Summary

  • Puerto Rico Internet Exchange has credible public signs of an active, multi-site local peering service, but those signs do not establish that every site, route server and member circuit can survive a prolonged island-wide emergency.
  • One listed exchange facility publishes an estimated five days of generator runtime without refuelling, making fuel delivery and safe physical access—not nominal port speed—the clearest public clock on sustained operation.
  • PR-IX becomes resilience infrastructure only when powered exchange equipment, diverse member backhaul, working local content paths and disciplined recovery coordination remain available at the same time.

The first hours expose the real exchange

In the first hours after a hurricane crosses Puerto Rico, the useful packet is often an ordinary one. It may carry a message from a family member in Caguas to a service hosted in San Juan, a request from a clinic to a local platform, a university lookup, a news page or a connection from a municipal office to an island-based provider. If both endpoints can meet at a local internet exchange, that packet should not have to leave Puerto Rico, cross an international or continental path and return. The distance saved is not merely a latency improvement.

It removes an avoidable dependency at the moment when landing stations, long-haul links and remote networks may all be under pressure.

Yet “local” describes the preferred route, not the packet’s physical independence. The customer’s access equipment needs power. The provider’s neighbourhood node and metro aggregation equipment need power. The fibre between that provider and the exchange has to remain intact. The exchange switch and route servers have to be reachable through working optics, cooling and control systems. The destination network needs its own surviving path. A local route that exists in the Border Gateway Protocol can therefore become unusable while it still looks perfectly sensible on a diagram.

Hurricane Fiona made that distinction concrete. The Department of Energy’s public situation-report index records an island-wide blackout after the September 2022 storm. Its 26 September situation update reported roughly 608,000 customer outages eight days after landfall, or 41 per cent of customers, even after substantial restoration. The same update described high demand for diesel and gasoline as people continued to run backup generators. That is the environment in which an exchange’s published electrical design meets the logistics of roads, ports, queues, staff safety and refuelling.

Maria established the harsher outer bound. A World Bank assessment of telecommunications and natural hazards says Puerto Rico suffered a near-total loss of connectivity, with 95 per cent of cell sites offline and large portions of above-ground and last-mile fibre destroyed. It also reports that, into January 2018, an estimated 60 per cent of online telecommunications infrastructure depended on diesel generators. An FCC communications status report from February 2018, issued months after Maria, still treated commercial-power loss as an event capable of taking cell sites down. Those records do not describe PR-IX performance specifically. They describe the coupled system in which any Puerto Rican exchange must operate.

The first question after landfall is thus not whether PR-IX advertises a high-speed port. It is whether two affected networks still have a continuous electrical and optical path to the same functioning switching plane. The second question is how long that shared condition can last. A generator can preserve a rack while a severed lateral isolates a member. A surviving metro circuit can lead to a dark building. A healthy route server cannot repair a flooded splice enclosure. Resilience begins only when these elements overlap.

That is why the most revealing public number in the PR-IX story is not 100 gigabits per second. It is an estimate of about five days of facility runtime without refuelling at one of the sites the exchange identifies. Five days is long enough to bridge a short outage. It is not long enough to make fuel logistics irrelevant in a Maria-scale recovery. The countdown begins as soon as the grid fails.

What PR-IX is—and what it is not

Puerto Rico Internet Exchange presents itself as a neutral interconnection ecosystem intended to improve quality and reduce the cost of exchanging traffic between networks in Puerto Rico. The operator’s public site describes peering and content-delivery interconnection, names local and international entities, advertises access speeds up to 100 Gbps and lists exchange availability at sites in San Juan, Guaynabo and Bayamón. This is credible evidence of a service being marketed and maintained, not merely an abandoned idea.

The network identity is independently visible. A registration mirror for AS18804 reproduces ARIN data naming Puerto Rico Internet Exchange Inc. and shows the autonomous-system record under the PR-IX name. PeeringDB’s AS18804 record characterises the network as a route server, shows two operational 10 Gbps exchange connections with IPv4 and IPv6 addresses, and reports an open peering policy. Packet Clearing House’s PRIX entry calls the exchange active, identifies San Juan, records an Ethernet medium and gives 2019 as the establishment year. These records reinforce one another on the basic point: PR-IX has a public network identity and an exchange function.

