AIR Air Internet Service Co.,Ltd. makes a local connectivity bill depend on upstream routes and field repair

Summary

• AIR Air Internet Service Co.,Ltd. is the English corporate name of Tokyo-based Airnet. The AIRnet service began inside Adtex in April 1996; the present company acquired the internet business in 2005. Current corporate records place its head office in Shinagawa, while its public infrastructure footprint includes AS7503, a managed-hosting site described as the Tokyo No. 2 Data Center in Koto, and an Osaka disaster-recovery option.
• AS7503 is an active dual-stack network. On 10 July 2026, RIPEstat saw three IPv4 and two IPv6 prefixes, equivalent to 16,384 IPv4 addresses and 65,536 IPv6 /48s, from every full-feed RIS peer in its visibility set. Public routing observations identify IIJ, KDDI and ARTERIA as upstreams, while PeeringDB records a 10 Gbps JPNAP Tokyo connection. That is strong logical-diversity evidence, but it does not establish physically independent fibre entrances or carrier paths.
• Airnet's retail and business fixed-IP services depend on access lines supplied by NTT East or NTT West. The customer therefore buys one experience across separate responsibility domains: premises power and equipment, the NTT access and regional IP network, Airnet's access point and backbone, exchange or transit links, and the destination service. A cut or power failure before traffic reaches Airnet can defeat an otherwise healthy AS7503.
• Airnet publishes substantial facility claims for its Tokyo No. 2 Data Center, including two incoming power systems, UPS, on-site generation, redundant cooling and inter-data-centre connectivity. It also publishes maintenance examples in which another backbone line carried traffic automatically. Missing details include fuel autonomy, load-test results, route maps, carrier entrance diversity, spare-router stock, measured restoration times and the staffing split behind round-the-clock operations.
• The best current assessment is a Medium network evidence grade with strong logical-routing support: the operator, routes, exchange port, service contracts, maintenance behavior and facility controls are all visible. Resilience remains conditional because installed interfaces are not the same as usable peak capacity, and logical alternatives can still share ducts, buildings, power or the same limited group of people.

The bill bundles several operators into one promise

A fixed-IP invoice makes connectivity look like one product. It names a course, a number of addresses and a monthly fee. To the customer, that is sensible: the point of buying an internet service is to avoid managing every router, fibre handoff and routing agreement along the path. Yet the simplicity of the bill can obscure who must act when a connection fails. Airnet's case is especially useful because the company publishes enough detail to trace the boundaries, but not enough to assume that every component is independently protected.

The current AIRnet service terms define internet access as a connection between a router at an Airnet access point and a router assigned to a member, using the public telephone network, the regional IP networks managed by NTT East or NTT West, or another telecommunications line. The terms separately define an access line as a circuit that the member leases from a line operator. That language divides the physical chain before the first packet leaves the customer's premises. Airnet supplies the internet service and may supply or manage network equipment; another carrier can own and repair the access facility.

Airnet's fixed-IP product page makes the commercial arrangement equally clear. A user needs both an NTT East or NTT West FLET'S contract and a compatible AIRnet course. The page offers one, eight or sixteen fixed addresses across family, apartment, priority and business access types. The latest terms also list IPoE fixed-IP courses over FLET'S Hikari Next and FLET'S Cross. This is nationwide product compatibility within Japan, not evidence that Airnet owns nationwide poles, ducts or fibre. It is an access-aggregation business in which Airnet controls addressing, authentication, routing and support around a carrier-built last mile.

That distinction defines the first failure path. A powered Airnet backbone in Tokyo cannot restore a severed drop cable outside a customer's office. A healthy NTT optical line cannot reach the internet if Airnet's access router or upstream handoff is unavailable. A customer router with failed power can make both networks appear down. The operational value on the bill is therefore partly technical capacity and partly coordination: recognizing which domain failed, dispatching the correct organization, keeping the customer informed, and moving traffic where an alternative exists.

The terms do not promise that all of those dependencies are under one owner. They explicitly allow interruption when service equipment requires maintenance or construction, or when a line operator makes the service circuit unavailable. They also permit traffic restrictions when heavy demand is expected, when quality falls below Airnet's standard, or when sustained high-volume uses affect fairness. These clauses do not show poor service. They show that a connectivity product has finite capacity and dependencies outside the retail provider's direct control.

For a buyer, the practical question is not simply whether AIRnet is up. It is whether the complete path from the premises to the application has an alternative at each consequential point. A second upstream does not replace a second access line. A second access line that terminates on the same customer router does not replace backup power. Two logical circuits routed through one street duct do not create physical diversity. The invoice hides those distinctions; resilience depends on making them explicit.

