, so the cooling that swallows a fortune on Earth becomes a radiator panel. No grid queue, no water, no zoning board. Johnston's contention is not that space is romantic. It is that space is cheaper.

That is the contrarian core of him. A British applied mathematician who spent his early career writing high-frequency trading software, then advised national space agencies at McKinsey, Johnston tends to arrive at conclusions most people find absurd by way of a spreadsheet they cannot fault. The plan does not stop at one satellite. Starcloud intends to build a 5-gigawatt orbital facility - solar and cooling panels stretching roughly four kilometers on each side, a structure large enough to be a landmark if it were on the ground.

From trading algorithms to escape velocity.

Johnston was born in Guildford, Surrey, in December 1986, the youngest of five boys and an identical twin. (His brother Adrian went on to found his own AI company, Elyos AI - apparently the appetite is genetic.) He spent five childhood years in South Africa, took a first-class degree in applied mathematics from Nottingham, then a master's from Columbia. Later came an MBA from Wharton and an MPA in national security and technology from the Harvard Kennedy School, where he was elected student body president. Somewhere in there he became a CFA charterholder and a member of Mensa. The résumé reads like someone collecting credentials for a job that did not exist yet.

Before space there was retail. In 2021 he co-founded Opontia, a digital brand aggregator that raised more than $60 million and landed twelfth on Forbes Middle East's list of the most-funded startups. It was a fine outcome. It was also, evidently, not the thing. This is his second act, and it is the one with rockets.

Greetings, Earthlings, or as I prefer to think of you, a fascinating collection of blue and green.Starcloud-1's first transmission home, December 2025

Starcloud came together in El Segundo, California - the unglamorous Los Angeles suburb that has quietly become the workshop of the new space industry. Johnston built it with Adi Oltean, a veteran of SpaceX and Microsoft Azure, and Ezra Feilden, formerly of Airbus Defence and Space. The company went through Y Combinator's Summer 2024 batch, raised a $21 million seed, and then did the thing that turns a thesis into a company: it flew hardware.

The results stacked up fast. Starcloud-1 carried that H100 to orbit in November 2025. In December it ran Google DeepMind's Gemma model in space and trained Andrej Karpathy's nanoGPT there - the first in-orbit training of a language model. By March 2026 the company had raised a $170 million Series A and crossed a $1.1 billion valuation, making it the fastest unicorn in Y Combinator's history at just seventeen months past the program. The backers tell their own story: Nvidia, the venture arm of the chips powering the whole AI boom, and In-Q-Tel, the strategic investor with deep intelligence-community roots.

Johnston took the stage at TEDAI San Francisco in 2025 to make the case in public, and has worked the podcast circuit - Sequoia, Payload - with the steady patience of a man who knows the idea sounds mad until it doesn't. He even did a "mean tweets" bit about his own venture, which tells you something: he finds the skepticism funny, not threatening. The numbers, in his telling, do the arguing.

What makes him unusual among deep-tech founders is the flip in his framing. The space industry has spent decades figuring out how to bring data down from orbit - imagery, signals, weather. Johnston wants to send the compute up. The satellite is not the product. The satellite is the building. And if he is right that orbit is the cheaper address for a gigawatt of GPUs, then the strangest infrastructure pitch of the decade turns out to be a real-estate play - just with a much better view.