Starlink reconnected La Guaira. The bill came in watts.

On June 24, 2026, at 6:04 and 6:05 p.m. Venezuela time, two earthquakes of magnitude 7.2 and 7.5 tore through the state of Yaracuy 37 seconds apart^[3]. The epicenters lay some 250 kilometers from Caracas — near San Felipe and Yumare, not near the coast. And yet the two most devastated areas were La Guaira and Caracas: not by proximity, but by accumulated vulnerability — soils, building stock, decades of deferred maintenance.

Inside the US$37 billion in direct physical damage the UN estimates there is a line almost nobody read out loud: telecommunications, at roughly US$5 billion, was the single largest infrastructure loss, ahead of power generation and ahead of roads^[2]. UNDRR itself warns that the figure comes out of risk models rather than field inspections, and that it excludes losses from interrupted services; it is worth quoting with that label attached.

The figure invites an intuitive and mistaken conclusion: that the country was cut off and the satellite rescued it. What actually happened is more interesting, and more useful.

The curve

The International Telecommunication Union and the UN telecommunications cluster activated a Disaster Connectivity Map on June 25. It measured this: national connectivity fell from above 90% to roughly 59–65% immediately after the quakes, and recovered to close to 77% within 24 to 48 hours^[1].

Two days. That is what the Venezuelan network took to recover most of what it had lost, and it did not do so by satellite: CANTV, Movistar and Digitel did it by repairing. The report is explicit that the core networks never stopped operating^[1]. What broke was the perimeter — power systems, fiber optics, mobile installations, transmission equipment — plus the submarine cable, severed 1.8 kilometers from the Puerto Viejo landing station in Catia La Mar^[11].

But the curve never climbed again. As of July 13 — three weeks later — La Guaira was still the worst-affected area in the country, while Caracas had recovered in essentials^[1]. The national network stabilized quickly on a lower step, and there it stayed. It is on that step, and only there, that Starlink comes in.

What Starlink did, exactly

By July 9, more than 1,600 Starlink kits had been distributed in Venezuela^[9]. But the detail almost nobody reported correctly is what they were used for: not as home internet, but as backhaul — the link that connects a cell site to the rest of the network, standing in for the destroyed fiber and microwave links^[10]. Three Movistar base stations came back up on satellite transport; as of July 17, 9 of La Guaira's 11 parishes had temporary satellite backup^[10].

The distinction holds for everything that follows: an antenna connects a node, not a population. Where there was direct service to people was in the shelters, and at modest scale: the cluster installed Starlink at Playa Grande and at the César Nieves stadium camp, serving some 1,850 people across the two sites^[1].

And it did not install antennas. It installed antennas with solar panels, backup batteries and uninterruptible power supplies^[1]. The cluster's minutes record that the power backup was mounted explicitly “to mitigate the effects of the recent power cuts,” and that three weeks after the quake the World Food Programme was still buying Starlink equipment^[12].

The regulatory back door

Here is the part worth reading slowly, because it contradicts the easy headline. Starlink has no license to operate in Venezuela — neither a Habilitación Administrativa nor a spectrum Concesión, which is what the Ley Orgánica de Telecomunicaciones requires. It did not have one before the quake and does not have one today, even though it has been selling kits commercially in the country since February 2026^[4].

What was authorized after the earthquake was two things, and neither one is a license:

• June 26: CONATEL granted temporary spectrum at 1900 MHz to Movistar — not to Starlink — to provide direct-to-device satellite service over Starlink's network. Three months, geofenced to the state of La Guaira alone, on the condition of sharing capacity with Digitel and Movilnet, and at first for SMS only^[6][7].
• July 3: CONATEL gave Starlink a temporary “pilot test” permit and relaxed the power flux density limits in the Ku and Ka bands. The announcement sets no duration and clarifies that satellite telecommunications are an important technical facility “provided they are implemented under the Venezuelan regulatory framework”^[5].

And the justification with which the Venezuelan regulator opened that door deserves to be written down exactly as it stands: to fulfill “the duty to guarantee technological sovereignty and the people's right to communications”^[5]. That is: technological sovereignty was invoked in order to authorize, on a trial basis and without a license, the largest private constellation on the planet.

Meanwhile, and this is the strangest thing in the file: the importation of telecommunications equipment remains suspended by CONATEL, save for express authorization from the Foreign Ministry, under procedures that the UN's own customs bulletin describes as “not yet clearly defined”^[8]. Thousands of terminals belonging to an unlicensed operator were deployed while the importation of telecommunications equipment was formally closed.

The wear is already showing

On July 14, in the cluster's global call, the sentence that best sums up the balance was put in writing: “Starlinks have been deployed, but some congestion has been reported due to private connections adding to the humanitarian users. Even so, it remains the preferred solution”^[12].

