On 7 August 2026, more than 180,000 water accounts across San Juan and six surrounding municipalities went dry, and the pipes had nothing to do with it. The Puerto Rico Aqueduct and Sewer Authority, known as PRASA, shut the taps off on purpose, in rotating 48-hour blocks, cycling through neighbourhoods until reservoir levels recover enough to restore normal supply.
The trigger was a punishing drought. July 2026 was the driest month on record for San Juan in more than 120 years of measurement, and by early August close to 68 percent of the island sat in some level of drought, roughly a quarter of it severe. Governor Jenniffer González called the situation "out of our hands," pointing to the weather, and on the rainfall itself she is not wrong. Nobody in San Juan chose this dry season. What she left out is what happens to the water once it does fall.
That answer sits underground, in pipes that lose an estimated 59 to 65 percent of every gallon PRASA treats before it ever reaches a customer's tap. That figure is not new. PRASA has been losing water at roughly this rate for years, drought or no drought, and it is the part of this crisis that AI-driven infrastructure can actually do something about.
The Rationing Schedule Puerto Ricans Are Living With
PRASA's rotating cuts affect San Juan along with the municipalities of Carolina, Juncos, Gurabo, Trujillo Alto, Canovanas, and Loiza. The pattern is simple and brutal: water for 48 hours, then none for 48 hours, repeating on a fixed schedule expected to run through at least the end of August 2026. Residents describe filling every container they own during the "on" window, then rationing it through the "off" window for cooking, bathing, and flushing toilets.
The Federal Emergency Management Agency has stepped in with water trailers, distributed for an initial 30 days, and Puerto Rico's Public Housing Administration has begun installing community cisterns at housing complexes. The island's Fire Department is delivering potable water directly to some residents. These are real, necessary responses to an emergency already underway. None of them touch the structural problem sitting beneath the streets.
The Number That Explains Everything
Utilities call it non-revenue water: treated water that PRASA pays to pump, purify, and pressurise, then loses before it reaches a paying customer through leaking pipes, faulty meters, and ageing distribution lines. PRASA's rate sits at an estimated 59 to 65 percent. For every three gallons the utility treats, roughly two never reach a household tap or a bill.
Put that number next to the rationing schedule and the crisis reads differently. San Juan's problem isn't merely a shortage of rain; it's a system that throws away most of the rain it does manage to catch and treat. Recovering even a third of that loss would free up enough water to ease the pressure behind the current 48-hour cuts, without a single new reservoir, without waiting on the next storm season, and without asking one more household to cut back.
Until February 2026, PRASA operated its core infrastructure without a SCADA system, the supervisory control and data acquisition software that gives a utility automated, real-time visibility into pressure, flow, and failures across its network. A $15.8 million federal funding package finally brought that monitoring layer online this year. Before it arrived, PRASA found leaks the way utilities did a century ago: roughly 16 leak detectives using ultra-sensitive listening equipment, backed by around 100 field brigades of 12 workers each, walking pipe routes and testing sections by hand. A leak could run for weeks, sometimes longer, before a human being happened to find it.
What AI-Powered Leak Detection Actually Changes
The tools that fix this problem are not experimental. Water utilities in drought-prone regions elsewhere already run them at scale, and the underlying approach translates directly to a network the size of PRASA's.
Continuous Acoustic Monitoring
Fixed acoustic sensors clipped onto pipes listen continuously for the specific sound signature a leak makes, a signature distinct from normal flow noise. Machine learning models trained on large libraries of both leak and non-leak acoustic data learn to flag the anomaly automatically, at a scale no team of 16 detectives with handheld listening equipment could match walking pipe routes on foot. Instead of a human ear checking a section of pipe once every few months, the network reports on itself around the clock.
Pressure Anomaly Analytics
A sudden or unexplained drop in pressure across a district metering area is one of the earliest, most reliable signs of a new leak, often visible in the data hours or days before any visible sign appears above ground. AI models trained on a network's normal pressure patterns can flag a deviation immediately and narrow the likely fault to a short stretch of pipe, turning a search across an entire neighbourhood into a targeted repair job for a single field crew.
Satellite and Aerial Loss Mapping
Satellite imagery and thermal or radar-based aerial surveys can detect the subsurface moisture signature of a significant underground leak across large areas of pipe network in a single pass, without digging a single test trench. Paired with the acoustic and pressure data from the ground, this gives a utility a prioritised map of where to send crews first, focused on the leaks losing the most water rather than the ones a detective happens to walk past.
The Core Argument
PRASA cannot make it rain, but it can find where 59 to 65 percent of its treated water disappears before anyone gets billed for it, and AI-based leak detection is the only approach fast enough and thorough enough for a network this size. Rationing buys time. Fixing the leaks buys the water back for good.
