Talk Around Town: A Chinese Company Claims It'll Park a 1,080-Satellite AI Supercomputer in Orbit

THE QUICK TAKE

According to secondary reporting that traces back to Chinese tech outlet 36kr, a company called SpaceWit claims it plans to launch over 1,080 laser-linked satellites to run AI training and inference directly in orbit.
No Western trade press, regulatory filings, or independent analysts have corroborated SpaceWit's existence, funding, or technical specs, making this one of the thinner reeds in a fast-crowding field.
The orbital AI computing race itself is very real, with SpaceX filing FCC plans for up to one million such satellites and Starcloud confirmed to have already trained a large language model in space.

What the Chatter Says

Well, slap the barn door and call it a rocket — according to secondary reporting that traces back through Russian tech aggregator ixbt.com to Chinese outlet 36kr, a company called SpaceWit claims it has unveiled a project it describes as 'Space String.' The company says the effort would involve more than 1,080 satellites connected by laser links, all aimed at running AI model training and inference up there among the stars rather than down here where the kudzu grows. SpaceWit claims this would essentially be a distributed orbital supercomputer of its own description.

According to that same unverified source chain, SpaceWit describes its architecture as having two tiers: the company says more than 720 satellites would handle Earth observation and on-orbit AI inference tasks, while a separate group of 360-plus dedicated computing satellites would handle the heavier AI training workloads, all parked in sun-synchronous orbit. That is the company's own framing, not a finding this publication can independently confirm, because nobody outside that single source chain has weighed in.

The Timeline SpaceWit Claims

Now, if you believe the secondary reporting — and that is doing some real heavy lifting right there — SpaceWit says the project rolls out across three satellite generations it describes as G1 experimental, G2 standard, and G3 high-performance. The company reportedly says the first test satellite is slated for Q4 2027, with the full constellation characterized as a long-term goal rather than anything imminent. Whether 'long-term' means five years or 'whenever the money shows up,' the reporting does not specify, bless its heart.

The source chain for all of this is about as short as a June shadow: a Central Asian outlet summarized a Russian tech aggregator that summarized a Chinese tech site, with no original company documentation, no FCC-style regulatory filing, no funding announcement, and no independent analyst on record having heard of SpaceWit. That is not a knock on the outlets involved; it is just an honest accounting of how thin the ice is before you go skating on it.

What Is Actually Known: The Real Orbital AI Race

Here is where things get solid as a cast-iron skillet: the broader orbital AI computing race is confirmed by multiple independent top-tier sources, even if SpaceWit itself is not. In January 2026, SpaceX filed plans with the FCC for a constellation of up to one million orbital AI computing satellites, and the FCC accepted that filing on February 2, 2026, according to Data Center Dynamics and corroborating records. That is a real, verified filing from a company that has actually put hardware in orbit.

Meanwhile, GeekWire reported that Seattle-area startup Starcloud reached a $1.1 billion valuation and confirmed that its satellite — carrying an Nvidia H100 chip — successfully demonstrated AI workload processing in orbit, becoming the first outfit to train a large language model in space. On the Chinese side, ADA Space has separately deployed what it calls its 'Star-Compute Project,' and according to CB Insights, Alibaba's Qwen-3 became reportedly the first AI model to operate in orbit via that network. None of these confirmed sources mention SpaceWit or Space String anywhere.

What Remains Unverified

Lord have mercy, the list of what cannot be confirmed about SpaceWit is longer than a dirt road to nowhere. No verifiable English-language public documentation from the company has turned up. There are no regulatory filings, no independent funding records, no Western trade press mentions, and no industry analysts on record having evaluated the technical plans. The company's very existence as an operating entity — rather than a project name attached to a press release — has not been independently confirmed.

The technical specifications SpaceWit reportedly claims, the satellite generations, the orbit type, the compute capacity figures, and the 2027 launch date are all drawn entirely from that single secondary source chain. Until a primary source, a regulatory filing, or an independent technology analyst puts their name on any of it, every one of those details must be treated as an unconfirmed company claim according to what secondary reporting attributes to SpaceWit.

The 'Starlink Benchmark' Framing Is a Bit Off

The original reporting floated the notion that Space String would serve as some kind of benchmark against Starlink, which is about as accurate as calling a catfish a show dog just because they both live near water. Starlink is a broadband internet service delivering connectivity to consumers and businesses on the ground. What SpaceWit reportedly describes — and what SpaceX's own FCC filing targets — is orbital AI compute infrastructure: processing power in the sky for AI training and inference, not internet access. These are different critters wearing the same spacesuit.

Framing an orbital data center play as a Starlink rival is a marketing angle, not a technical characterization. It may be that SpaceWit's reported description of its own project uses that comparison as a benchmark for ambition or scale, but this publication's analysis is that readers should not let that framing blur the actual distinction between satellite broadband and orbital computing infrastructure.

Analysis: The Economics Are Doing a Lot of Work

This is analysis, not reporting: even if SpaceWit is a fully funded, technically capable company that somehow escaped all public documentation, the economic case for orbital AI compute remains genuinely contested. Skeptics at Varda Space Industries, cited in industry roundups, have argued that orbital computing runs roughly three times more expensive per watt than its terrestrial equivalent, which is a gap wide enough to lose a whole pickup truck in. Microsoft President Brad Smith has publicly stated he is not wagering on the orbital data-center strategy, signaling that major cloud incumbents remain unconvinced about the near-term numbers.

That said, the strategic logic has enough believers — SpaceX, Starcloud, ADA Space, and reportedly several others — that dismissing the entire concept would be premature. The confirmed players in this race are real and funded. Whether SpaceWit belongs in that conversation at all is, as of this writing, entirely unanswered. If Space String turns out to be a genuine project with real hardware and real money behind it, the crowded nature of this field means the company would be arriving late to a party that already has too many guests and not enough chairs.

Who is doing the hollering

These links show where the chatter came from. A link is attribution, not our endorsement or independent confirmation.

Revision record

Last checked Sep 29, 2026, 9:06 PM EDT. Talk Around Town: SpaceWit and its 'Space String' project are reported by a single, non-primary source chain originating from Chinese tech media. No regulatory filings, independent funding records, or Western press coverage have been found to verify the company's existence, technical plans, or 2027–2028 timeline. Treat all attributed capabilities and dates as unconfirmed company claims only.