A 1996 service sits inside a company formed later

The operator has a longer network history than the present corporate shell. Airnet's company history says Adtex began the AIR Internet Service as an internal business in April 1996. It joined JPNIC that December and began domain-name and IP-address allocation work. A Tokyo network operations centre opened in the KDDI Otemachi building in June 1997, the service connected to NSPIXP2 in 1998, and an AIR Tokyo data centre opened in KDDI's Otemachi building in July 2001.

The current corporation was established in December 2002 under a different name to sell security systems. In August 2005 it acquired Adtex's internet access, data-centre and application-service activities, changed its name, and began operating as Airnet in September. Aeria became its parent in November 2005. This sequence matters because a statement such as "founded in 1996" can blur the difference between service continuity and corporate incorporation. The service lineage starts in 1996; the present legal company dates from 2002 and took over the business in 2005.

A 2024 JPNIC member-company interview adds operational texture. Airnet director Masahiro Tanaka said the original service ran from a corner of a factory in Fujisawa and reached the internet through IIJ. Statutory power inspections at the factory helped motivate the move to a data centre in Otemachi. That is a small but revealing infrastructure lesson: the service did not become more resilient merely by having servers and an upstream. Its location and power regime had to change as availability expectations rose.

The same interview reported 33 employees as of July 2024, a capital base of JPY100 million, and a Shinagawa head office. It said corporate customers dominated both customer count and revenue, and that managed hosting accounted for about 40 percent of sales at the time. Airnet described a business built around designing, owning and operating dedicated customer systems rather than simply renting generic server space. Long customer relationships and customized configurations are central to that proposition.

Airnet's current company profile lists business cloud, managed hosting and ISP service as its principal activities. As of June 2026 it named Takashi Yoshimura as president, kept the JPY100 million capital figure, and listed Aeria and Newtech as major shareholders. Aeria's 2025 securities report identifies Airnet as a consolidated IT-services subsidiary in Shinagawa and reports an 89.4 percent voting interest. The same filing records 78 employees across Aeria's IT-services segment, not at Airnet alone, so it cannot be used as a current Airnet headcount.

The ownership evidence supports continuity but not unlimited financial backing. Aeria's 2025 full-year presentation says Airnet continued to earn stable revenue from data services, while the wider IT-services segment declined because of another subsidiary's payment and affiliate-advertising activities. There is no separate public Airnet income statement in that presentation. It is reasonable to say Airnet is an established operating subsidiary; it would be too strong to infer its network investment budget, spare inventory or disaster-recovery expenditure from group segment totals.

This history also explains why Airnet does not fit neatly into a rural last-mile category. It began as an ISP, still sells access, and operates a public autonomous system, but much of its current value sits in business hosting, enterprise email, managed infrastructure and cloud interconnection. Its physical centre is metropolitan data-centre infrastructure, while the customer edge often rides on NTT access. The relevant labour is therefore not only cable repair.

It includes network engineering, carrier escalation, server replacement, security response, customer-specific configuration and the judgement required to isolate a fault across organizational boundaries.

The visible network edge is mature and dual-stack

AS7503 gives Airnet a public routing identity that can be tested independently of product copy. JPNIC registration data reproduced by bgp.tools identifies the holder as Air Internet Service Co.,Ltd., the AS name as AIR, and the allocation date as 2 April 1997. That date fits the company's account of opening its Tokyo network operations centre in 1997. It also makes AS7503 an established routing resource rather than a recently acquired label.

On 10 July 2026, RIPEstat's AS overview marked the network announced. Its routing-status view reported three IPv4 prefixes covering 16,384 addresses and two IPv6 prefixes equivalent to 65,536 /48 networks. Every full-feed RIS peer in the relevant visibility sets saw the routes: 327 of 327 for IPv4 and 321 of 321 for IPv6. The first route in the historical view, 210.166.64.0/19, was observed in August 2000.

The announced-prefix view listed 210.159.64.0/19, 210.166.64.0/19, 210.166.92.0/22, 2402:3800::/32 and 2402:3800:dc03::/48. The first two /19s each represent 8,192 IPv4 addresses; the /22 represents 1,024, but it sits within the 210.166.64.0/19 address span and is a more-specific route rather than an additional 1,024 unique addresses. RIPEstat therefore correctly reports 16,384 unique IPv4 addresses, not 17,408.

This arithmetic is more than tidiness. Prefix counts are often mistaken for capacity or customer scale. An address block can host access customers, servers, mail systems, network equipment and infrastructure functions. Network address translation can place many users behind one public address, while dedicated hosting can assign several addresses to one customer. The IPv6 /32 is extremely large when expressed as /48s because IPv6 allocation conventions reserve generous address space. None of these numbers tells us subscriber count, peak throughput or available headroom.