They do not establish the same thing as a single, fully owned physical network. An internet exchange is a coordination and switching surface. Its entities usually own or lease their own access circuits. Colocation operators control the rooms, utility entrances, cooling plants, generators and physical-access procedures. Fibre carriers control metro spans and repair crews. Content networks decide which caches and routes remain available. Upstream providers and submarine-system operators control the off-island paths.

PR-IX can set technical rules, operate switching and route-server functions, manage ports and coordinate members, but it does not thereby own every dependency that makes a packet deliverable.

This boundary matters in both directions. It would be unfair to attribute every member outage, cut fibre or failed generator to the exchange company. It would be equally misleading to treat a powered exchange switch as proof that local connectivity survived. The product is the intersection of systems controlled by different organisations. A meaningful resilience account must say which party owns each component, who can observe it, who can repair it and how the exchange communicates when a failure sits beyond its direct control.

The route server adds another important boundary. It can simplify multilateral peering by allowing a member to establish a session with a common service rather than negotiating a separate session with every peer. It can apply routing policy and distribute reachability. It does not carry traffic in the data path in the way an exchange switch does, and it cannot create physical reachability when a port or member circuit is down. Direct bilateral sessions can also exist alongside it.

Therefore, a route-server failure may reduce or withdraw routes for members that depend on it, while members with direct sessions could continue to exchange traffic. Conversely, two healthy route servers offer limited protection if they share a switch, power domain, facility or backhaul failure.

The exchange should consequently be assessed as a layered service: legal operator, switching fabric, route-control layer, colocation footprint, member access, content presence and operating team. Public records are strong on the first two layers’ existence and useful on the announced footprint. They are thin on failure-domain separation, emergency staffing and tested recovery. That is not proof of fragility. It is the line between what outsiders can verify and what still needs to be demonstrated.

A metropolitan footprint with several kinds of diversity

PR-IX’s site page names five access locations: Telxius CLS at Punta Las Marías in San Juan; the NetWave data centre on Avenida de la Constitución in San Juan; an AeroNet point of presence and the FiberX data centre in Metro Office Park, Guaynabo; and Engine-4 in Bayamón. This footprint is more geographically distributed than a single-rack exchange. It spans several municipalities in the metropolitan region and places the service near multiple carrier and facility ecosystems.

The broader public datasets describe an even wider logical presence. PeeringDB lists AS18804 at Puerto Rican facilities as well as at sites in Jacksonville, Atlanta, northern Virginia, California and Oregon. Those continental entries may support remote peering, control, transport or other operational arrangements; they should not be read as proof that a Puerto Rican member has physically diverse access to every listed building. PeeringDB also labels facilities rather than drawing the actual fibre paths between them. Its useful contribution is to show where the AS says it is present, not how packets are protected between any two points.

Internet Society Pulse provides the clearest current aggregate. Its Puerto Rico IXP tracker reported one active exchange, 28 members and 15 physical locations in July 2026. It estimated that 71 per cent of 95 active Puerto Rican networks were either exchange members or customers of members, and that 21 per cent of the 1,000 most-visited sites had an in-country server or cache. Pulse explicitly says that its IXP list is based on self-reported PeeringDB data. The figures are therefore valuable indicators of reach, not audited counts of simultaneously usable disaster capacity.

The named local facilities expose different physical claims. NetWave’s network page says its Puerto Rican backbone has six nodes, including San Juan network operations and data-centre facilities, a second data-centre facility in Ponce, and nodes in Caguas, Arecibo, Mayagüez and Humacao. That suggests an operator with island-wide reach beyond the exchange room. It does not reveal which of those paths are used by PR-IX members, whether they share conduits, or whether the San Juan access circuit has a protected route.

