Summary
- Jiangsu Dongyun is a real, currently active Zhenjiang infrastructure and network-security business, not merely a routing-registry label. Its public website is selling server rental, colocation and DDoS-protection services in 2026, while a June 2026 local report describes nearly 1,000 high-power racks, more than 10,000 servers and more than 6 Tbps of claimed egress at Jiangsu security nodes.
- Those large operating claims are not the same as independently verified usable capacity. Public material does not disclose utility capacity, live IT load, rack occupancy, generator runtime, cooling topology, commissioned accelerator inventory, sold capacity, reserved headroom or a tested restoration objective.
- APNIC records associate Jiangsu Dongyun with AS131483, nine IPv4 /22 allocations and one IPv6 /32. On 15 July 2026, however, the company's own autonomous system was continuously originating only 103.85.86.0/24 and 103.85.87.0/24 in the preceding two weeks, with DuoTu Cloud as the single visible adjacent upstream. Other Dongyun-registered blocks appeared through China Telecom, China Mobile, China Unicom, DuoTu Cloud or more-specific third-party origins.
- Multiple carrier origin options can improve reachability, yet they do not prove physically diverse building entrances, ducts, meet-me rooms, power feeds or operating teams. Buyers need a service-by-service dependency map and evidence from failover tests, not a count of carrier names or cities.
A control room full of traffic, and a much smaller edge
The most revealing public image of Jiangsu Dongyun is not a generic cloud graphic. It is a reporter standing with technical director Yang Wei in June 2026 while a monitoring screen showed traffic across domestic and overseas security nodes. The accompanying Jinshan Network report said the company had built nearly 1,000 high-power racks, installed more than 500 high-end hardware firewalls and core switches, and exceeded 10,000 servers. It put aggregate egress at the company's self-built Jiangsu high-defense nodes above 6 Tbps.
That scene makes the business tangible. There are people, equipment and active traffic. It also makes the central analytical problem sharper. A customer does not consume a headline number. A customer consumes a particular protected IP, server, rack, virtual instance, storage service, transit path or scrubbing policy at a particular time. The relevant capacity is what remains reachable and supportable when a power feed, cooling loop, carrier handoff, firewall cluster, routing session, billing account or human escalation path fails.
The public Internet shows only one part of that operating system. RIPEstat's current AS overview marks AS131483 as announced and identifies its holder as DYIDC, Jiangsu Dongyun Cloud computing. Its route collectors show a small self-originated edge. That does not disprove a large server estate. Much of Dongyun's address space is originated by carrier autonomous systems, and protected services can sit behind addresses that do not use AS131483 as their origin. But it does mean that the rack count and the AS131483 view cannot be treated as two measurements of the same thing.
This distinction matters because the company's product is resilience. A normal hosting provider can sell compute and bandwidth. A DDoS-mitigation provider sells the promise that abnormal traffic will be detected, diverted, absorbed and cleaned while legitimate users still reach the application. That promise crosses far more systems than a server specification suggests. Dongyun's public evidence supports the existence of an operating business and meaningful network assets. It does not yet close the gap between portfolio scale and service-specific survivability.
Establishing the company behind the English registry name
The directory name, "Jiangsu Dongyun Cloud computing co., LTD", is the English description held in the regional Internet registry. The operating company presents itself in Chinese as Jiangsu Dongyun Cloud Computing Co., Ltd., written 江苏冬云云计算股份有限公司. The different romanisation and the Chinese character 冬 rather than 东 make careless matching risky. There is also a separate Jiangsu Dongyun Interconnection Network Technology company using the "East iCloud" name. Its own website footer identifies a different legal company, so its products, data centres and certifications should not be attributed to the subject here.
The identity match for the subject rests on several converging records. APNIC's RDAP entry for AS131483 uses the DYIDC name, the English company description and a Runzhou District, Zhenjiang contact address at 36 Huangshan South Road, Building 8, Room 2208. The company's Chinese commercial record is tied to the same Zhenjiang address in public corporate information, while its current Dongyun Security website identifies Jiangsu Dongyun Cloud Computing Co., Ltd. as the owner of the service and publishes its ICP filing number. The Jiangsu provincial small-business service platform also lists Jiangsu Dongyun in Zhenjiang's Runzhou District, with registered capital of RMB10 million and cloud, data-centre, security and content-distribution activities.