Both halves matter. It is the preferred solution — and it is already congested by private users competing with the humanitarian ones, three weeks after the quake. The free service Starlink announced for Venezuela expires on July 25^[13].

The size of the thing we are tying ourselves to

It helps to size up the actor. As of July 18, 2026 there are 10,828 operational Starlink satellites in orbit, out of 12,552 launched^[14]. The network went from roughly 40 Tbps at the end of 2022 to roughly 445 Tbps at the end of 2025, serves 12 million subscribers in some 166 countries, and billed US$11.4 billion in 2025^[14]. Latin America contributes close to 2 million connections and grew 114% in 2025.

What is coming is of another order: the V3 satellites promise 1 Tbps of downlink each — some ten times the V2 mini — and roughly 60 Tbps of added capacity per Starship launch. With a caveat that has to be stated in the present tense: as this edition closes, no V3 has reached orbit; Starship's flight 13 aborted at T-0 on July 16^[15].

And one number that orders everything above: the analyst Tim Farrar estimates that the current network supports around 20 million broadband subscribers^[16]. With 12 million already connected, Starlink sits at roughly 60% of its own ceiling. V3 and Starship are not ambition: they are necessity.

Disclosure: Since July 19, Grupo Causa Común has been running a campaign of free access to Mentor Hispano for students in La Guaira affected by these earthquakes. We are analyzing a disaster in which we have an active product; saying so struck us as more honest than leaving it out.

Those are the facts. What they reveal about how this region's infrastructure gets built, powered and repaired is the business of the section that follows.

The quakes did not break the logical architecture of Venezuelan telecommunications: the core networks kept operating^[1]. What gave way was everything that holds them up physically — power, fiber, transmission equipment, the submarine cable. For anyone who plans, operates or builds infrastructure, that is the lesson that orders all the others: resilience is being bought at the wrong layer. You pay for redundant routes and mesh topology, and what fails is the current in the last kilometer.

The curve says it: from above 90% to ~60%, back to 77% within 24 to 48 hours, and flat thereafter — three weeks later La Guaira was still the worst-affected area in the country^[1]. The first stretch recovers almost on its own: those are the big nodes, with their own power and staff nearby. The expensive work is the tail, and it coincides with the most damaged and the poorest areas. A plan measured in national percentage declares victory right before the hard part begins.

The second finding is one of sizing. A cell site's battery bank is calculated for outages of hours, not days, and generators sit only at the critical sites. With national generation already below 40% of its installed capacity before the quake^[18], that assumption had been false for years. The response confirms it: the cluster deployed not antennas but UPS units, solar panels and generators, and it noted that the generators were what failed most^[12]. Sustaining a link is a problem of watts before it is one of bandwidth — 75–100 per terminal, ~1.6 kWh a day^[17] — and in Venezuela there is something no spreadsheet captures: the power backup of an unattended station is itself the target of the theft.

Out of that comes the question that matters for what is next, because reconstruction is going to demand more energy than there was before the quake, not less: hospitals, construction, pumping, telecommunications. Restoring centralized generation and transmission is a program of years. Between what the grid delivers today and what reconstruction is going to ask of it, an immediate gap opens, and the only resource that deploys at that speed and that granularity is solar with storage: it is installed site by site rather than network by network, in days rather than years, and it removes the fuel logistics chain — precisely where the response got stuck^[12]. Just as the satellite is a crutch for the link and not a backbone — a LEO network covers between 0.9% and 14.7% of the capacity lost when a cable goes down^[19] — the panel is a bridge and not a substitute for generation. But a bridge that can be laid now, and one that turns the blackout from an exception into a design assumption.

Where does AI come in without selling smoke? Not in predicting the quake, but where the scarce resource is physical and the decision is one of allocation. Energy-aware orchestration: which sites are going to go down, given battery, load and population served, so as to degrade on purpose and make the ones holding up the most people last the longest. Dispatch of crews, fuel and spare parts under real constraints of access and safety — and, by the same arithmetic, where to put the next panel to protect the most people per dollar. And allocation of scarce capacity on the emergency links, the congestion the cluster documented when private connections began competing with the humanitarian ones^[12]. None of that generates a watt or lays a meter of fiber: it makes every watt and every crew go further, which is the only thing left to optimize once the infrastructure is already on the ground.

THE PLAY

1. Measure sites in hours of autonomy, not in availability. A fleet at 99.9% with four hours of battery does not exist on the second day — and today the fastest way to buy autonomy is solar with storage, site by site.
2. Plan for the tail. The 77% arrives on its own; the stretch that remains takes the budget, the fuel and the crew hours, and it decides how many people stay cut off and for how long.
3. Put AI where the scarce thing is allocated, not where the inevitable is predicted. Energy, crews, emergency capacity and solar siting are allocation under constraint, with data that already exists and measurable returns. There is a product there; in earthquake forecasting, there is not.