Internet Data Center Service has to prove whether marketed data-centre capacity can survive power and carrier constraints

Summary

• Internet Data Center Service is the registrant name behind AS134166, TOT-IDC-AS-AP, a Thailand autonomous system registered in April 2015 to TOT Public Company Limited's internet data-centre function and still active in APNIC RDAP.
• The route is real rather than merely administrative. On 12 July 2026, RIPEstat showed AS134166 announcing 31 IPv4 prefixes and one IPv6 prefix, covering 8,704 IPv4 addresses and a /40 IPv6 route, with full visibility across RIPE's IPv4 and IPv6 full-table peers.
• The network footprint is narrow. RIPEstat saw only two neighbouring ASNs, CAIDA counted two providers and no customers, and PeeringDB returned no public network entity for AS134166. That makes carrier, exchange and facility evidence more important than the ASN label alone.
• National Telecom's data-centre pages show an identifiable estate: Bangrak at 2.7 MW and 1,700 square metres, Nonthaburi 1 and Nonthaburi 2 at 1.7 MW and 1,000 square metres each, Sriracha at 1.0 MW and 400 square metres, plus renovation sites. The pages also show important asymmetry, including Nonthaburi language that pairs a Tier 3 design claim with a single power source and two transformers.
• The evidence grade is Medium. A live route, named facilities, public service pages, rack-power options, carrier-neutral claims and third-party IX references support operation. Public evidence still does not show current utilisation, customer failover tests, generator fuel records, cross-connect diversity below the marketing layer, or enough proof that all advertised capacity is simultaneously usable after a power, cooling or carrier event.

A live ASN does not settle the facility question

The first useful fact about Internet Data Center Service is that the network exists in the public routing system. The APNIC RDAP record for AS134166 names the autonomous system TOT-IDC-AS-AP, describes it as Internet Data Center Service, gives country code TH, and records registration on 28 April 2015 with a last change on 5 September 2023. The registrant is Internet Data Center Service at TOT Public Company Limited, with a Chaeng Watthana Road address in Bangkok and abuse contact material that now includes an ntplc.co.th address. That is a legacy label, but it is not an orphaned label.

The corporate surface around it has changed. National Telecom's own public material says NT was established in 2021 from the merger of CAT Telecom and TOT Public Company Limited. NT Data Center's site presents NTDC as being under National Telecom Public Company Limited and says it provides data storage and data-centre services, with colocation, carrier-neutral data centre, connectivity and Smart Hands among its services on the About NT Data Center page. The live commercial brand therefore gives a plausible operating home for a registry entity that still carries the old TOT name.

That identification matters because this is not a cloud-software-only service. The reliability question sits in a physical chain: utility feeders, transformers, switchgear, uninterruptible power, generator fuel, cooling distribution, data-hall airflow, fibre entrances, meet-me rooms, cross-connect records, field engineers, guards, spares and a support desk that can act when something breaks. The AS number can show that a network announces prefixes. It cannot show whether the facility supporting a customer has spare power after one transformer is out, enough cooling after maintenance, or a second fibre path that avoids the same duct.

AS134166 is materially stronger than the dormant ASNs that often appear in small infrastructure dossiers. RIPE's AS overview marked it announced at the 12 July 2026 cut-off. RIPE's routing-status response showed first visibility in June 2017 and current visibility on 12 July 2026. It also showed 31 IPv4 prefixes, 8,704 IPv4 addresses, one IPv6 prefix and 256 IPv6 /48 equivalents. That is a real public edge.

It is also a modest public edge for an operator advertising a national estate. The announced-prefixes response listed 32 current entries when the IPv6 prefix is counted alongside IPv4 routes. The prefixes include address blocks such as 1.179.246.0/24, 1.179.247.0/24, 118.174.8.0/24, 118.175.24.0/22, 180.180.242.0/24, 203.113.8.0/22 and 2001:ec0:100::/40. The route count is enough to support live services, but it is not by itself proof of large multi-site colocation activity or independent customer traffic.