The company was formed in 2016. A 2018 report from China Jiangsu Network described revenue of RMB10 million in its first year and signed projects worth RMB80 million by mid-2018. Those figures are historical and came through local reporting, so they are not a current financial statement. They do show that the business identity predates AS131483's present routing configuration and that its commercial focus has long combined hosting with network defence.
There is also fresh independent evidence of trading activity. The 2025 annual report of Tianjin Yingxuntong Technology, published in April 2026, lists Jiangsu Dongyun as its fourth-largest customer, with RMB7.51 million of sales and no related-party relationship. The annual report table does not say which specific system-integration or technical service Dongyun bought. It therefore supports current commercial activity, not a particular facility, capacity figure or supplier dependency.
What customers are actually buying
Dongyun's present offer is more concrete than its broad cloud label. The company website advertises dedicated-server rental, server and full-rack colocation, high-defense IP, high-defense CDN, a game-protection service and operational support. Its colocation page offers a 42U rack option with customised addresses and bandwidth, and a 1U server option described as using three-line BGP with 50 Mbps of dedicated bandwidth, redundant power and 24-hour support. Those are commercial descriptions, not audited service measurements, but they define the customer dependency more precisely than "cloud computing" does.
The DDoS products work by placing an intermediate defence surface between users and the customer's origin. Dongyun's CDN product page says traffic can be distributed across China Telecom, China Unicom and China Mobile paths, with the protected address concealing the origin. The game-protection page similarly describes distributed entry points and a maintained origin server. In both cases the service's value depends on more than raw scrubbing throughput. DNS or address steering must send users to the protection layer; the protection layer must recognise legitimate traffic; cleaned traffic must reach the origin; and the origin must accept it without exposing an easy bypass.
The company site promises 24-hour support and publishes a support telephone number. That is evidence of a reachable service channel, but not of response performance under stress. There is no public severity matrix, initial-response target, escalation tree, service-credit schedule, maintenance-notice policy or customer-visible incident archive. A buyer should distinguish "support is advertised" from "support recovery has been demonstrated" in exactly the same way that it distinguishes an advertised rack from a powered, occupied rack.
Licensing evidence supports the basic service category. Industry coverage of Ministry of Industry and Information Technology licence releases reported Dongyun among companies receiving an Internet resource collaboration, or cloud-service, authorisation in 2018, and among companies receiving CDN authority in 2019. A 2022 licence-release digest records a renewal for B1-20172723. The current detailed scope and expiry should still be checked in the ministry's live licence system before contracting; a historical award or renewal does not prove that every currently marketed location and product sits within the same permission.
Guantang moved from a project to an operating claim
Dongyun's physical story centres on the Guantang Supercomputing Center in Zhenjiang. In September 2022, China National Radio's Jiangsu service reported that a Dongyun subsidiary, Dongyun Data Technology Zhenjiang, had begun investing in the project in March 2021. At that stage the centre was a build intended to support network security, data disaster recovery and compute services. The article also quoted 1.5 Tbps of then-current DDoS cleaning ability across the platform, a service measure separate from the future centre's compute design.
By May 2024, a Zhenjiang report on advanced computing placed the centre at 3 Guanfu Road in Runzhou District and described it as erected but expected to enter operation in September. It gave an investment of about RMB240 million, 660 deployed racks, room for 10,000 compute servers, a 100-petaflop artificial-intelligence peak and a 400-petaflop high-performance-computing peak. Xinhua's Jiangsu channel repeated the same construction-stage account in July 2024. In August, the Jiangsu provincial Party news site said the project was in its final stage and expected to operate that year.
Those reports prove a substantial project and establish a dated status sequence. They do not, by themselves, prove that every designed rack was powered in September 2024, that all planned accelerator boards arrived, or that the peak calculations were benchmarked and customer-available. The wording matters: "can carry" is a design envelope; "deployed racks" can mean installed cabinets; "expected to operate" is a schedule; and a peak floating-point figure is not the same as sustained application performance after network, storage, power and cooling overhead.
The June 2026 on-site report materially advances the operating evidence. It describes live monitoring and says Dongyun had built nearly 1,000 high-power racks and exceeded 10,000 servers. Yet it speaks about Dongyun's estate and Jiangsu high-defense nodes, not an audited inventory for the Guantang building alone. It does not reconcile the earlier 660-rack plan with the later near-1,000-rack total. The company's own about page introduces another version: 12,000 square metres, 1,600 high-performance racks, 1,000 high-end compute servers, 5 Tbps of egress and 100 petaflops of artificial-intelligence performance. That copy also says the project started in 2022, while the earlier report says investment began in March 2021.