The routes are also covered by origin authorization. For example, RIPEstat's RPKI validation reports a valid route-origin authorization for AS7503 and 210.159.64.0/19; bgp.tools marks all five listed origins as valid. This reduces the risk that a correctly validating network will accept an unauthorized origin for those prefixes. It does not authenticate every hop, prevent a route leak, guarantee traffic delivery or prove that the physical network is redundant.

Cloudflare Radar's AS7503 page observes traffic and identifies the network in Japan. Its estimated user population was about 75 when reviewed. That estimate should not be read as a subscription count. Airnet hosts business systems and email, and Cloudflare's population measure is derived from observable end-user activity rather than customer contracts. A low estimate is consistent with a network whose commercial weight is concentrated in enterprise and hosted services, but it cannot establish revenue concentration or actual connected endpoints.

Taken together, the registry, route visibility, IPv6 presence and observed traffic support a live, mature network edge. They settle the operating-status question more strongly than a company listing alone. They do not reveal where every border router is installed, which prefixes serve access customers rather than hosted systems, how much of the route capacity is used, or how quickly engineers can recover a failed chassis.

Three visible upstreams reduce one kind of concentration

Public BGP observations show more than a single path out. bgp.tools classifies IIJ AS2497, KDDI AS2516 and ARTERIA Networks AS2519 as Airnet's upstreams for both IPv4 and IPv6. RIPEstat's neighbour view also sees those three prominently to the left of AS7503 in observed paths. Airnet's own 2025 maintenance notice named an NTT Communications line and said another backbone line would take over automatically during work, adding direct operating evidence that the company has used alternate backbone carriage.

These are materially better signals than a generic claim that a network is "redundant." Three upstream organizations can reduce dependence on one carrier's commercial network or routing policy. BGP can withdraw an unavailable path and select another. The 2025 backbone-router maintenance notice said traffic might temporarily take a different path but customer service should not be affected. A May 2026 router-maintenance notice similarly anticipated changed paths without a communications stop.

Airnet also has a public exchange connection. PeeringDB lists AS7503 at JPNAP Tokyo with a 10 Gbps port, IPv4 and IPv6 addresses, route-server participation and an open peering policy. JPNAP's own customer list independently includes Air Internet Service Co., Ltd. and AS7503. An exchange port lets Airnet exchange traffic with participating networks without sending every packet through paid transit, subject to peering arrangements and route-server policy.

The port is installed capacity at one interconnection service, not a measure of total internet capacity. PeeringDB leaves Airnet's traffic level, traffic ratio and geographic scope undisclosed. It lists one exchange and no interconnection facility. The missing facility record is important: JPNAP offers service at multiple Tokyo points of presence, but the public Airnet entry does not disclose which building hosts its port. Nor does the 10 Gbps figure show average use, burst headroom, packet loss during busy periods or the capacity of each upstream transit contract.

The distinction between logical and physical diversity is decisive. IIJ, KDDI and ARTERIA are different networks, yet their circuits could enter an Airnet site through the same meet-me room, cross the same bridge, share a metro duct or depend on the same building power. BGP would see separate autonomous systems even if a backhoe or building incident could interrupt all of them together. Conversely, physically diverse circuits from one carrier might provide useful protection that public route observations cannot distinguish.

Airnet's managed-hosting page says its data centres are redundantly interconnected and peer with major domestic ISPs through an internet exchange. That supports a deliberate multi-path design, but the public material does not provide route diagrams, carrier entrance points or shared-risk groups. The right conclusion is therefore specific: AS7503 has credible logical upstream diversity and a current exchange port. Physical route independence remains unverified.

The access network belongs to a different resilience domain

The public AS can remain fully reachable while an individual customer is offline. Airnet's fixed-IP service is built around NTT East and NTT West FLET'S access products. A household or office line reaches an NTT regional IP network before it reaches Airnet's access point. That architecture moves much of the trench, pole, local fibre, optical termination and field-repair burden to the access carrier.

This is not merely an inference from the product name. The AIRnet terms say the member leases the access line from a telecommunications operator and describe the access point as the junction between that line and Airnet's service circuit. The fixed-IP page says the NTT access contract and AIRnet course are both required. Current application forms list compatible FLET'S Hikari Next, Priority, Business and Cross products. The division is contractual as well as technical.