FiberX is more specific about its Guaynabo building. Its colocation description says the facility is about five miles from Puerto Rico’s central business district, outside identified flood and tsunami evacuation zones, hardened for hurricane winds, and served through diverse fibre entrances. It publishes two 250 kW backup generators in an N+1 arrangement, two 100 kW managed uninterruptible power systems, 1,000 gallons of above-ground fuel and an estimated five days of runtime without refuelling. It also identifies a fuel-priority arrangement. These are unusually useful design details, but they remain the facility operator’s statements; the load assumed for the runtime and the current maintenance state are not public.

HUB787 offers a useful comparison, even though the available PR-IX records reviewed here do not identify it as one of the exchange’s announced access sites. The HUB787 facility page says the cable-landing and data-centre building ran for 42 continuous days on generators after Irma and Maria before utility service returned. That is a historical operating result, not a guarantee for a future event and not evidence about PR-IX. It demonstrates why duration and refuelling history are more informative than a generic assertion that a building has backup power.

Geographic separation is only one kind of diversity. Five addresses can still depend on a common carrier, conduit corridor, utility substation, fuel distributor, control system or staff pool. Conversely, two sites close together can reduce risk if they have genuinely independent power, fibre entrances and operating teams. The public footprint is therefore a promising starting point. Its resilience value cannot be inferred from pin count alone.

The route-server signal is strong but incomplete

AS18804 is unusually visible for a small-island exchange because the public records expose both route-server addressing and operating signals. PeeringDB shows two operational 10 Gbps connections for AS18804 at PR-IX, using 204.138.0.251 and 204.138.0.252 for IPv4 and corresponding IPv6 addresses. It reports the route-server network’s traffic band as 100–200 Gbps, with balanced traffic, and notes support for unicast IPv4, multicast and IPv6. The profile was last updated in July 2025, while its public peering information carries an earlier update date.

Those fields answer some narrow questions well. Two server addresses indicate a paired service rather than a lone advertised endpoint. IPv6 is not merely promised in prose; addressing is shown. A 100–200 Gbps band is consistent with an exchange carrying material traffic. The open policy and absence of ratio or contract requirements are consistent with reducing barriers to local interconnection.

They do not answer the failure questions. PeeringDB does not identify the chassis hosting each route server, whether the two instances run in different buildings, whether their power supplies land in independent electrical paths, whether their management access is out of band, or whether configuration state can be restored without the primary site. The two displayed 10 Gbps connections are route-server ports, not a total of exchange fabric capacity. Nor does the 100–200 Gbps band specify a measurement interval, peak, average, direction or independently observed counter.

Packet Clearing House provides a useful counterweight. It calls PRIX active and displays IPv4 and IPv6 exchange subnets, but its page shows no switch inventory and no populated facility section. That absence does not mean PR-IX has no switches or sites; the operator and PeeringDB clearly describe sites. It means PCH’s record is incomplete for physical verification. The disagreement is about dataset coverage, not necessarily about operations.

Cloudflare Radar’s AS18804 page also recognises PR-IX and associates the autonomous system with Puerto Rico. Radar’s HTTP-oriented observations can show traffic trends when enough traffic is visible to Cloudflare, but they cannot measure all layer-two exchange traffic or prove the state of private member-to-member flows. A quiet chart could reflect limited observability rather than an outage. A busy chart could show traffic reaching Cloudflare without demonstrating that every exchange site survived.

This distinction between control plane, data plane and observation is essential. The route server distributes reachability. The switching fabric forwards Ethernet frames between member ports. External datasets observe selected announcements or application traffic. These layers can fail separately. During an emergency, a route may remain visible briefly after a circuit is cut; a healthy switch may forward only among the handful of members whose access survived; external monitoring may see the exchange ASN while local peering at one site has collapsed.

A defensible status statement is therefore modest but positive: public data strongly supports an operating route-server network, dual-stack exchange addressing and non-trivial traffic. It supports a multi-site service at the level of announced presence. It does not establish route-server geographic separation, switch redundancy, per-site load, spare-port availability or disaster-state usable capacity. The difference is exactly what a resilience publication should make visible.