The $580 Million Bet on Smart Metering
FEMA has approved a $580 million award to replace roughly 1.4 million mechanical water meters across Puerto Rico with smart ultrasonic meters connected through an Advanced Metering Infrastructure network. That is a serious commitment of federal money, and it matters because ultrasonic meters are more accurate than the mechanical meters they replace and report consumption continuously rather than once a month.
But hardware alone does not recover a gallon of water. A smart meter that simply replaces an old meter with a more accurate version of the same monthly bill is an upgrade, not a fix. The value only appears once the continuous data those meters generate feeds an analytics layer built to act on it: models that flag a customer-side leak the moment consumption spikes overnight for no reason, catch a stuck or failed meter before it silently under-reports for months, and spot a district-wide flow anomaly the same day it starts rather than at the next quarterly review.
PRASA's fiscal plan already frames the metering rollout as a route to more accurate billing, which is reasonable as far as it goes. The larger opportunity is treating 1.4 million new data points as a real-time sensor network for the entire island's water system, feeding directly into the same AI models doing acoustic and pressure-based leak detection upstream. Skip that analytics layer and the $580 million buys better bills, nothing more. Add it, and the same investment becomes the closest thing PRASA has ever had to a live map of exactly where its water goes missing.
Why This Is a Caribbean Story, Not Just a Puerto Rico Story
Non-revenue water loss on this scale is not unique to Puerto Rico. An Auditor-General's report reviewed by the Jamaica Gleaner in 2026 found that Jamaica's National Water Commission loses between 40 and 60 percent of the water it treats, a figure the utility has been working to reduce through a district metering partnership in Kingston and St Andrew since 2015. Utilities across smaller island states, working with similarly ageing pipe networks and limited monitoring budgets, face the same structural exposure.
What makes 2026 different is that the drought hitting Puerto Rico is not an isolated weather event. It sits inside a wider pattern of El Nino-linked dry conditions that has also pushed Jamaica into an August 2026 water prohibition order across seven parishes, driven up produce prices at Kingston's Coronation Market, and put drought response costs across the region into the hundreds of millions of dollars. Every Caribbean territory carrying a high non-revenue water rate into a drought year is one dry season away from its own version of Puerto Rico's 48-hour rationing schedule.
That is the pattern regional utilities and CARICOM governments should take from Puerto Rico's crisis: the drought triggers the emergency, but an unrepaired pipe network is what turns a bad dry season into a territory-wide one. A utility that instruments its network with AI-based leak detection and demand forecasting before the next dry season hits can usually absorb a bad year without resorting to blanket cuts. One that waits, the way PRASA largely did, ends up building that same monitoring layer in the middle of an emergency instead, at higher cost, under public pressure, with FEMA trailers parked outside housing complexes standing in for a fix that should have started years earlier.
Beyond Leaks: Forecasting the Next Drought Before It Arrives
Leak detection recovers water already being lost. A second layer of AI application, drought and demand forecasting, helps a utility manage the water it has left. Models combining satellite soil-moisture data, rainfall forecasts, and reservoir drawdown trends can flag a developing shortfall weeks before a utility's board is forced into an emergency rationing announcement, giving planners time to phase in conservation measures gradually rather than switching an entire municipality to 48-hour cuts overnight.
The same forecasting layer supports smarter rationing design when cuts do become necessary. Rather than a single blanket schedule applied uniformly across San Juan, Carolina, and five other municipalities, demand models built on the smart meter data PRASA is now installing could identify which district metering areas can tolerate longer intervals between supply and which, home to hospitals, schools, or vulnerable populations, need priority restoration. Data does not eliminate rationing; it just makes the rationing smarter, cutting the avoidable hardship for roughly the same volume of water saved.
What Should Happen Next
Three actions would move Puerto Rico, and the wider Caribbean watching this crisis unfold, from managing a shortage to closing the gap that caused it.
First, PRASA should pair its new SCADA system and $580 million smart meter rollout with AI-based leak detection and anomaly analytics from day one, rather than treating the hardware as a standalone billing upgrade to be analysed later. The data infrastructure being built right now is the same infrastructure a functioning leak-detection system needs. Building it once, designed from the start to serve both purposes, costs far less than retrofitting analytics onto a finished metering network later.
Second, Caribbean utilities carrying non-revenue water rates above 40 percent, which includes Jamaica's National Water Commission among others, should treat Puerto Rico's 2026 crisis as an early warning rather than a distant news story. The instrumentation that would have let PRASA head off some portion of this drought's severity, acoustic sensors, pressure monitoring, satellite loss mapping, costs a fraction of what an emergency rationing response and federal disaster funding ultimately requires.