The physical chain at a typical site can therefore be read in order. Customer equipment needs local electricity and a working optical network terminal or router. A drop cable reaches building or street distribution. NTT's access network carries the session through its regional facilities. Airnet terminates or authenticates the service at an access point, assigns the relevant address service, and carries traffic into AS7503. The packet then leaves through a peer or upstream and continues across other networks. Each step can be available while another is not.

For customers, this creates a support problem before it creates a routing problem. A failed browser does not identify whether the cause is Wi-Fi, premises power, customer equipment, the access line, the regional IP network, Airnet authentication, DNS, a backbone path or the destination. The ISP's local value lies in diagnosing that sequence and opening the correct escalation. A small but skilled support and network team can outperform a larger one if it has good telemetry, carrier contacts and tested procedures; a thin shift or unclear boundary can extend restoration even when replacement fibre crews exist elsewhere.

Airnet's standard retail access does not by itself demonstrate a second local route. A business can buy a high-grade FLET'S product or a standby VPN option, but product names do not establish that two services leave the building through different conduits or terminate on independent equipment. Proper access resilience requires address-level detail: two line providers where feasible, separately routed building entrances, independent optical termination, redundant customer routers, diverse power and a failover method that has been tested under load.

This is where the planned title's reference to field repair becomes concrete. Airnet can operate AS7503 and coordinate incidents, but NTT or another contracted line operator may dispatch the crew that repairs a broken access fibre. Data-centre operators may maintain generators and cooling. An upstream carrier may replace a failed metro optic. Airnet's own engineers may replace border routers, servers or managed customer equipment. The repair chain is distributed, and the time to restore service is set by the slowest unresolved dependency.

Public information does not give Airnet-specific mean time to repair for access circuits, carrier escalation targets, after-hours staffing, spare customer routers or the number of field technicians available through partners. The company should not be described as weak simply because those commercial details are private. But buyers should not treat a 24-hour monitoring statement as proof that a person with the right spare can reach every failed asset within a fixed time.

Tokyo facility controls are concrete, but autonomy is not quantified

Airnet publishes unusually specific facility claims for its managed-hosting environment. The Tokyo No. 2 Data Center operating page places the facility in Koto, Tokyo, while withholding the exact address for security. It describes a structure designed for Japan Meteorological Agency seismic intensity 7 class shaking, ground improvement and liquefaction countermeasures, two incoming power systems using a closed-loop arrangement, UPS, on-site generation, redundant large water-cooled air-conditioning equipment, inert-gas fire suppression, sprinklers and controlled access.

Those details identify real failure domains. Two utility feeds can reduce exposure to one incoming circuit. UPS can bridge the interval between a utility disturbance and generator pickup. On-site generation can extend service through an outage. Redundant cooling can protect servers when one chiller or pump fails. Seismic and liquefaction measures address hazards relevant to Koto's reclaimed and low-lying districts. Physical access controls reduce the risk that an unauthorized person reaches customer equipment.

The claims still leave decisive quantities unknown. The page does not disclose generator fuel autonomy at design load, refuelling contracts, black-start performance, battery age, UPS topology, simultaneous-maintenance tolerances, cooling redundancy level or the most recent integrated load test. "Two systems" can describe different utility routes or two feeds that converge upstream. A generator can exist without enough fuel for a prolonged regional outage. A water-cooled plant can be redundant at the chiller level while sharing another component.

The same caution applies to geography. Airnet's history records an older data centre in KDDI's Otemachi building, while the current managed-service page emphasizes Tokyo No. 2 in Koto. Public materials do not provide a complete current facility inventory or say which AS7503 routers and customer services sit at each site. PeeringDB does not disclose the building behind the JPNAP port. A customer therefore cannot infer that the exchange port, upstream handoffs and hosted systems occupy separate facilities merely because several Tokyo locations appear in the company's history.

Airnet does offer an Osaka disaster-recovery option. A 2023 contract revision added a DR environment described as rented virtual servers for standby or backup systems in an Osaka backup data centre. Colt's Airnet customer case says Airnet used Colt On Demand while reconsidering the network design for its own DR site, seeking flexible bandwidth because normal traffic did not flow there. This is meaningful evidence of geographic recovery planning.

It is not evidence that every customer service is synchronously replicated to Osaka or can fail over without manual work. A standby virtual server may have different recovery-point and recovery-time objectives depending on data replication, licensing, DNS, routing, application state and customer contract. Colt connectivity can shorten provisioning and provide flexible capacity without proving that the primary and recovery paths share no carrier facilities. The relevant customer question is which service components are duplicated, how current the copy is, and when the last end-to-end recovery exercise succeeded.