The neighbour picture is even narrower. RIPE's ASN-neighbours response showed two neighbours, AS38040 and AS131293, both on the left side of the observed path. CAIDA's AS Rank entry marked AS134166 seen, but counted two providers, no peers and no customers. That fits a network that is visible and externally connected but not publicly evidenced as an interconnection hub in its own right. The public network view therefore supports operation while keeping the buyer's focus on whose carrier contracts, facility rooms and cross-connects are actually involved.

The public estate has named sites, but capacity is uneven

NT Data Center's locations page says the business has nine data centers across eight provinces. The visible named list, at the time reviewed, showed Bangrak, Nonthaburi 1, Nonthaburi 2, Sriracha, Phra Khanong, Chiang Mai, Khon Kaen and Hat Yai. It also described Phra Khanong, Chiang Mai, Khon Kaen and Hat Yai as under renovation, with an estimated completion date of June 2026. That means the public estate combines currently marketed sites and sites whose readiness requires separate verification.

The four named operating pages provide the clearest current capacity numbers. NT Data Center Bangrak is listed at 2.7 MW of IT power and 1,700 square metres of data-hall area. NT Data Center Nonthaburi 1 is listed at 1.7 MW and 1,000 square metres. NT Data Center Nonthaburi 2 is also listed at 1.7 MW and 1,000 square metres. NT Data Center Sriracha is listed at 1.0 MW and 400 square metres. Those pages add up to 7.1 MW and 4,100 square metres for the four named facility pages, before any contribution from renovation sites or other space that is not surfaced with the same detail.

Those numbers are useful because power and floor area are the main constraints a colocation buyer eventually hits. They are not the same as sellable capacity. A facility can have nameplate IT power but still lack available racks, distribution capacity, cooling margin, power whips, security cages, cross-connect availability or staff coverage for a particular deployment date. A customer also needs recoverable capacity, which is the amount that remains available after removing the largest credible fault domain. A site that is 80 per cent occupied may have enough space for a normal order and too little space to absorb a hall-level recovery event.

The site language encourages a careful reading. Bangrak is described as a carrier-neutral central Bangkok facility with dual power sources and diverse communication routes. Sriracha is described as carrier-neutral, near Thailand's submarine cable station and close to the Eastern Economic Corridor, with dual power sources and advanced cooling. Nonthaburi 1 and 2 are also described as carrier-neutral, with three diverse communication routes, advanced power and cooling systems, and ISO/IEC 27001 and ISO/IEC 27701 claims. But the Nonthaburi pages also say the reliable-infrastructure design uses a single power source and two transformers.

That is not fatal; a facility can still use redundant components behind a single upstream utility source. It does mean the failure analysis must distinguish dual rack feeds or 2N internal equipment from independent utility supply.

This is the core capacity problem. Marketing pages commonly collapse several different concepts into one resilience story: Tier design, dual rack feeds, redundant UPS modules, generator sets, carrier-neutral access and 24-hour support. Each concept can be true while the combined service still has a shared point of failure. If two rack feeds originate behind the same utility source, they help with rack distribution failure but not with all upstream supply events. If three communication routes share a building entrance, they help with carrier choice but not with an excavation outside that entrance.

If a site is "near" a submarine cable station, it improves route economics and latency potential but does not automatically prove diverse long-haul paths into the room.

Power evidence is the first procurement gate

Data-centre reliability starts with electricity because every other service depends on it. A rack customer experiences power as a simple pair of sockets, but the chain behind those sockets is long: utility substation, medium-voltage feed, transformer, switchgear, UPS, battery, generator, fuel system, power distribution unit, branch circuit and rack power strip. A failure at any shared stage can turn a redundant-looking rack into a single-risk system.