The correct conclusion is not to choose the largest number. The correct conclusion is that public descriptions use different scopes and maturity states. The 660 racks were a 2024 project figure. The 1,600 racks are a current company-site statement whose installed, powered and occupied status is not defined. The near-1,000 racks are a June 2026 company-wide construction claim conveyed by a local reporter. None discloses critical IT megawatts, utility reservation, live load, rack power density distribution or customer occupancy. For procurement, usable capacity remains unknown.
The registered address portfolio is much larger than AS131483
The most durable quantitative evidence is in the Internet registry, but it must be read correctly. An inverse APNIC lookup tied to Dongyun's incident-response contact identifies nine IPv4 /22 allocations: 43.248.76.0/22, 43.248.96.0/22, 43.248.100.0/22, 43.248.116.0/22, 43.248.128.0/22, 43.248.132.0/22, 43.248.136.0/22, 43.248.140.0/22 and 103.85.84.0/22. Together they contain 9,216 IPv4 addresses before exclusions for network and broadcast use in individual subnets. APNIC also records the IPv6 allocation 2403:4ec0::/32.
These are registration holdings, not a count of active servers and not proof that Dongyun physically owns the routers using every address. The APNIC record for 43.248.116.0/22, for example, names Dongyun and marks the allocation active. The record for 43.248.128.0/22 does the same. A hosting company can announce portable space through its own AS, authorise a carrier to originate it, divide it into more-specific customer or service routes, leave some space unannounced, or change the origin during mitigation. Registry control and live route origination answer different questions.
This distinction explains why a simple AS lookup severely understates Dongyun's address estate. RIPEstat's announced-prefix history for AS131483 shows 103.85.86.0/24 and 103.85.87.0/24 continuously present through the 1-15 July 2026 observation window. A handful of 43.226.x /24 routes appeared intermittently, but APNIC registers those ranges to other organisations; their temporary presence under AS131483 does not turn them into Dongyun property.
At the allocation level, RIPE's collectors see different carrier origins. The routing status for 43.248.76.0/22 shows the aggregate originated by AS23650, the China Telecom Jiangsu network. The 43.248.96.0/22 view shows AS56046, China Mobile Jiangsu, with less-specific coverage also visible through AS153911 and AS23650. The 43.248.116.0/22 view shows China Telecom's origin plus China Mobile more specifics. The 103.85.84.0/22 allocation is split among China Unicom, China Mobile, DuoTu Cloud and Dongyun's own AS at different prefix lengths.
This is meaningful operational evidence. Dongyun has arranged for address space to be reachable through several networks, which is consistent with multi-carrier hosting and traffic-cleaning products. It is not evidence that every service is simultaneously multi-homed, that all origins are active at one building, or that any two paths avoid a common trench, metro ring, carrier room or regional power event.
AS131483 is active, validly authorised and visibly narrow
AS131483 was registered in January 2017 and remains active in APNIC. On 15 July 2026, two high-visibility routes under Dongyun's own allocation were originated by it: 103.85.86.0/24 and 103.85.87.0/24. RIPE's RPKI validator marks both origins valid against a route-origin authorisation covering 103.85.84.0/22 with a maximum length of /24. That is a positive control. It allows networks enforcing route-origin validation to accept these announcements as authorised rather than invalid.
Authorisation is not availability. The RIPE looking-glass view for 103.85.86.0/24 repeatedly places AS153911 immediately before AS131483 in observed paths. The live bgp.tools view of AS131483 likewise identifies DuoTu Cloud as the one visible upstream and reports no originated IPv6 route. A single observable adjacent provider is a concentration signal for this particular AS edge, even though Dongyun's other carrier-originated address blocks create additional service paths elsewhere.
The route path beyond DuoTu includes other networks, and the carrier-originated blocks can provide distinct logical entrances. Yet customer resilience cannot be inferred by combining every route in the company portfolio. A server on 103.85.86.0/24 may depend on AS153911. A server on a China Telecom-originated aggregate may have a different dependency. A CDN-protected domain may switch among distributed addresses. A private-cloud or colocation customer may use its own address space and BGP. Each product needs its own path inventory.
