"Space Data Centers Aim at the Wrong Target" — LinkedIn Article

Companion photo

AI-generated illustration (not a photograph): a small orbital data center dwarfed by Earth, with the app's thesis in one line — a golden beam of light landing on a single glowing laptop. Midnight, violet, and teal palette to match the brand. Post it as the article's hero image.

What the article points to — what we're focused on

Space data centers are having a moment. A startup called Starcloud — formerly Lumen Orbit, Y Combinator-backed, $170M Series A this spring at a $1.1B valuation — put an NVIDIA H100 in orbit in November 2025 and says that within ten years, most new data centers will be built in space for the energy. Google, Microsoft, and others have published or funded similar orbital-compute roadmaps. The pitch is genuinely clever: unlimited solar, radiative cooling to deep space, no NIMBYs, no water fights.

I think the industry is aiming at the wrong target — and I want to say why, as an engineering opinion, not a knock on the people doing honest hard work up there.

The problem with putting the data center in space is that it accepts the premise. The premise is: humanity's data keeps growing without limit, so the only question is WHERE the racks go — a desert, the ocean floor, orbit. Space answers the WHERE question beautifully. It never asks the SHOULD question: how much of this data needs to be stored and shipped at all?

Here is the uncomfortable math that no orbital roadmap addresses: most of the growth in storage is not precious data. It is redundancy — the same image uploaded at 20x the size it needs to be, the same video stored at three bitrates nobody finishes watching, cold archives kept at full resolution forever "just in case," logs nobody will ever open. When we actually compress that data before it is ever written — lossless text transforms, perceptual image re-encoding, foveated-style downscaling on video that nobody examines frame by frame — the measured ratios in our own browser-verified trials run from about 90% smaller on folder ingests to roughly 98% on JPEG re-encoding. Read that again: not 20% savings. Around 90–98% of the bytes were never needed.

A data center you never fill beats a data center you don't have to cool, don't have to launch, and don't have to maintain with robots that don't exist yet.

And it's not just storage. The second lever is where the compute happens. Nearly everything an everyday app does — rendering, diffing, converting, simulating, even AI image generation in a browser — can run on the device in your hand. Our entire platform computes in the user's browser: the server stores rows of text, not rendered video or processed images. A whole category of software that used to require a rack now requires a single file, with the data in the user's own storage. The best edge device ever manufactured is already in eight billion pockets, fully paid for, with free cooling, zero launch cost, and no latency penalty.

So here is the honest comparison, scoped carefully:

Where space data centers win, for real: dense AI training runs that genuinely need gigawatt-class continuous power and don't mind round-trip latency. If training demand grows the way its advocates project, someone will want those. I'm not claiming orbital compute is a scam — Starcloud-1 training a model in orbit is a real technical achievement, and I don't dispute it.

Where the compression-first approach wins: everything else. The daily write-amplification of redundant data. The cold archives. The app workloads that never needed a server in the first place. That is the large majority of projected growth, and the physics of "never store the byte" is better than the physics of "store the byte in vacuum." One is subtraction. The other is logistics.

I'd rather be the person who deleted the petabytes than the person who rented more racks — even racks with a great view.

None of this means data centers disappear, and I won't fib about that. Compute, networking, and latency-sensitive workloads still need real hardware on the ground. Our own numbers are labeled honestly: measured compression ratios are real; scale projections are modeled estimates, not audited results.

But if you're deciding where the next decade of infrastructure talent and capital goes, I'd bet on the subtraction: compress at write time, compute at the edge, ship software as files instead of farms. The satellite can wait. The redundant bytes can't.

What do you think — is orbital compute the future, or a beautiful answer to the wrong question?

#buildinpublic #honestsoftware #datacenters #space #compression #sustainability