The public pages give different levels of confidence by site. Bangrak states dual power sources. Sriracha states dual power sources. Nonthaburi 1 and Nonthaburi 2 state a single power source and two transformers while also presenting themselves as Tier 3 design facilities. The public specification-sheet links on the facility pages, including NTBangrak.pdf, NTNonthaburi_1.pdf, NTNonthaburi_2.pdf and NTSriracha.pdf, are therefore not decorative files. They are the documents a buyer should demand, preserve and compare with the signed order.

The clearest publicly indexed technical detail appears around Nonthaburi. Publicly available copies of the Nonthaburi material describe a site at 299 Tiwanon Road, a fourth-floor data centre in Building 4, an 800 kg per square metre floor load, downflow air conditioning, environmental controls in the 18 to 25 C and 45 to 60 per cent humidity range, a single substation, 2N transformers, 2N UPS, dual 230V rack feeds, N+1 generators and 12 hours of full-load fuel backup. That is a much richer power picture than a short web page headline, and it also shows why the words must be separated.

A single substation and 2N downstream equipment do not describe the same redundancy boundary.

For a high-dependency customer, the minimum evidence is a one-line diagram and a maintenance record. The diagram should show where A and B power paths separate, where they converge and which components can be maintained without customer load shedding. The maintenance record should show at least one recent generator load test, UPS test, battery inspection and switchgear exercise. Fuel records should state usable runtime at the contracted load, not only tank capacity or a generic "full load" phrase.

A contract should state whether the customer can request the last test date and whether critical maintenance is communicated before work begins.

Power also turns advertised rack choices into economic constraints. NT's services page lists colocation choices from shared rack units through quarter, half and full racks, including full rack options at 3 kW and 5 kW and cage options at 5 kW. Those offers are practical and concrete, but they show the arithmetic. A 1.7 MW site is not a boundless pool of 5 kW racks. Once cooling reserve, stranded power, network rooms, house load, high-density customers and recovery headroom are considered, the sellable rack count can be materially lower than a simple division of megawatts by rack rating.

Carrier neutrality is valuable, but it must be traced below the slogan

NT Data Center repeatedly describes its sites as carrier-neutral. That is commercially important. A carrier-neutral facility can let customers choose among telecom providers, place routers near exchange fabrics, use cross-connects to reach carriers or peers, and avoid being locked to one upstream. NT's services page says connectivity supports Thailand IX, BBIX Thailand, NT IIG and local carrier providers, with cross-connects, external fibre connections and on-demand services. The same page says Carrier Data Center service is offered for telecom-licensed carriers and describes the facilities as 100 per cent carrier-neutral.

There is outside support for the Bangrak interconnection story. BBIX announced in 2021 that it had established a Thailand node in the CAT Tower BangRak Data Center provided by National Telecom. BBIX's IX service list also shows BBIX Bangkok No.2 at NT Bangrak Tower. That is meaningful evidence that Bangrak is not just a sales location. It is a place in which an exchange operator publicly places service.

The public route evidence for AS134166 is thinner than the facility-neutrality story. PeeringDB's network query for ASN 134166 returned no public network entity at the cut-off. PeeringDB participation is voluntary, so the absence does not prove that NT Data Center has no peers, no facilities or no interconnection policy. But it removes a common independent source for facility presence, traffic level, peering policy, exchange participation and operations contacts. The buyer is left with NT's own pages, BBIX references and routing telemetry.

RIPE's routing data shows the public AS has two observed neighbours, not a mesh of public peers. CAIDA likewise counts two providers and no customers. This does not make the facility carrier-neutral claim false. The facility can be carrier-neutral while the specific AS134166 route is small, transit-oriented or used for selected IDC services. It does mean a customer should not infer route diversity from carrier-neutral marketing. Carrier neutrality is a real estate and access property. Route diversity is an end-to-end property from the customer's device to the rest of the internet.