There is also a warning about self-reported interconnection data. PeeringDB's page for ASN 131483 still labels the network "Joynet Shanghai Network Services" and lists a 1 Gbps CN-IX Shanghai connection, even though APNIC's authoritative record names Dongyun and was last changed in 2023. The PeeringDB record was last updated in 2022. It may be stale from an earlier user account or ASN association. It should not be used as proof that Dongyun currently owns or operates that exchange port. Registry hierarchy, update dates and conflicting identities matter.
Logical carrier diversity is not physical redundancy
Dongyun's website uses the familiar phrase "three-line BGP" and names China Telecom, China Unicom and China Mobile. The routing record gives that claim some substance at portfolio level: Dongyun-registered space is visible behind autonomous systems belonging to all three, as well as DuoTu Cloud. But BGP diversity is a logical property. Physical resilience requires different evidence.
Two carriers can enter a building through the same duct. Two services can terminate on the same meet-me-room patch panel. Separate routers can share one power distribution unit. Separate city nodes can depend on one central control system, one DNS provider, one policy database or one support team. A traffic-cleaning platform can announce an address from multiple origins yet still pass clean traffic to one origin server over a shared tunnel. None of those shared components appears in a public AS path.
For the Guantang site, no public document reviewed here states the number of fibre entrances, the route of each entrance, the carrier meet-me rooms, the distance between ducts, the cross-connect providers or the optical protection arrangement. No public source establishes dual utility substations, dual transformers, generator count, fuel runtime, cooling redundancy, battery autonomy or the maintenance regime. It is therefore not possible to rate physical path diversity or facility fault tolerance from public evidence.
The same limit applies to geography. The 2024 reports place the project at 3 Guanfu Road. Dongyun's current website gives 8 Guanfu Road. The difference between No. 3 and No. 8 remains unresolved. APNIC gives the older corporate contact at 36 Huangshan South Road, which is an office address rather than evidence of a server hall. These references establish Zhenjiang and the Guantang area, but not a verified parcel boundary or the separation between operational buildings.
IP geolocation labels such as Nanjing or Zhenjiang are not substitutes for facility records; they can reflect registry contacts, network topology or database inference rather than where a server stands.
Capacity has at least six different meanings here
Dongyun's public figures become more useful when assigned to distinct capacity classes.
First is building capacity: floor area and the number of rack positions that can physically fit. The company page states 12,000 square metres and 1,600 racks, but does not say whether that is design maximum, installed inventory or occupied inventory.
Second is powered capacity: the number of racks that can receive their contracted kilowatts at the same time under normal and degraded conditions. No public critical-load figure, utility connection or rack-density schedule is available. A "high-power rack" is not a unit unless its kW rating and operating state are specified.
Third is compute capacity: installed processors and accelerators, their precision and their sustained throughput. The 2024 project description gives 100 petaflops for artificial intelligence and 400 petaflops for high-performance computing. It does not identify FP64, FP32, FP16, BF16 or integer precision, the hardware mix, benchmark method, installed share or customer allocation. Those numbers cannot be combined or compared without a precision definition.
Fourth is server capacity. "Can carry 10,000 servers" in 2024 was an envelope. "More than 10,000 servers" in June 2026 sounds like an installed estate, but the report does not separate owned servers, customer colocation equipment, security appliances, powered-off inventory or machines spread across partner sites.
Fifth is network capacity. The 2026 report says Jiangsu security nodes have more than 6 Tbps of aggregate egress. Dongyun's website says 5 Tbps for Guantang, while other pages advertise 1.5 or 1.6 Tbps of cleaning. Egress, committed transit, burst capacity and DDoS cleaning throughput are not interchangeable. A defence platform may ingest attack traffic at one rate, deliver clean traffic at another and black-hole a target when policy or upstream limits are crossed.
Sixth is saleable capacity. This is what a new customer can actually reserve with the required power, ports, addresses, hardware and support. No public source provides rack occupancy, available kW, spare firewall slots, accelerator availability, bandwidth commits, sold capacity or reserved headroom. Saleable capacity is unknown.
This classification prevents a common error: converting an impressive design number into an availability promise. It also gives customers a practical request list. Ask for the capacity unit, facility scope, measurement date, installed state, powered state, occupied state, oversubscription policy, failure-state performance and evidence from the last test.