Airnet's public description is strongest at the component level: power feeds, UPS, generator, cooling, redundant server designs, backbone routing and a DR option. It is weaker at the system level: common-mode dependencies, test results and measured recovery. That is enough to treat the facility as a serious operating environment, but not enough to assign a guaranteed autonomy or restoration time.

Installed capacity is not the capacity a customer can use

Several public numbers can be mistaken for one another. AS7503 has 16,384 unique IPv4 addresses. Its JPNAP port is listed at 10 Gbps. Airnet advertises a 100 Mbps guaranteed-bandwidth option for managed systems. FLET'S access products have their own nominal line rates. None of those figures answers the same question.

The 10 Gbps exchange port is a physical interface into JPNAP Tokyo. It applies to traffic exchanged through that port, not necessarily to transit traffic through IIJ, KDDI or ARTERIA. It does not show whether Airnet has one or several links in a bundle, whether the port is close to saturation, or how much route-server traffic it carries. The total available internet capacity could be larger or smaller in particular directions depending on private contracts and traffic engineering.

The 100 Mbps service described on the managed-hosting network page is narrower than a blanket internet guarantee. Airnet says the assured bandwidth applies from a customer system built in the data centre to Airnet's own backbone, and requires a dedicated firewall contract. Charging uses a monthly average. That can protect the customer's access to Airnet's backbone while leaving onward performance dependent on backbone congestion, peering, transit and the remote destination.

Address holdings are not bandwidth. A /19 can hold customer servers or access assignments without carrying a predictable volume of traffic. IPv6 space is even less useful as a load indicator. Cloudflare's observed population estimate does not repair this gap because hosted applications may serve users outside AS7503 and enterprise mail flows do not map cleanly to resident users. A business-heavy network can matter operationally while appearing small in consumer-population measurements.

Airnet does publish examples of controlled maintenance without customer impact, which is useful evidence that some spare path capacity exists. But a successful maintenance window occurs at a chosen time and may not reproduce peak traffic, a simultaneous carrier failure or a disaster in which several systems fail together. The strongest capacity evidence would include time-series utilization by edge, packet loss and latency under failover, transit commit and burst limits, JPNAP port headroom, and test results showing that backup paths can carry the production load.

This is also an economic question. Capacity held idle for rare failures costs money. Colt's case explains that Airnet wanted flexible DR bandwidth because the recovery site did not carry normal traffic. The rational design may be to buy capacity on demand, maintain standby contracts or accept a defined recovery interval rather than duplicate everything at full rate. That can be a sound business choice if customers understand the objective. It should not be described as instantaneous active-active resilience without supporting detail.

For a customer comparing plans, the important number is therefore not the largest printed interface speed. It is usable capacity across the complete failure path. How much traffic can the alternate route carry? Does failover preserve the customer's fixed address? Will the application continue if DNS, authentication or storage is unavailable? Does a backup circuit rely on the same premises router and power outlet? Those questions convert installed components into service continuity.

Repair labour is part of the network architecture

Airnet repeatedly describes 24-hour, 365-day monitoring and incident response for managed hosting and enterprise mail. Its managed dedicated-server page says it handles design, construction, monitoring and failure response as one service. The ALL in One Mail Pro page similarly describes round-the-clock operation in a domestic data centre, with an optional secondary mail server to preserve mail flow after a primary hardware failure.

Monitoring is essential, but it is not the same as repair capacity. A monitoring system can detect a failed server in seconds. Restoration may still require a remote command, a technician at a rack, a replacement component, a carrier dispatch, a vendor escalation or a customer decision. The speed of each step depends on people, spares, access permission and the clarity of responsibility.

The 2024 JPNIC interview gives rare insight into Airnet's staffing constraint. Tanaka said the company looked for people already involved in operations and monitoring who wanted to broaden their work, because the role required curiosity about preventing incidents rather than merely following steps. He also said recruiting infrastructure engineers and technically knowledgeable sales staff was difficult. New hires were expected to work toward an internet-competency qualification, learn under senior staff and gain experience on company systems before taking on customer environments.

That account supports the Local support labour topic more directly than a generic claim that all ISPs need technicians. Airnet's product is customized operation. Engineers must know customer-specific systems, the Airnet backbone, mail behavior, cloud interconnects and carrier escalation. The company said long relationships help it understand where risk sits in each customer's system. That knowledge can accelerate diagnosis, but it also creates concentration risk if too few people hold it.

The reported 33 employees in July 2024 cannot be divided into network engineers, server operators, developers, support staff, sales and management from public information. Nor does the current recruitment page, which says the company is not hiring, establish whether staffing is abundant or constrained. A 24/7 service can be supported by on-call rotations, data-centre personnel, contractors and carrier operations centres as well as direct employees. Headcount alone cannot measure shift depth.