Power starts a clock that port capacity cannot stop

The exchange’s advertised ceiling of up to 100 Gbps tells a prospective member what class of port may be available. It says almost nothing about endurance. When the grid fails, the relevant quantities change to battery ride-through, generator start success, usable fuel, cooling load, refuelling interval and the time required for a technician to reach the building.

FiberX’s published design makes that conversion possible at one announced PR-IX location. Two 250 kW generators and two 100 kW managed UPS units are described as N+1 systems. The site lists 1,000 gallons of fuel and approximately five days of runtime without a refill. If those statements reflect current configuration, the building has meaningful protection against short interruptions and a defined bridge into a longer emergency. But N+1 is a topology description, not a probability of success. It depends on maintenance, transfer gear, fuel quality, starting batteries, cooling controls and a load that stays within the protected envelope.

The five-day estimate is particularly revealing because Maria and Fiona exceeded it. DOE’s 2023 grid-recovery review says Fiona knocked out the entire grid and left some areas without power for as long as four weeks. DOE’s Puerto Rico recovery page says telecommunications, water, transportation, health and manufacturing were severely disrupted after the 2017 collapse, often because of electrical damage or pre-existing instability. A five-day tank therefore moves the problem from immediate shutdown to replenishment; it does not remove the problem.

Puerto Rico’s power risk is not solely historical. In February 2026, DOE renewed emergency orders for grid generation and vegetation work as the island approached hurricane season. The announcement said the actions had helped restore up to 820 MW of baseload generation capacity and described continuing reliability challenges. An earlier DOE account of building a more resilient Puerto Rican grid observed that Maria took the entire grid down. These are system-level statements, not forecasts for any particular exchange facility, but they justify treating prolonged utility loss as a design condition rather than a remote exception.

Puerto Rico’s housing department likewise calls energy the most critical factor in long-term recovery. Its Electrical Power Reliability and Resilience programme allocates funding toward decentralised generation, distribution and storage after the 2017 blackout. Such investments may improve the environment in which communications facilities operate. They do not substitute for site-specific autonomy because a hardened exchange cannot assume that a general grid programme will restore its feeder before its fuel runs out.

Fuel is part of the power system. DOE’s Fiona update noted that Puerto Rico imports petroleum products, distributes them from ports and terminals, and has almost no petroleum pipelines. Backup-generator fuel must ultimately move through a transport network competing with hospitals, water systems, public safety, businesses and households. A priority agreement helps, but performance depends on the supplier’s stock, the road, the delivery vehicle, site access and safe handling. The exchange’s endurance is therefore bounded by logistics outside the equipment room.

The public gap is clear. No comparable runtime, generator, fuel or battery specification is available across all PR-IX sites. One facility’s detailed disclosure cannot be copied onto Telxius, NetWave, AeroNet or Engine-4. The exchange may have stronger arrangements than the public can see. Until those arrangements are stated or demonstrated, the prudent conclusion is that power redundancy varies by location and that the shortest surviving member path—not the best-equipped room—sets actual service continuity.

Member backhaul starts a second, independent clock

An exchange switch can remain powered throughout a storm and still lose most of its usefulness as member circuits fail. This is the second clock: how long the terrestrial paths from access and aggregation networks to the peering fabric remain intact, and how quickly damaged spans can be isolated or replaced.

Puerto Rico’s terrain makes that a physical question. Fibre may be buried in conduit, attached to poles, carried across bridges or road cuts, or handed between carriers at facilities. Flooding can affect vaults and building entrances. Wind, trees and falling poles can damage aerial plant. Landslides can isolate mountain routes. Repair crews need access, accurate records, spares, fuel and permission to enter affected areas. Two commercial circuit orders are not diverse if both eventually use the same duct, bridge, pole line or carrier aggregation node.

The World Bank’s Maria case study attributes much of the telecommunications damage to extensive above-ground deployment and reports severe destruction of above-ground and last-mile fibre. GAO’s review of the federal telecommunications response found extensive damage to communications infrastructure and unprecedented restoration challenges. It also found that responsibilities and public communication around federal support were not sufficiently clear. The lesson for an exchange is operational: after a disaster, technical restoration is entangled with information quality, prioritisation and coordination among organisations.