The failure path begins before an attack reaches a firewall
For a protected service, DNS and route control determine where traffic goes. A mistaken record, expired certificate, route withdrawal or failed policy update can make an otherwise healthy server unreachable. If Dongyun originates the prefix, its BGP session and adjacent provider matter. If a carrier originates the prefix, the carrier's acceptance policy, route objects and route-origin authorisation matter. If the product hides an origin behind a CDN or high-defense address, the mapping between public entry point and origin becomes a critical configuration asset.
Attack absorption then depends on upstream willingness and capacity. A claimed 6 Tbps site egress does not guarantee that every protected address can receive 6 Tbps of hostile traffic. Carrier contracts can set per-port or per-customer limits. Scrubbing appliances can have packet-per-second, connection-state, TLS or application-layer constraints below their raw bit-rate headline. A volumetric attack may also trigger remote black-holing before traffic reaches Dongyun's own equipment. Public material does not disclose these thresholds.
After cleaning, legitimate traffic still needs a path to the application. If the origin is in the same facility, power and cooling concentration become important. If it is remote, the clean-return tunnel or private line can fail. If multiple scrubbing nodes share a central policy service, control-plane failure can affect several cities at once. If the customer's allow-list recognises only one cleaning range, a failover node may be technically ready but operationally blocked.
Finally, people decide what happens when software is limited public evidence. Dongyun advertises 24-hour support, but there is no public evidence of staffing levels, bilingual coverage for overseas nodes, change approval, incident command, spare-parts logistics or escalation to carriers. Hardware stock matters because high-end firewall line cards, optics, power supplies and specialised accelerators cannot always be replaced from local retail inventory. The service survives only if the organisation can diagnose the failure, obtain access, make a safe change and communicate with customers.
A power or cooling event would cut across product labels
The Guantang centre is presented as a common base for computing, data storage, security and distribution. Shared infrastructure creates efficiency, but it can also create correlated failure. A utility loss can affect racks, firewall clusters, storage, control systems and support tooling together. A cooling incident can force load shedding even while generators and network links remain healthy. A maintenance mistake in a common electrical bus can defeat nominal device redundancy.
No public evidence gives Dongyun's utility topology or tested generator runtime. It is therefore impossible to state whether a full-site utility outage would leave every advertised service available, for how long, or at what reduced capacity. The phrase "redundant power" on a colocation product page may refer to dual server power connections, dual rack feeds or a broader facility design; the page does not define it. A customer should ask for the one-line electrical diagram, maintenance boundaries, last integrated systems test and the actual result under load.
Recovery also depends on data. A rented bare-metal server can be replaced only if an equivalent machine, network configuration and customer data are available. A virtual service can restart elsewhere only if images, keys, volumes and orchestration state have been replicated. A CDN can steer around a node, but the origin may still be the bottleneck. Dongyun's public pages do not disclose backup scope, replication distance, recovery-point objectives or recovery-time objectives. Those values are unknown, not zero and not automatically included.
For customers, this means resilience must be designed above the provider contract. Maintain application-level backups outside the failure domain. Keep infrastructure definitions, certificates and DNS authority under recoverable control. Test restoration to another provider or site. Document which addresses must be changed and how long propagation takes. A provider's hardware and support can reduce operational work, but it cannot remove the customer's responsibility for portability.
Route security is better evidenced than route diversity
The valid RPKI state for the two visible AS131483 /24s is a concrete strength. It reduces the chance that networks enforcing validation will accept an unauthorised origin for those exact routes. The broader 103.85.84.0/22 portfolio also carries authorisations for several carrier origins, consistent with flexible use of China Telecom, China Unicom, China Mobile, DuoTu Cloud and Dongyun itself.
That flexibility creates governance work. Each authorised origin expands the set of networks allowed to announce the space. Old authorisations should be removed when contracts or designs change. Internet routing registry entities should match the intended paths. More-specific announcements should be monitored because they can override aggregates. The intermittent appearance of third-party-registered 43.226.x prefixes under AS131483 is not necessarily an error; it could reflect customer service or mitigation. It does show why ownership, authorisation and current origin must be checked separately.
IPv6 is the clearest gap. APNIC assigns Dongyun 2403:4ec0::/32, but RIPE's collectors did not see that /32 or any more-specific IPv6 route on 15 July 2026, and AS131483 was not originating IPv6 in the public views reviewed. This does not prove that every Dongyun service lacks IPv6; carrier networks or private arrangements may carry it. It does mean that the company's own public AS does not provide visible dual-stack evidence comparable to its two IPv4 /24s.
