China’s orbital AI ambitions gained a new private-sector proposal on September 25, when Oriental Starlink Spacetime Intelligence (Shandong) Technology Development Co., Ltd. and Diwei-2 Space Technology (Hangzhou) Co., Ltd. jointly announced the Space String computing constellation. The companies said construction had formally begun at an event held at the Hangzhou International Expo Center.
The announcement places satellite-based computing alongside China’s broader effort to build AI infrastructure beyond conventional terrestrial data centers. It remains a construction plan rather than a deployed network. The first G1 verification satellite is expected to launch in the fourth quarter of 2027, according to IT之家. Neither the primary report nor the independently published event coverage establishes that the full constellation is in orbit, commercially live, or available for customer use.
The proposal is nevertheless notable for its division of labor. Oriental Starlink is described as responsible for overall design and operations, while Diwei-2 is described as providing AI capability support and expanding international markets. Together, the companies are presenting Space String as infrastructure organized around computing services, rather than simply a collection of satellites gathering data.
Space String is a privately announced architecture with prospective capabilities. EastFrontier’s earlier report on Shanghai’s Mingzhu plan covered a separate orbital-compute proposal; neither report establishes Space String as an operating network.
A two-layer architecture separates orbital data work from heavier computing
IT之家 reported that Space String is planned around two layers. The business layer would deploy more than 720 data satellites, also called inference satellites, for data acquisition and business tasks. A separate computing layer would deploy more than 360 computing satellites, also called training satellites, in layered dawn and dusk orbits.
Those figures describe intended deployments, not satellites already installed or launched. The plan does not establish a firm delivery timetable for either layer, and the combined figure should not be treated as a confirmed in-orbit fleet. The companies’ announcement instead outlines an architecture that could distribute tasks between satellites designed to collect or process data and nodes intended for more demanding computing work.
The two layers are planned to connect through inter-satellite laser links. Under the proposed design, business satellites would access resources in the computing layer and coordinate with other nodes and ground systems. That arrangement could allow the network to treat orbital computing as a shared pool rather than requiring every data-gathering satellite to carry the same processing capability.
The wording is important. The sources describe a system intended to gradually form schedulable, user-callable space-computing capability. They do not show that the laser-linked network is operational or that users can call on it today. The practical test will come later, when the companies move from architecture and verification hardware toward any eventual networked service.
The public announcement establishes the companies’ planned architecture and roles rather than operating results. It does not provide evidence that Space String has completed a performance test or reached commercial service.
G1, G2 and G3 stages put the first test in late 2027
Space String is planned to progress through three generations: G1 verification satellites, G2 standard satellites, and G3 flagship satellites. IT之家 reported that the G1 satellite is expected to launch in the fourth quarter of 2027. That date is a projection, not evidence that a launch has been completed or that a launch-provider commitment has been secured.
The staged approach gives the companies a sequence for testing the concept before any larger deployment. G1 is positioned as the verification phase, followed by standard satellites and then flagship satellites. The fact card does not establish how many satellites will be built in each generation, so the announced sequence should not be read as a detailed production schedule.
凤凰网’s event coverage adds design-stage estimates for the generations’ computing payloads. Peak power is listed at 11.5 kilowatts for G1, 48 kilowatts for G2, and 140 kilowatts for G3. The same report gives approximate FP8 design peak computing ranges of 40 to 60 PFLOPS, 160 to 310 PFLOPS, and 460 to 1,000 PFLOPS, respectively.
These numbers are reference-configured design estimates, not measured in-orbit performance. They show how the proposed generations are meant to scale, but they do not demonstrate achieved throughput, sustained operation, or efficiency under space conditions. For investors and infrastructure companies, the distinction is central: the announcement provides an engineering direction, not operating metrics or revenue evidence.
A thousand-satellite target raises execution and commercialization questions
The second source describes Space String as a thousand-satellite blueprint, with 2030 and 2035 identified as key target nodes for reaching thousand-satellite networking. Those dates are planning targets attributed to company or event representatives. They are not a confirmed deployment schedule, and the announcement does not establish financing, customers, regulatory approvals, or revenue projections.
That makes the project relevant to China’s startup and infrastructure landscape without making it a live business. A large orbital AI network would require progress across satellite manufacturing, launch access, optical communications, onboard computing, ground coordination, and customer demand. The fact card confirms none of those commercial commitments for Oriental Starlink or Diwei-2.
The companies’ stated roles suggest an attempt to pair an overall system operator with a partner focused on AI capability and international markets. Whether that division becomes a working commercial model will depend on results from the planned verification stage. The G1 milestone should therefore attract more attention than the headline satellite count, because it is the earliest stated point at which the concept could begin to move from announcement toward demonstrable hardware.
No financing for Space String is verified in the checked sources. EastFrontier’s separate report on satellite-compute financing provides background on a different project, while Space String remains an announced construction effort with future milestones.
For now, Space String should be read as a private blueprint for China’s orbital AI push, not as a completed constellation. The September 25 event established the project and its proposed architecture. The next hard evidence will be whether G1 reaches launch in the expected fourth quarter of 2027, followed by proof that the planned layers, laser links, and computing workloads can operate together. Until then, the project’s importance lies in the infrastructure model it is putting forward and the execution questions it has made visible.