The due-diligence request should therefore be physical. Ask for the carrier list available in the room, the meet-me room or demarcation point, cross-connect lead times, dual-entry status, internal fibre path diversity, and the exact carriers on the customer service order. Ask whether AS134166 is in the customer path or whether another NT, partner or customer ASN originates the traffic. Ask for route tests from both rack feeds or both edge routers. A traceroute cannot prove fibre-route diversity, but a controlled failover report can show whether the customer remains reachable when one carrier service or one edge router is withdrawn.

Bangrak is the strongest public interconnection signal

Bangrak carries the strongest public case because three evidence layers line up. The NT page lists the site as central Bangkok, 2.7 MW, 1,700 square metres, Tier 3, dual power sources, carrier-neutral and connected through Thailand IX and NT IIG. BBIX separately places an exchange node at NT Bangrak Tower. The broader NT Data Center site describes carrier data center service and cross-connect connectivity. For customers needing low-latency Bangkok access or exchange presence, that is enough to justify a serious procurement conversation.

It still does not settle the most important operational question: what happens when the building or carrier-meet layer is impaired? Bangrak's value comes partly from concentration. It is in central Bangkok, near telecom infrastructure, exchange access and business demand. Concentration creates short paths and market liquidity, but it also concentrates risk. A single tower, cable entry zone, shared electrical room or meet-me room can become the limiting point even when many carriers are available on paper.

The BBIX reference helps but should not be overread. An exchange node proves that an operator saw value in the facility and placed infrastructure there. It does not prove that every NT colocation customer has direct access to that fabric, that all cross-connects are quick to deliver, that the exchange node is in the same room as the customer's rack, or that traffic can shift automatically to another city if the Bangrak room has a problem. A buyer using Bangrak for internet-facing workloads should decide whether a second NT site, another Bangkok carrier hotel, or an external cloud region is the recovery target.

Bangrak also illustrates the difference between installed capacity and recoverable capacity. A 2.7 MW central site can serve many normal deployments. But if customers use it as their sole Thailand edge, the ability to recover from a major room or building event depends on pre-built capacity elsewhere. Recovery cannot be improvised by ordering cross-connects after an incident begins. It requires spare racks, reserved ports, routing plans, security access, tested data replication and a contract that lets engineers act quickly.

Nonthaburi is the power-boundary test

Nonthaburi 1 and Nonthaburi 2 look similar in public presentation: each is 1.7 MW, 1,000 square metres, outside Bangkok, carrier-neutral, linked to Thailand IX and NT IIG, and supported by 24-hour operations. Each page refers to three diverse communication routes and advanced power and cooling systems. Each also says the reliable-infrastructure design has a single power source and two transformers.

That combination is plausible, but it changes how the buyer should read the risk. A single power source can be normal for many facilities if downstream redundancy and generators are strong. It is not the same thing as independent utility feeds from separate substations. Two transformers can protect against transformer failure and maintenance. They cannot protect against all upstream substation, utility distribution or site-level switching failures. N+1 generators can protect against extended grid loss if they start, carry load and have fuel. They do not eliminate transfer risk, fuel logistics or common controls.

Nonthaburi therefore needs a more granular power answer than Bangrak. Which loads are dual-fed? Are all customer racks delivered with A and B feeds? Are the A and B paths separated through independent UPS systems and distribution boards? Is any part of the path shared before the rack? How recently were generators tested under load? How much fuel is available at the committed rack load? Is replenishment contracted during regional disruption? What maintenance can be performed while customer load remains protected?

The public pages also mention ISO/IEC 27001 and ISO/IEC 27701 certifications. Those are useful controls signals, especially for access, information-security management and privacy management. They do not replace electrical and mechanical proof. A secure process cannot keep servers alive during a failed transfer switch. A privacy management certificate cannot prove that a chilled-water valve or CRAC unit has a spare. For Nonthaburi, the evidence has to be engineering-specific.

The location near Bangkok is a strength if it gives customers a second metro site with different flood, utility and cable exposure from Bangrak. It is weaker if Nonthaburi is used only as another room in the same operational envelope without true route and power independence. The public data does not prove either way. It gives enough detail to ask the right questions and not enough detail to skip them.