Maintenance notices show that Airnet performs planned router work and server upgrades with redundancy in mind. A 2025 DNS maintenance notice described work on one side of a redundant DNS arrangement with no expected service impact. Another 2025 notice said mail-delivery servers would be upgraded sequentially across a multi-server configuration. These are useful signs of operational discipline. They do not disclose whether the same coverage holds during simultaneous incidents or whether a replacement router, optic, firewall and server are stocked at every relevant site.

The labour chain extends beyond Airnet. NTT crews repair access plant. Data-centre teams maintain utility interfaces, generators and cooling. Upstream carriers maintain transport and border equipment. Hardware vendors may replace failed parts. Airnet's support team must identify and coordinate the right response. A resilient service therefore needs both technical alternatives and tested human handoffs. A second circuit with no after-hours escalation can be less useful than expected; a skilled engineer with no spare optic can still be unable to restore a link.

The most informative missing measures are operational: median and high-percentile repair times by failure type, after-hours acknowledgement, carrier dispatch performance, spare holdings, on-call depth, cross-training, and the frequency of failover and recovery exercises. Until those are available, Airnet's long operating history and published maintenance behavior support confidence in competence, while staffing depth and restoration speed remain unquantified.

Failure does not have one shape

An access cut is the simplest local failure. Construction damage, a building fibre break or failed optical termination can isolate one customer while AS7503 remains healthy. If the customer has only one FLET'S circuit, the restoration clock belongs largely to the access carrier's diagnosis and field dispatch. An Airnet support engineer can verify that the session is absent and escalate it, but cannot route around a severed physical drop unless a second access path already exists.

A premises power failure is even closer to the customer. Fibre can remain lit upstream while the optical network terminal, router, firewall, Wi-Fi access point or local switch is dark. A laptop battery may create the impression that the network alone failed. Resilience requires UPS capacity at the customer site and a decision about which equipment receives it. Airnet's data-centre generator cannot protect an office router connected to an unprotected socket.

A regional access outage can affect many customers before traffic reaches Airnet. Because fixed-IP service depends on NTT regional IP networks, an NTT control-plane or aggregation fault can interrupt sessions across a wider area. A second AIRnet backbone path would not solve that failure if both customer circuits use the same regional access system. A genuinely diverse design may need another access technology or carrier as well as automatic failover at the premises.

An Airnet edge or backbone-router failure occurs later in the chain. Here the company's BGP design and alternate routes matter. Published maintenance records indicate that traffic can move to another backbone line, and the network is visible through three upstreams. The residual risks are shared chassis, software defects, configuration errors, common power and common fibre entrances. Planned maintenance demonstrates one controlled withdrawal, not every unplanned failure combination.

An upstream loss is the best-supported failover case. If one transit relationship or circuit becomes unavailable, BGP should select remaining routes. The result may still be slower or more expensive. Paths can lengthen, latency can rise, and a backup contract may have less capacity. Airnet's 2025 notice explicitly warned that traffic could take an unusual route during maintenance. Availability can be preserved while performance changes.

An exchange-port failure removes one route to peers and content networks but should not remove transit if the upstreams remain available. The reverse is also true: exchange peering cannot replace all transit because not every destination is reachable through bilateral peers or a route server. Airnet's 10 Gbps JPNAP port improves path choice and economics, but resilience depends on the transit links around it.

A data-centre power or cooling incident affects hosted services rather than only access users. The Tokyo No. 2 site claims dual feeds, UPS, generation and redundant cooling. A prolonged outage tests fuel, maintenance and refuelling rather than the mere presence of a generator. A facility-wide network incident tests inter-data-centre links and the recovery design. Customers with an Osaka standby environment may have an option, but only if data, application state, DNS and network access are ready to move.

Congestion is a partial failure. Packets still flow, but an application becomes unusable. The 100 Mbps data-centre access guarantee applies to a defined segment, while public records do not show upstream utilization. A viral campaign, attack or carrier reroute can shift traffic suddenly. Capacity planning must consider the load on alternate paths, not only normal-state averages.

DNS, mail and authentication failures can look like general internet outages. Airnet operates authoritative and service DNS hosts, enterprise email and hosted applications. Redundant DNS maintenance and secondary mail options reduce some risks, but a bad configuration can propagate across redundant systems. The public incident page reports only the most recent two weeks and showed no current incident when reviewed. That is a useful live status, not a long-term availability record.

Finally, a labour shortage can turn any technical fault into a long outage. If the correct engineer, carrier contact, access credential or spare is unavailable, redundant hardware may wait unused. Airnet's customized systems make customer knowledge valuable. Cross-training and documented escalation are therefore infrastructure controls, even though they do not appear on a route map.