The federal support programme now places those dependencies in explicit terms. USAC’s Bringing Puerto Rico Together programme page says supported carriers must maintain disaster preparation and response plans addressing infrastructure strengthening, network diversity, backup power, monitoring and emergency procedures. It also describes reporting for buried fibre, aerial plant, fixed wireless and miles of backup or redundant network. These obligations apply to funded carriers, not automatically to PR-IX. They establish a useful standard for the members and access providers on which the exchange depends.

NetWave’s six-node island backbone is another positive signal, but a list of nodes is not a route-diversity proof. To evaluate its PR-IX contribution, a member would need to know which entrance and aggregation path reaches the San Juan exchange site, what alternate site is available, whether failover is automatic, and how the alternate avoids common civil infrastructure. The same questions apply to every carrier selling access to Guaynabo or Bayamón.

Backhaul survival also determines who can benefit from local content. A cache in a powered metropolitan data centre helps only if the access provider still has a path to it and can deliver service through its last mile. Internet Society’s estimate that 21 per cent of popular sites have in-country instances describes a useful pool, not guaranteed emergency reachability. Some content may withdraw routes, disable a cache, exhaust capacity or become unreachable from affected access networks even while other exchange traffic flows.

The best resilience design is not necessarily a dedicated circuit to every PR-IX location. Cost, network size and traffic patterns matter. It is, however, a design in which the operator and member understand their common risks: at least two physically separated access paths where continuity warrants the expense, deliberate placement at distinct failure domains, documented failover, and the ability to operate if a preferred route server or facility is absent. Without that discipline, a multi-site exchange can be reduced in practice to one shared metro corridor.

Off-island diversity still decides what “local” can do

The purpose of an island exchange is not to eliminate off-island connectivity. It is to avoid using it when two networks can exchange appropriate traffic locally, and to make the remaining international capacity work more efficiently. Many essential destinations will still sit elsewhere: cloud control systems, software services, external news and government sites, identity providers, payment systems, remote data copies and content not cached in Puerto Rico. Even a vigorous local exchange cannot turn an island’s internet into a closed system.

The relevant resilience question is therefore two-part. Can local paths continue without an international detour, and can the island reach enough independent international routes for everything that cannot remain local? Failure in the first part wastes scarce submarine capacity and may isolate local services unnecessarily. Failure in the second part leaves a functioning local fabric connected to only a subset of the wider internet.

Puerto Rico’s government has documented the concentration it wants to reduce. The broadband programme’s five-year action plan maps multiple submarine systems but places the main landing-station ecosystem in the greater San Juan metropolitan area, identifying facilities such as HUB787, Telxius San Juan and other carrier sites. The document is useful at regional scale. It is not a survey-grade route map, and its lines should not be read as exact seabed or terrestrial paths.

More recent public investment treats landing diversity as unfinished work. A Puerto Rico Broadband Program procurement for a submarine-cable and landing-station study sought analysis for a more resilient and redundant system. The US Treasury’s Puerto Rico Capital Projects Fund page describes an $85.7 million submarine-cable resilience programme intended to add a route involving the Dominican Republic and US Virgin Islands and create three new landing stations on Puerto Rico’s coastline.

The Puerto Rican government’s 2025 performance report makes the status distinction especially important. It describes the desktop-study procurement and contemplated east, west and south coast landing locations, plus connections toward the Dominican Republic, US Virgin Islands and an existing northern station. It also describes later work for marine survey, cable manufacture, installation and dry-plant construction. Those are planning and procurement milestones. They are not proof that a new cable is installed, lit or available to carry emergency traffic.

Older systems remain physically relevant. An archived FCC notice concerning the Taino-Carib system identifies landing facilities at Miramar in San Juan and Isla Verde in Carolina, and describes ownership and operating interests. Because this is a filing record reproduced by a third-party archive, it is best used to establish historical landing arrangements, not current capacity or present operating condition. Telxius, meanwhile, describes its CELIA project as a new submarine connection between Puerto Rico and Boca Raton, Florida. A project description indicates intended route diversity; it should not be counted as usable capacity without a ready-for-service date and operating confirmation.