For a customer that requires IPv6, the contract should name the actual prefix, origin AS, upstreams, protection method and failover behaviour. "IPv6 allocation exists" is not enough. The same standard should apply to IPv4: identify the service prefix and path rather than relying on company-wide counts.
Geographic reach is a service map, not an ownership map
The June 2026 report says Dongyun's security nodes cover Jiangsu, Shandong, Zhejiang, Guangxi and Hong Kong, with strategic deployments in Tokyo and Seoul. The 2022 report described storage and security nodes in eight Chinese cities and company or laboratory presences in Jiangsu, Guangxi, Beijing and Tokyo. These sources support a regional network of service locations and ambitions beyond mainland China.
They do not specify which sites Dongyun owns, leases, colocates in, operates through a subsidiary or accesses through a partner. The 2026 article explicitly says future international nodes may be co-built with partners, which reinforces the need to separate operational control from a city label. A node can be real and useful without being a Dongyun-owned building. Ownership affects access rights, maintenance control, insurance, equipment replacement and the consequences of a provider-contract dispute.
Data locality also follows the service path, not the company headquarters. A customer's origin may remain in mainland China while protective traffic is steered through Hong Kong, Tokyo or Seoul. Logs, packet metadata, attack telemetry, customer identifiers or cached content may be processed in a different jurisdiction from the origin. Conversely, a globally marketed service may keep all customer data inside one mainland facility. Dongyun's public pages do not provide a location-by-data-type schedule, subprocessor list, retention period or cross-border transfer design.
Buyers should request a location matrix for compute, storage, backups, logs, security telemetry, management access and support. They should also ask whether an overseas node is continuously active, available only during mitigation, or still a planned deployment. A dot on a service map cannot answer those questions.
Who is affected when Dongyun fails
The direct customer may be a game publisher, e-commerce operator, public-sector system, financial service, cloud platform or another hosting company. The 2026 report names large cloud companies among Dongyun's customers, although those names come through a company interview and are not accompanied by contracts. If Dongyun acts as an upstream security supplier to another cloud, end users may never see its name even while depending on its cleaning capacity.
A failure can therefore propagate in several ways. Loss of a protected address makes the customer's application disappear. A false positive blocks legitimate users while the service appears technically online. A false negative allows attack traffic to exhaust the origin. A route leak can send traffic to the wrong network or create unstable paths. A control-panel or support failure can prevent customers from changing policies during an incident. A billing or contract dispute can remove service without any equipment fault. A facility event can affect multiple products marketed under different labels.
The concentration is not necessarily at Guantang. A customer using a carrier-originated Dongyun block elsewhere may be independent of that building but dependent on the same support organisation or policy system. Another customer may have servers at Guantang but use its own carrier and addresses. The right unit of analysis is the specific service chain: legal counterparty, facility operator, rack, power path, network origin, upstreams, protection cluster, origin return path, control system and escalation contacts.
For downstream customers, transparency should continue one layer further. If a managed-service provider resells Dongyun protection, it should disclose the relevant failure domain and provide an exit route. Otherwise a buyer can unintentionally purchase two "independent" services that share Dongyun, the same carrier or the same Zhenjiang facility beneath different brands.
What a credible resilience package would contain
Dongyun has enough public evidence to justify detailed diligence. It is not a case where nothing can be verified. The company operates a current commercial site, appears in current reporting, holds substantial registered address resources, originates valid routes and is named in a 2025 counterparty disclosure. The next step is to turn broad scale into service-level proof.
For Guantang, that proof would include the legal owner and operator of the building; the relationship between Jiangsu Dongyun and the investing subsidiary; the facility's exact address; commissioning dates by hall; utility and generator capacity; cooling design; fire suppression; rack counts by installed, powered, occupied and available state; and independent certifications with their scope and validity dates. A photo tour is useful, but test records and maintenance boundaries are more important.
For network services, it would include a prefix-by-prefix table naming the origin AS, contracted upstreams, port capacity, physical entrances, route-origin authorisations, traffic-engineering policy and black-hole thresholds. Customer-specific diagrams should show the path from user to scrubbing node to origin. A failover report should record the date, injected failure, measured packet loss, convergence time, reduced capacity and any manual action.