The economics reward coordination, not ownership of every asset

Airnet's structure illustrates how a specialist provider can sell resilient service without owning the entire physical path. NTT finances and maintains broad access infrastructure. Internet exchanges aggregate opportunities to peer. Transit carriers sell reachability. Data-centre operators supply hardened buildings and utility systems. Airnet combines those inputs with addresses, routing, managed equipment, hosting and support.

This arrangement reduces the capital required to replicate a nationwide access network. It also creates recurring supplier expenses and boundary risk. The retail price must cover NTT-facing access coordination, upstream transit, exchange ports, rack and power, equipment depreciation, software, security controls, customer support and skilled labour. A fixed-IP plan can look expensive compared with mass-market broadband because the paid value includes stable addressing and business support as well as raw bits.

Managed hosting pushes the economics further toward labour and customer retention. Airnet told JPNIC that it buys and capitalizes hardware for dedicated customer systems, designs configurations around each customer's needs and operates them over long relationships. Older, larger customers can carry higher revenue per account, but customized estates demand knowledge and make migrations expensive. Reliability becomes an economic asset because failures threaten both immediate service and long-term trust.

Peering can reduce transit expense for suitable traffic and improve paths to nearby networks. Airnet's 10 Gbps JPNAP connection and open policy expand that option. Yet the port itself costs money, and useful peering requires traffic, routing operations and a physical connection to the exchange. Transit remains necessary for universal reachability. The efficient mix depends on traffic ratios and destinations, which Airnet does not disclose publicly.

Disaster recovery presents the classic idle-capacity trade-off. Full duplication at equal capacity is expensive. Colt's case says Airnet sought on-demand connectivity because its DR site carried no normal traffic. Flexible bandwidth can align cost with recovery use, but it may add provisioning or activation dependencies. The customer should know whether recovery capacity is reserved, immediately available or ordered after an incident.

The human side has the same tension. Enough engineers are needed for round-the-clock response, maintenance, leave, training and simultaneous incidents. Specialist staff are expensive and difficult to recruit. Outsourcing and carrier support extend the pool, but every handoff can add delay. The 2024 interview makes clear that Airnet values engineers who can prevent problems and reason beyond fixed procedures. That capability is central to the service, not overhead detached from the network.

This is why the local connectivity bill depends on upstream routes and field repair. The customer is paying Airnet to make a collection of leased and owned components behave as one service. Logical diversity, exchange access and hardened facilities make that possible. Fast diagnosis, carrier escalation and hands-on repair determine whether the design works during the moment that matters.

What the public record proves, and what it leaves open

The strongest claim is identity and operation. Air Internet Service Co.,Ltd. is the legal English name used by Airnet, a Tokyo company with a service lineage to 1996. AS7503 has been assigned since 1997, is currently announced, carries IPv4 and IPv6 routes, and is visible across global route collectors. Current service pages, contracts, maintenance notices and customer-facing operations show that this is not a dormant registration.

The next strongest claim is logical network diversity. Three upstreams appear in current route observations. Airnet has a 10 Gbps JPNAP Tokyo port and publishes use of BGP. A maintenance notice identifies automatic backup through another backbone line. These facts support multiple logical ways to exchange traffic.

Facility resilience is credible but self-described. The Koto site claims two power systems, UPS, generation, redundant cooling and seismic measures. Airnet describes redundant data-centre links and round-the-clock monitoring. The Osaka DR option and Colt case show geographic recovery intent. No public independent audit gives fuel runtime, full topology or recovery-test results.

Access ownership is clear at a high level. NTT East and NTT West provide the compatible FLET'S lines, while Airnet provides the service that connects those lines to its network. The exact physical route, repair organization and service level vary by customer contract. No public map identifies customer locations or diverse local entrances.

Capacity remains partly opaque. Route collectors establish address space and reachability. PeeringDB establishes one exchange-port rate. Airnet defines a 100 Mbps guarantee for a particular in-data-centre segment. Traffic levels, transit-port sizes, peak utilization and backup-path headroom are undisclosed.

Recovery labour is the least quantified layer. The company has a long operating history, describes 24/7 operations, and discusses training and recruitment frankly. Public records do not show shift rosters, spare inventories, carrier repair targets or measured restoration times. That uncertainty should lower confidence in a precise recovery estimate, not erase the strong evidence that an experienced operation exists.

Several documents would materially improve the assessment. A current topology showing facilities and shared-risk groups would test physical path diversity. Carrier letters could confirm diverse building entrances. Redacted capacity charts could show peak and failover headroom. Generator and UPS test summaries could establish power autonomy. A DR exercise report could show recovery point and recovery time. Incident statistics could demonstrate repair performance. None requires exposing customer identities or sensitive router configuration.