This distinction between route shown, route built and route usable also applies to the exchange’s continental facility entries. A PR-IX presence in Jacksonville or Oregon may offer valuable operational options, but a map pin off island does not show how a Puerto Rican member reaches it, which submarine system carries the traffic or whether two apparent paths converge at a landing station. True diversity must be traced from the member’s equipment through metro fibre, the exchange fabric, the landing facility, the wet segment and the far-end station.

A shared failure anywhere along that chain can defeat diversity that looks persuasive at a higher level.

Local peering still pays a resilience dividend before every international weakness is fixed. It reduces the volume that must traverse those paths, keeps suitable traffic close to users and gives Puerto Rican networks a common place to coordinate. Its contribution is greatest when caches, authoritative services and locally important applications remain reachable at the exchange. But that dividend should be measured as avoided dependency, not mistaken for independence.

The combined-failure audit is harder than a generator test

A hurricane rarely presents one clean fault at a time. The demanding case is a coupled sequence: commercial power disappears, one or more terrestrial routes are cut, a facility is difficult to reach, fuel demand rises, some members lose their access networks, and operators must work with incomplete information. PR-IX should be judged against that sequence because Puerto Rico’s recent history shows that its ingredients can arrive together.

In the first minutes, batteries and uninterruptible power systems carry equipment through the transition to generators. Some member sessions drop as remote nodes lose power or fibre. Routing converges around what remains. The exchange can look healthy in a global reachability view even though the set of usable local peers has contracted sharply.

During the first day, facility and carrier operations become the deciding layer. Staff verify generator load, cooling, fuel, optical levels and alarms. Members determine whether the lost route is their circuit, a shared carrier segment or the exchange port. If management connectivity depends on the failed production path, diagnosis slows. If the route servers are in one affected domain, multilateral routes may disappear even while bilateral peers could still exchange traffic.

By the third day, repair and replenishment compete for roads, vehicles, spares and people. The DOE Fiona report’s description of fuel queues matters because a generator’s nameplate capacity cannot guarantee a delivery. Sites with longer autonomy can absorb delay; sites with shorter autonomy need priority and a confirmed route. A member whose access node has only hours of backup may have vanished long before a well-protected exchange facility approaches its own limit.

Around the published five-day runtime at FiberX, the distinction between design and operation becomes stark. If grid service is still absent, continued operation depends on at least one successful refill or a managed reduction in load. The five-day figure is approximate and facility-wide conditions may alter it. More importantly, keeping FiberX powered would not prove that the other PR-IX access sites or the member circuits into Guaynabo were available. The resilience unit is an end-to-end peering path.

Federal records show why the outer scenario must remain severe. GAO’s account of 2017 grid restoration says the blackout lasted about 11 months for full restoration and that island logistics made crews and materials harder to mobilise than on the mainland. Its later review of Puerto Rico grid recovery found that the territory faced a complex, long-term rebuilding task after temporary and partial repairs. These findings concern the grid as a whole, but every exchange facility consumes that grid or substitutes an on-site power chain for it.

Public-safety planning also treats isolation as a first-order concern. An NTIA Puerto Rico public-safety broadband proposal describes the need for locally sustained capabilities when outside help may take time to arrive and warns that critical connectivity can be lost if supporting infrastructure is hit. A later FirstNet environmental assessment chapter for Puerto Rico describes the island’s hurricane exposure, difficult geography and reliance on imported petroleum moving through ports such as San Juan, Fajardo and Ponce. Neither document evaluates PR-IX. Together they explain why local staff, fuel and communications cannot be assumed to arrive on a mainland timetable.

A credible combined-failure exercise would therefore remove more than utility power. It would withdraw a site, impair one carrier corridor, make one route server unavailable, constrain refuelling and require operators to communicate through an alternate channel. The useful result would not be a flawless demonstration. It would be measured reconvergence, known surviving peers, documented fuel margins, clear authority and a list of common dependencies that can be reduced before the next storm.