Sriracha is a connectivity story with a different failure map

Sriracha is the smaller named site by public metrics: 1.0 MW and 400 square metres. Its page positions it near Thailand's submarine cable station and the Eastern Economic Corridor, with direct access to cables including AAG and ADC. That makes the site strategically different from the Bangkok and Nonthaburi pages. It is not just another metro room. It is marketed as an international-connectivity and industrial-corridor location.

That location can matter. A site close to submarine cable infrastructure can reduce latency, simplify international-carrier access and support customers whose traffic patterns are tied to manufacturing, port activity, logistics or regional enterprise systems. It can also serve as a non-Bangkok recovery point if the facility, power and terrestrial routes are truly independent from Bangkok metro exposure.

The public evidence still needs translation into service claims. "Near" a cable station is not the same as a direct, diverse, contracted path into multiple subsea systems. Direct access to AAG and ADC should be backed by carrier service orders, cable landing arrangements, protected terrestrial backhaul and route maps. Multiple communication routes should be shown as physical diversity, not simply multiple logical services through a common access path. Dual power sources should be mapped to upstream supply, not only rack distribution.

Sriracha's smaller power envelope also affects failover. A 1.0 MW site may be excellent as a regional edge or a smaller colocation point. It may not be able to absorb a large Bangrak customer base in a broader outage. Customers should size Sriracha against the workload they intend to recover there, not against the idea of a national NT estate. If the recovery plan assumes that Sriracha will take Bangkok load, the contract should reserve the necessary racks, ports, power and cross-connects before the incident.

The site can still be valuable precisely because it is different. A well-prepared customer might use Bangrak for exchange density, Nonthaburi for metro separation and Sriracha for subsea-proximate or corridor-facing resilience. That architecture only works if replication, routing and support are tested as a combined system. Three locations on a map do not become a recovery design until they can run the workload while one location is deliberately removed.

Route security is adequate in samples, but routing is only one layer

Public routing security for AS134166 is not the main weakness in the evidence. A RIPE RPKI validation query for 1.179.247.0/24 returned valid status for the sampled current route, with a validating ROA for origin AS134166 and max length 24. That is a positive sign. It means at least one visible customer-relevant route has a route-origin authorization matching the observed origin at the cut-off.

The broader RIPE routing-consistency response also showed current prefixes present in BGP and in whois or routing-registry sources for the sample returned by the API. The imports and exports for observed peers were in BGP but not in whois in the portion inspected. That is not unusual, but it means public routing-registry policy is not the only evidence to use.

Routing controls reduce one class of incident: mistaken or unauthorized origin announcements. They do not prove that the data centre has two carriers, that a cross-connect is installed, that a router has redundant power, or that staff can change filters under pressure. A buyer should ask for maximum-prefix controls, route-filter policy, RPKI origin validation settings, DDoS diversion arrangements and out-of-band management access. Those questions belong beside the power and facility questions, not after them.

The public AS also appears to sit without visible downstream customers. CAIDA's provider-degree result and the lack of a PeeringDB profile suggest a network used for NT's own IDC service or internal service boundaries rather than a public transit ecosystem under AS134166. That is fine if customers buy colocation and bring their own networks. It is more important if customers buy bundled internet from NT and assume the AS itself represents carrier diversity. The contract should say which ASN originates the customer's service and which carrier path is used for each option.

Certifications and Tier language need a scope test

NT Data Center pages repeatedly use Tier 3 language and list ISO/IEC 27001 and ISO/IEC 27701. The About page says NTDC adheres to those international security standards and describes advanced environmental controls, fire-safety systems, dual power sources and 24-hour expert support. Those are useful statements. They are not a substitute for a certificate scope and an engineering acceptance test.