Third, CARICOM member states and Puerto Rico's own utility planners should share what they learn as this rollout proceeds. A region facing the same El Nino-linked drought pattern, the same ageing colonial-era pipe infrastructure in many cases, and the same limited public utility budgets has little reason to solve this problem in isolation, island by island, at full cost each time.
The Choice Underground
Nobody could have stopped this drought, and Puerto Rico's rationing schedule is, at bottom, a response to weather nobody controlled. The 59 to 65 percent of treated water disappearing into leaking pipes is a separate problem entirely, years in the making, built out of ageing infrastructure and monitoring budgets that never kept pace with the network they were supposed to watch. Unlike the drought, this part is fixable, and faster than most people assume: AI-based leak detection deployed at the scale PRASA's network needs could start recovering real volume within months, not the decade it took to get here.
The $580 million now going into smart metering is a start, not a solution. It becomes one only if PRASA builds the analytics layer alongside the hardware instead of treating it as an afterthought, because the alternative is watching this same argument play out again the next time the reservoirs drop.
Frequently Asked Questions
Why is Puerto Rico rationing water in August 2026?
PRASA began rotating 48-hour water rationing on 7 August 2026 after reservoirs fell sharply during a record drought. More than 180,000 customer accounts across San Juan and the municipalities of Carolina, Juncos, Gurabo, Trujillo Alto, Canovanas, and Loiza lose water supply for 48 hours at a time, then have it restored for 48 hours, on a rotating basis. July 2026 was the driest month on record for San Juan in more than 120 years, and close to 68 percent of the island was in some level of drought by early August.
How much water does PRASA lose to leaks before rationing even started?
PRASA loses an estimated 59 to 65 percent of the water it treats before it reaches a paying customer, through leaking pipes, faulty meters, and ageing distribution lines. For every three gallons PRASA pumps and treats, roughly two never reach a household tap or a bill. That gap existed long before the current drought and is the structural reason a dry summer turned into a territory-wide emergency.
How does AI-powered leak detection work on a water network?
AI-powered leak detection places acoustic sensors and pressure monitors along a pipe network and feeds their continuous readings into machine learning models trained to recognise a leak's sound and pressure signature against normal flow. Instead of a technician walking pipe routes with a listening device, the network reports its own anomalies in near real time, narrowing a leak's likely location to a short stretch of pipe rather than a whole neighbourhood.
How is PRASA currently finding leaks in Puerto Rico?
PRASA relies on roughly 16 leak detectives using ultra-sensitive listening equipment and around 100 field brigades of 12 workers each, walking and testing pipe sections by hand. Until a $15.8 million federal funding package arrived in February 2026, PRASA operated its core infrastructure without a SCADA system, meaning it had no automated, real-time monitoring to flag pipe failures as they happened.
What is the $580 million smart meter programme in Puerto Rico?
FEMA approved a $580 million award to replace roughly 1.4 million mechanical water meters with smart ultrasonic meters connected through an Advanced Metering Infrastructure network. These meters report consumption continuously, giving PRASA the raw data an AI system needs to spot a leak or an unusual drop in flow within hours instead of a billing cycle. The hardware alone recovers little water; the value comes from pairing it with analytics that act on what it reports.
Could AI have prevented Puerto Rico's water rationing?
AI could not have prevented the drought itself, but a fully instrumented network with AI-driven leak detection and demand forecasting in place years earlier would likely have meant rationing a smaller, better-targeted volume of water rather than blanket 48-hour cuts. Recovering even a third of the water currently lost to leaks would free up enough treated water to materially ease the pressure driving the current rationing schedule.
Is this a Puerto Rico problem or a wider Caribbean problem?
Non-revenue water loss on this scale is common across the Caribbean. Jamaica's National Water Commission has reported losing between 40 and 60 percent of the water it treats, according to a 2026 Auditor-General's report, and utilities across smaller island states face similarly ageing pipe networks with limited monitoring. Puerto Rico's crisis is the most visible current example of a pattern that applies across the region.
What should other Caribbean utilities learn from Puerto Rico's crisis?
The clearest lesson is sequencing. Puerto Rico is installing smart meters and SCADA monitoring during an active emergency, the most expensive and least efficient time to build infrastructure. Caribbean utilities that instrument their networks with acoustic sensors, pressure monitors, and AI-based anomaly detection before a drought hits can recover lost water gradually, at lower cost, and use the same data to plan rationing by district rather than announcing blanket cuts once reservoirs have already dropped to crisis levels.