The people affected are several networks away

PR-IX does not sell a household broadband line, dispatch a cellular tower repair crew or operate the island’s electric grid. Most Puerto Ricans will never see its name on a bill. Its failure can still reach them because an exchange sits between the networks that do have direct relationships with residents, businesses and public institutions.

For an internet service provider, loss of the local peering fabric can shift traffic onto paid transit or a longer private path. That may increase latency and fill an international link at the same time demand is rising and other capacity is unavailable. A provider can remain nominally online while delivering a slower, narrower and less predictable service. Small networks are especially exposed if they have fewer alternative interconnections or if their only exchange access follows the failed metro route.

For content and application networks, the exchange is a distribution point. A surviving cache can serve many users efficiently, but only if its host facility, exchange port and the access networks all remain connected. If the cache or its route is withdrawn, requests move to a remote origin. That consumes off-island capacity and can turn a modest local disruption into congestion visible across multiple providers.

Universities, hosting operators and technical organisations appear among the exchange’s published entities. Their traffic can include research, education, domain-name infrastructure, business systems and locally hosted services. Not every service is life-critical, and membership alone does not identify emergency priority. The wider point is that an exchange concentrates opportunities for local reachability among institutions whose users are distributed across the island.

Public services are affected indirectly. Emergency communications use dedicated systems as well as commercial networks, and no public record reviewed here establishes PR-IX as the sole path for a particular emergency service. It would be wrong to imply that an exchange outage automatically disables 911 or public-safety radio. But GAO’s telecommunications review describes how broad network damage complicated post-Maria restoration, while the World Bank report says callers struggled to reach emergency services and response teams had difficulty coordinating. Commercial internet reachability is one strand in a larger communications lifeline.

The consequences also vary geographically. PR-IX’s facilities cluster in the metropolitan area, while members may serve communities across Puerto Rico. A provider in the west or central mountains can benefit from local peering in San Juan only through its terrestrial backhaul. That circuit may cross more hazards than a metro member’s connection. An island-wide member count therefore does not imply equal disaster access from every municipality.

Failure can also split the network rather than erase it. Some peers may continue exchanging at one site while others are isolated. Users of one provider may reach an island-hosted service while customers of another cannot. IPv4 may converge differently from IPv6. Direct sessions may survive a route-server problem. This partial state is operationally difficult because broad “up” or “down” labels conceal who is actually connected.

The exchange’s most valuable public role during such a period would be precise status communication: which sites are forwarding, whether route servers are reachable, which maintenance windows are active and where members should move traffic. That information cannot repair a fibre cut, but it helps member operators distinguish exchange faults from their own failures and use scarce field resources intelligently. Clear communications are part of resilience because uncertainty consumes time.

The missing measurements define the next standard

PR-IX has enough public information to move beyond a binary question about whether it exists. The next standard should describe how the service behaves under stress without exposing sensitive security details. A useful disclosure would connect every high-level statement to an operating state.

At the physical layer, the exchange could identify which advertised access locations host its own switching equipment and which are remote extension points. It could state whether two sites depend on the same switching chassis, transport ring, utility feeder or facility operator. General route diversity could be expressed by independently verified failure domains without publishing exact fibre alignments. This would let members distinguish geographic spread from actual separation.

At the electrical layer, per-site disclosure could report the protected load class, UPS autonomy, generator arrangement, tested transfer interval, typical fuel autonomy and refuelling method. FiberX already shows how useful this can be. The remaining uncertainty is whether that design is current, what load underlies the runtime and how it relates to PR-IX equipment. Comparable facts from the other access locations would reveal whether a member can choose sites with genuinely different endurance.

At the network layer, PR-IX could identify route-server placement by failure domain, publish aggregate availability for IPv4 and IPv6, and distinguish installed port inventory from connected and usable capacity. A total traffic graph should define whether it displays average or peak throughput and whether all sites are included. During a disruption, a count of active member ports by site would be more meaningful than an autonomous system remaining visible somewhere.