Uptime Institute's public material is helpful here because it separates phases. Its Tier Certification overview describes certification across design, construction and ongoing operations. The Operational Sustainability page describes operations as a later phase, and the Construction page distinguishes verification of the constructed facility from design intent. Uptime's award terms also describe design-document awards and constructed-facility awards as different kinds of evidence. The procurement implication is simple: a page saying "Tier 3 design" should be read as a design claim unless the operator supplies the actual award, site name, phase and capacity covered.

The ISO scope question is similar. ISO/IEC 27001 and ISO/IEC 27701 can be strong evidence of management controls if the certificate covers the site, service and operating unit being purchased. If the certificate covers a broader NT function, a subset of sites or a specific management office, it may not prove that every advertised hall operates under the same audited controls. Customers should ask for the certificate, scope statement, expiry date, certification body and included sites.

Fire and environmental controls also require site-specific proof. The public pages mention advanced cooling and fire safety, and the indexed Nonthaburi specification material refers to gas suppression, photoelectric smoke detection, alarm signalling and environmental set points. These controls matter, but they must be checked against customer equipment density. A 5 kW rack in a low-density room may be easy to cool. A denser cage or mixed-density row can create local hot spots even when the room average looks healthy.

The buyer should ask for maximum rack density by zone, containment design, cooling redundancy, alarm thresholds and escalation rules.

Flood exposure should be checked separately. Publicly indexed Nonthaburi material describes a flood-free area, but the buyer should request the flood basis, floor level, drainage assumptions and whether external utility and fibre paths share the same exposure. Facility resilience is not only the data hall. It includes the routes that let power, fuel, people and fibre reach the hall.

Renovation sites should not be counted as current recovery capacity

The renovation sites are important because they can make the estate look broader than the verified operating footprint. The locations page lists Phra Khanong, Chiang Mai, Khon Kaen and Hat Yai as under renovation with estimated completion in June 2026. It also shows capacities such as 150 racks for Phra Khanong and 25 racks for Chiang Mai and Khon Kaen. Those are useful expansion signals, not current proof of recovery capacity.

Completion estimates should be treated as hypotheses until they are backed by commissioning evidence. The question is not only whether construction work finished. A site becomes useful to a customer when the facility has power, cooling, network access, security controls, operating staff, cross-connect ordering, customer contracts, monitoring, incident escalation and enough spare capacity for the promised service. A site under renovation can be strategically important while still being irrelevant to a live failover plan.

This matters for national customers. A government agency, enterprise, carrier or content provider may see nine locations and infer geographic resilience across Thailand. If only some sites are ready, if some lack carrier diversity, or if some are not yet accepting the customer's rack density, the usable estate is smaller. A buyer should separate current sellable sites, commissioned but unavailable capacity, renovation sites and planned future locations.

There is nothing wrong with a public operator showing expansion. The problem is counting expansion as resilience before it is commissioned and tested. A June 2026 estimate, viewed on 12 July 2026, calls for a direct status update: which sites were completed, which are taking orders, which have live carriers, and which can carry production load today. The article's grade stays at Medium partly because the expansion story is plausible but not fully settled by public evidence.

Who is affected when the system fails

The affected users are not only retail website owners. NT Data Center's public service set includes colocation, carrier data center service, connectivity, Smart Hands, shared web hosting, email hosting, temporary office and add-on services. That means a failure can hit rack tenants, carriers, internet providers, cloud and content customers, enterprises using cross-connects, small customers using hosted services, and organisations using the facilities as backup office or continuity infrastructure.

The failure modes differ by customer. A colocation customer may lose power to a rack, lose cooling margin, lose cross-connect access or need remote hands to replace equipment. A carrier customer may lose a neutral meet point or a cross-connect path to another provider. A connectivity customer may lose external fibre, exchange reachability, NT IIG access or a local carrier handoff. A Smart Hands customer may discover that the most important constraint is not the server but technician availability and authority to touch the equipment.

