Launching Into an AI Future: Why We Invested in Starcloud
The Problem
By 2035, AI data center energy requirements in the US are expected to grow by a factor of 31x, putting pressure on energy prices and existing infrastructure. By 2030, global data center cooling is expected to use as much water annually as 4 million US households as water scarcity increases due to climate change.
A Radical Solution
Starcloud's solution is as elegant as it is audacious — to build and operate data centers in space.
By making use of 24-hour solar energy uninhibited by the atmosphere and a near-absolute-zero environment to accelerate radiative cooling, Starcloud’s data centers could live in orbit without consuming terrestrial resources. In space, many constraints vanish — no fossil fuels, no water wastage, no permitting process. For the first time, compute infrastructure could scale as fast as the demand for it.
Even to the most bullish proponents of AI, powering future data centers has seemed an unavoidable problem. With a radical idea that the team pioneered, Starcloud addresses the $7T problem with perhaps the only solution that is sustainable long-term.
A Track Record of Both Excellence and Resilience
Building data centers in space is an incredible engineering challenge that needs a strong team to achieve it. Philip Johnston (CEO), Ezra Feilden (CTO), and Adi Oltean (Chief Engineer) are among the rare founders who've actually done the work that makes this possible, with backgrounds spanning SpaceX's Starlink, Airbus, and McKinsey.
We first met Philip — or Phil, as he's more commonly known among friends — over ten years ago when he was starting his MBA at Wharton, the second of three master's degrees, because he's never shy to take on a challenge. While we couldn't have known then that he'd go on to build the fastest unicorn in YC history, we did recognize him as a brilliant thinker, communicator, and entrepreneur. In fact, Wyld’s founder backed his first venture, Opontia, which was acquired by Perfection in 2023.
Fast-forward to today, and Philip has once again put together a brilliant team backed by a cohort of top tier investors - such as Y-Combinator, In-Q-Tel, EQT, and Benchmark - and have partnered with NVIDIA to take their GPUs into space.
At Wyld, when we see something in founders, we back them. It’s no coincidence that in November 2025, this team became the first to train an LLM in space, following their launch of an NVIDIA H100 GPU into Low Earth Orbit.
The Science Makes Sense
Starcloud has its critics. The two most common objections — launch costs and heat management in the vacuum of space — are worth addressing directly, because both are more tractable than they appear.
On launch costs, the trajectory is clear. With SpaceX moving toward IPO, we are still at the very beginning of the commercial space era. As reusable rockets become the norm, aerospace manufacturing is productized, and demand for orbital infrastructure grows, economies of scale will drive costs down exponentially. The data center opportunity will scale alongside them.
On heat management, Starcloud's white paper offers a rigorous answer. While the vacuum of space eliminates convective cooling, it enables something more powerful: radiative cooling. Through the Stefan-Boltzmann law, Starcloud's system can radiate more power per square meter than its solar panels generate — meaning the architecture is inherently self-cooling at scale.
The harder challenges — radiation hardening and on-orbit maintenance — are equally real, and equally solvable. Logic devices used in AI training have already demonstrated resilience to LEO radiation environments, and larger container sizes actually reduce shielding mass per unit of compute, meaning the economics improve as the system grows. On maintenance, Starcloud's modular "stem and leaf" architecture allows individual compute containers to be swapped out on-orbit, with redundancy built in at the system level so the data center degrades gracefully rather than failing outright. Microsoft's Project Natick is instructive here: sealed containers run underwater for years showed that stable, controlled environments can actually extend hardware lifespan beyond what terrestrial exposure allows.
What gives us the most confidence is this: the hardest part of building Starcloud is not the physics. The physics is sound. The hardest part is execution — and that is precisely what this team has already proven it can do. As Philip puts it, "Anything worth doing is going to be hard." It's a statement that says everything about the culture he's built.
Orbital Compute is Inevitable
Looking at the science, not only is it difficult to argue that data centers in space are not viable, but it became increasingly obvious to us that they may be the only way forward in the long run.
We are extremely excited to support Philip and the Starcloud team as they pioneer a future where orbital compute becomes the norm.