At the recovery layer, the exchange could publish outcomes from periodic exercises: which site or route server was withdrawn, how quickly routes reconverged, whether out-of-band management worked, how members were notified and what dependency failed unexpectedly. The result need not disclose customer configurations. It should demonstrate that operating procedures have been exercised with facility and carrier partners rather than assumed from equipment specifications.

Capacity needs the same precision. “Up to 100 Gbps” is a product ceiling. The two 10 Gbps route-server connections in PeeringDB are specific ports. The 100–200 Gbps traffic band is a self-reported range. None is the total installed switching capacity, sold capacity, spare capacity or guaranteed emergency throughput. A switch may have enormous design headroom yet be unusable without power; a smaller surviving site may become constrained when traffic fails over. Installed, lit, powered, operational and usable are separate states.

The public data also needs dates. The official site’s facility and member lists, PeeringDB’s update timestamps, Pulse’s July 2026 aggregation and older disaster records answer different questions. A current status page can show what is intended to operate today, while a dated exercise shows what actually survived a controlled failure. Neither should be substituted for the other.

These are not demands for perfect transparency. Network operators have legitimate commercial and security constraints. They are the minimum distinctions needed to evaluate a resilience service responsibly. PR-IX’s present public signal is strong enough to justify deeper scrutiny, and the gaps are specific enough to be closed.

Puerto Rico IX is real infrastructure with an unproven outer limit

Puerto Rico Internet Exchange has crossed the first credibility threshold. Its operator names facilities and entities. AS18804 is registered to the company and appears as a route-server network. PeeringDB exposes dual-stack sessions, an open policy and a material traffic band. Packet Clearing House calls the exchange active. Internet Society reports one active Puerto Rican IXP with 28 members and 15 physical locations. These independent signals make “operating exchange” the best-supported conclusion.

The physical story is also more substantial than a single San Juan rack. The announced access surface reaches San Juan, Guaynabo and Bayamón. Facility and carrier descriptions show hardened rooms, generator systems, multiple fibre entrances and broader island networks. Public investment is moving toward more diverse submarine landing geography. These are the ingredients from which resilient local interconnection can be built.

But the hardest claim remains unproven: that members can continue to exchange local traffic through a prolonged island-wide power and transport emergency. No public account ties every PR-IX site to a current power-autonomy figure. No published topology shows that route servers and switches occupy independent failure domains. No member-path record demonstrates physically diverse backhaul into the fabric. No disaster exercise shows the surviving peer set after the simultaneous loss of grid power, a facility and a carrier corridor.

The five-day number captures both the progress and the limit. At FiberX, a disclosed N+1 generator design, fuel storage and estimated runtime provide a concrete basis for planning. Five days of autonomy could preserve valuable local connectivity through a shorter outage. In a longer emergency, the design depends on successful refuelling. Other PR-IX sites and member networks will have different clocks, many of them shorter and undisclosed. The first clock to expire on an end-to-end path decides whether that path survives.

PR-IX should therefore be treated as meaningful Puerto Rican infrastructure, but not as a self-contained hurricane shelter for the island’s internet. Its resilience contribution is conditional and potentially large: keep suitable traffic local, reduce avoidable pressure on submarine links, preserve access to local content and give networks a common interconnection surface. The conditions are powered facilities, intact and diverse member backhaul, surviving endpoints, operational route control, reachable staff and a replenishment chain that works under stress.

The strongest next step is evidence of coordinated operation across those boundaries. A per-site power and network matrix, a clear distinction between remote access and independent switching, aggregate traffic definitions, dated failover exercises and public incident status would transform a collection of credible signals into a demonstrable resilience case. The task is not to promise that every packet survives every hurricane. It is to show which local packets can survive the combined failure Puerto Rico has already experienced—and for how long.

That is the standard appropriate to an island exchange. Latency and port speed explain why networks join. The ability to keep a modest, useful local internet alive while the grid is dark, roads are constrained and off-island capacity is uncertain explains why the exchange matters to Puerto Rico.