For compute and storage, Dongyun should define the precision behind its petaflop figures, identify installed hardware, report available rather than theoretical capacity, and distinguish owned equipment from customer colocation. It should disclose oversubscription, data-protection options, replacement inventory, recovery objectives and the boundary between included resilience and customer responsibility.
For operations, buyers should receive support targets, escalation contacts, maintenance notice, incident communications, data-export procedures and termination assistance. A tested migration matters because the ability to leave is part of resilience. Without export time, bandwidth limits, media handling and configuration portability, backup capacity at another provider may not be usable when needed.
What remains unknown on 15 July 2026
The company's operating status is positive. The public evidence does not support downgrading Jiangsu Dongyun to a dormant registration. The website is active, the service catalogue is current, an on-site report is recent, the AS and multiple address blocks are routed, and a third-party annual report records 2025 purchases by the company.
The unknowns are nevertheless material. Public records do not establish the current ownership structure of every operating subsidiary, audited Dongyun revenue, facility title, utility supply, live IT megawatts, power usage effectiveness, generator runtime, cooling redundancy, rack occupancy, accelerator inventory, sold bandwidth, available bandwidth, support staffing, incident history, customer churn or insurance. The current licence scope needs direct confirmation. Physical fibre diversity is unknown. The operating role and exact asset scope of each domestic and overseas node are unknown.
The 2026 capacity numbers also require reconciliation. "Nearly 1,000 high-power racks" cannot automatically be combined with the website's 1,600-rack figure or the 2024 project's 660 racks. "More than 6 Tbps egress" cannot automatically be combined with 1.5 Tbps cleaning, 1.6 Tbps protection or 5 Tbps at Guantang. "More than 10,000 servers" does not reveal how many are powered, customer-owned or available for new workloads. These are different measures with different scopes.
Network evidence is strongest at registration and route-origin level. It is weaker at customer-service and physical-path level. The portfolio has multiple carrier origins, while AS131483 itself has one visible adjacent upstream. Both facts are true. Neither tells a customer whether its particular service survives a cut cable, failed router, carrier withdrawal or control-plane error.
The watchpoints are measurable
The most useful future signal would be a commissioning or facility document that states live power and hall status at Guantang. A utility connection notice, construction acceptance, fire acceptance, energy filing, operating certificate or audited facility inventory would move the rack story from reported scale toward verified usable capacity. Until then, changes in marketing copy should not be treated as commissioning evidence.
Routing can be monitored more directly. Watch whether AS131483 gains a second independent adjacent upstream, whether its two /24s remain stable, whether the IPv6 /32 begins to appear, whether more-specific origins change, and whether RPKI authorisations stay aligned with intended carriers. A second AS path would still require physical verification, but it would improve the public logical-diversity evidence.
Watch the distinction between owned and partner nodes as expansion continues. Dongyun's stated plan is to build several more large security-computing centres in southern, western and Southeast Asian markets while co-building international nodes. New city names should be accompanied by operator, facility, launch date, active capacity and service-prefix evidence. Announced construction is not available capacity, and a partner point of presence is not an owned data centre.
Finally, watch for customer-facing operational evidence: status history, incident reports, maintenance notices, service-level terms, export procedures and tested recovery results. These are less spectacular than a petaflop or terabit number, but they reveal whether the company can keep promises through failure.
The evidence points to a real operator, not a completed resilience case
Jiangsu Dongyun has more infrastructure substance than the thin AS131483 view initially suggests. Nine APNIC-registered IPv4 /22s, an IPv6 /32, multi-carrier origin arrangements, current services, a major Zhenjiang construction project and recent independent commercial references form a credible operating footprint. The June 2026 control-room visit is especially important because it moves the story beyond an old allocation and a proposed building.
But the evidence also shows why scale claims need boundaries. A near-1,000-rack estate does not tell a customer which racks are powered or available. More than 6 Tbps of egress does not define cleaning performance during a specific attack. Several carrier origins do not prove separate fibre. A valid route-origin authorisation does not create a second upstream. A node in another city does not reveal who owns the equipment or where data travels.
The decisive question for Dongyun is therefore not whether it has infrastructure. It does. The question is how much of that infrastructure remains usable when one component fails, and whether the answer can be demonstrated for the exact service a customer buys. On the public evidence available on 15 July 2026, the company can show scale, address control and active operations. It cannot yet show a complete, independently verifiable chain from utility supply and occupied rack to diverse carrier path, tested restoration and portable customer data. That is the gap procurement teams should measure rather than fill with assumptions.