Bundled services add account and support risks. Shared hosting or email customers may not know which facility or ASN supports them. Temporary-office customers may depend on both the physical room and the connectivity service. Add-on services such as extra IP addresses, extra internet ports and excess electricity can become bottlenecks during growth or recovery. If a customer assumes extra power or ports can be added during an incident, it may be surprised by lead times, approval requirements or exhausted panels.

The public pages mention 24-hour expert support and monitoring. That is a positive signal, but incident capability depends on scope. A guard can open a door. A remote-hands technician can reseat a cable. A network engineer can change a route. A facility engineer can transfer load or inspect cooling. A commercial manager can approve an emergency order. A major incident may need all of them. The customer should ask who is on site, who is on call, who can authorize carrier or utility escalation, and how the contact path works if the normal portal is down.

What evidence would raise the grade

The first missing evidence is utilisation. Public pages give megawatts, square metres and rack-product shapes, but not occupied racks, free power, reserved recovery capacity, port availability or cooling headroom. A buyer does not need every customer's inventory. It does need a current capacity statement for the site, hall and row being purchased, including the power density and growth reserve.

The second missing evidence is tested failover. The operator should be able to show a recent exercise in which one power component, one carrier path or one facility zone was withdrawn and customer service remained inside a defined objective. A paper design matters, but a measured exercise reveals hidden dependencies: shared management networks, DNS assumptions, route filters, staff permissions, spares, access-control delays and communication gaps.

The third missing evidence is carrier path diversity. The public estate has carrier-neutral claims and a third-party IX node at Bangrak, but each customer path needs a specific carrier map. The map should name the carriers, building entrances, meet-me points, cross-connects, edge routers and ASN used for the service. If the customer buys two links, it should know whether they share a duct, room, router, optical shelf or long-haul provider.

The fourth missing evidence is certificate and Tier scope. For each site, the buyer should collect the actual ISO certificates and any Tier award documents, including site names, expiration dates and certification phase. If the public site says "Tier 3 design", the buyer should not silently convert that into a constructed-facility or operations award. If the certificate covers only some locations, the contract should not imply all locations are identical.

The fifth missing evidence is exit and continuity. Colocation customers need equipment-removal procedures, access windows and termination rules. Hosted customers need backup exports, DNS control, IP portability and data-retention terms. Carrier customers need cross-connect cancellation and migration terms. Every customer needs an out-of-band contact path. Resilience is not only avoiding outage; it is being able to leave or move when the risk model changes.

Verdict: operational, but not yet fully evidenced as resilient capacity

Internet Data Center Service deserves a stronger starting point than a thin or dormant operator. AS134166 is live. The registrant and abuse data tie it to the old TOT identity now inside National Telecom. RIPE sees current prefixes with full collector visibility. CAIDA sees the AS. NT Data Center publishes named facilities, power and area figures, rack options, carrier-neutral service lines, Smart Hands, connectivity options and renovation sites. BBIX independently supports the Bangrak exchange-location story.

The downgrade is not about existence. It is about proof of recoverable capacity. The public estate combines strong claims and unresolved boundaries. Bangrak has the best interconnection signal. Nonthaburi has useful capacity but a power story that needs careful separation between single source, two transformers and downstream redundancy. Sriracha has international-connectivity logic but a smaller power envelope. Renovation sites should not be counted as current recovery resources until commissioning is proven.

For a small rack, the evidence may be sufficient to begin normal procurement. For a critical service, the buyer should require site-specific diagrams, carrier path records, certification scopes, maintenance logs, generator and fuel records, port availability, support escalation, failover tests and exit terms. The fact that AS134166 is routed helps. It does not prove that every marketed facility can keep operating through the failure paths that matter.

The grade is Medium. The public evidence supports an operating national data-centre service with live routes and real facilities. It does not yet support a stronger grade because the most important resilience claims still sit at the marketing and specification level, while measured failover, utilisation and path-diversity evidence remains private or unpublished.