Until then, the network should be judged in layers. The logical internet edge is well evidenced. The data-centre environment has substantial stated controls. The customer access path depends on third-party plant. The human recovery system appears experienced but thinly quantified. That is a more accurate picture than either calling Airnet a fully self-contained carrier or dismissing it as a reseller.

A resilience purchase should be specified end to end

A buyer evaluating AIRnet should begin at the premises. Which device terminates the access line? Is it duplicated? What powers it, for how long, and can the backup circuit operate if the primary router fails? Are two access lines physically separated, or do they share the building entrance and street route? The answer determines whether the first failure can be routed around.

The next question is the access-carrier boundary. Who opens the NTT fault, what service level applies, and who can authorize an after-hours dispatch? If there are two FLET'S services, do they terminate in separate NTT equipment and regional paths? If the backup uses another technology, has automatic failover been tested with the customer's fixed-IP, VPN and firewall rules?

At Airnet's edge, the customer should ask which service elements use AS7503 and which depend on other providers. Three public upstreams are encouraging, but contract-level assurance requires knowing whether the relevant product can use them, whether paths enter through separate facilities, and whether backup capacity can carry the customer's peak load. The JPNAP port improves route choice but should not be mistaken for a second last mile.

Hosted customers need facility and recovery detail. Which Airnet data centre contains the production system? Does it use the stated dual-power and generator environment? What is the contracted backup and restore scope? Is the Osaka DR environment warm, cold or active? How frequently is data copied, and who declares a disaster? Does the application require customer action before traffic moves?

Support commitments should name people and clocks. Twenty-four-hour monitoring is valuable, but customers need acknowledgement, diagnosis, dispatch and restoration targets. They should know whether Airnet, the data-centre operator, NTT or another carrier owns each action. Escalation contacts should be tested before an incident, especially for customers whose applications support reservations, email or other time-sensitive business processes.

Capacity commitments should also be segment-specific. A 100 Mbps guarantee to Airnet's backbone does not state performance to every destination. A 10 Gbps exchange port does not promise 10 Gbps to one customer. A FLET'S line rate does not establish application throughput. Useful commitments define the measurement point, time interval, exclusions, loss, latency and behavior during failover.

The final test is rehearsal. Disconnect the primary access line. Remove one premises power feed. Withdraw one upstream in a maintenance window. Restore a server from backup. Shift an application toward Osaka. Confirm that DNS, certificates, authentication, monitoring and customer communications follow. Resilience that exists only in component descriptions can fail at the joins; an end-to-end exercise tests the joins themselves.

Airnet's published record gives a customer a better starting point than many small providers offer. The company names the access-carrier dependency, shows a current autonomous system, discloses an exchange connection, describes facility controls and publishes examples of backup routing during maintenance. The missing evidence is not whether a network exists. It is whether every important customer service has purchased, configured and tested the right combination of those capabilities.

The correct conclusion is confidence with boundaries

Airnet is not a thin, placeless operator. It is a Japan-focused ISP and managed-services company with nearly three decades of service history, a current Tokyo corporate base, AS7503, dual-stack address space, three visible upstreams, JPNAP Tokyo participation, hardened hosting claims and an Osaka recovery option. The Global metadata understates how specifically the operating evidence points to Japan and should be corrected to the site's Japan or Asia-Pacific classification.

Nor is Airnet a vertically integrated owner of the full connection. Its fixed-IP service explicitly relies on NTT East and NTT West access lines. The company operates the service edge, public routing and managed environments, while access carriers, data-centre operators and upstream networks control other assets. Customers experience one service across those boundaries.

The logical network evidence deserves confidence. Routes are active and globally visible. IPv6 is present. Origin authorizations are valid. Multiple upstreams and an exchange port reduce straightforward transit concentration. Airnet has shown that it can move traffic to another backbone line during planned work.

The physical and recovery evidence deserves conditions. Public information does not establish diverse carrier entrances, independent metro corridors, generator autonomy, full backup-path capacity, spare holdings or repair-time performance. A three-upstream diagram can still share one fibre route; a generator can still face a long outage; a 24/7 monitor can still wait for a field crew.

That boundary is the heart of the company-specific story. Airnet's value is not a single cable. It is the operating work that joins NTT access, AS7503, peering, transit, data-centre systems and customer support into a service that feels ordinary. The bill remains resilient only when the upstream alternatives are physically meaningful, the backup capacity is usable, and the right people can reach the failed component in time.