The Dark Grid problem

Critical infrastructure failure is an intelligence problem.

Every grid, pipeline, water main, and installation fails the same way — not because the iron is inherently unreliable, but because the systems that operate it have no live picture of their own state, no useful prediction, and no autonomous response. Civilization is permanently one step behind its own plant.

The Dark Grid is not a utility problem. It is a civilization problem. The blindness that costs a desk four hours to find a feeder fault is the same blindness that lets a pipeline run as a leak while the monitor calls it normal, and the same posture that lets contamination travel a main for hours. Kyber exists to end that posture across every critical system.

4–8 hrs
Typical US restoration time for a standard distribution fault
$78B
Annual US outage losses to businesses and households; more than $150B globally
2,950
Municipal utilities and rural cooperatives with no path to modern grid software
42M
Americans those 2,950 operators serve, often on 30- to 50-year-old plant

Section 01 · Electric utilities

A grid no one can fully see — including the people who run it.

The United States power system serves about 145 million customers through more than 7,300 plants, roughly 160,000 miles of high-voltage transmission, and millions of miles of distribution. It was built over a century by thousands of organizations with incompatible records. The result is a network that is never fully known at a given moment.

4–8 hrs
Average restoration for a standard distribution fault. Storms stretch that to days or weeks.
$78B
Approximate annual US outage losses. A single major storm can cost a mid-sized utility $50–200 million in credits, overtime, penalties, and damaged plant.
SAIDI / SAIFI
Regulators score minutes and events per customer. A 30 percent improvement is worth tens of millions a year to a mid-sized utility — and poor scores cost rate cases and public trust.
507,000
Additional US workers needed in transmission, grid, and energy construction by 2030. Utilities cannot hire their way out. They need the crew they have to see more and wait less.
+224 GW
NERC has sharply raised North American summer peak-demand outlooks over the next decade — on the order of 200-plus gigawatts, well above prior-year cases — driven by AI data centers. Every gigawatt added to a blind grid raises the chance and the cost of the next failure.
2,950
Municipal utilities and rural cooperatives, serving about 42 million people, still lack modern grid intelligence. The systems that exist are priced for investor-owned utilities ten times their size.

Early self-healing programs at large IOUs — Duke Energy among them — have cut outage duration on the order of 25 percent with first-generation schemes. Kyber’s target, as an AI-native stack rather than an overlay on legacy OMS, is a 40 to 60 percent improvement over baseline. That is a design goal, not a published field result.

What Kyber does here

Atlas rebuilds live topology from the SCADA, AMI, and IoT a utility already owns. Command takes restoration from hours toward under sixty minutes with dispatch and switching on that twin. Sentinel isolates in under five milliseconds so most customers are back before a truck leaves the yard.

Continue in the platform →

Section 02 · Water & wastewater

The same blindness, in a different medium.

The United States has on the order of a million miles of distribution pipe. The average water main is about 45 years old; many large cities still run on pre-war cast iron. About 240,000 main breaks occur each year — roughly 650 a day.

20–30%
Share of treated water lost to leaks before a tap. Older urban systems can exceed 40 percent — water already pumped, filtered, and dosed, then lost to the ground.
$2.6B+
Annual US cost of main-break emergency repairs alone, before property damage, road repair, interruption, or hauled water.
4–8 hrs
Typical time from a break to detection on conventional monitoring. Discovery is often a customer with no pressure, or a street that starts to flood.
100,000
Residents affected in Flint, Michigan, in 2014. Conventional testing took months. A live quality network would have seen the corrosion signature at the first affected point in hours, not seasons.

EPA rules require contamination response plans. Most utilities still have no real-time detection — they rely on scheduled grabs days or weeks apart. Water topology is as poorly mapped as electric GIS. Crews routinely find reroutes, abandoned connections, and valves the records never caught. That is the Dark Grid, underground.

What Kyber does here

Atlas reads pressure, flow, and quality telemetry to map the real network and flag leak and contamination signatures. Command isolates the broken or contaminated segment and keeps pressure on hospitals and hydrants. Sentinel executes the physical close at the edge so the event does not travel the rest of the main.

Continue in the platform →

Section 03 · Military & defense

Installations that still respond like a municipal desk.

The United States military operates hundreds of installations worldwide. Most still take primary power from the civilian grid even when they are designated critical. A civilian feeder event can degrade or remove operational capability.

Metcalf, 2013
A physical rifle attack on a California transmission substation, not a cyber intrusion, took 17 transformers offline, caused about $15 million in damage, and took 27 days to restore. The FBI called it the most significant attack on the US grid to that point. It exploited the same lack of live awareness Kyber is built to close.
Colonial, 2021
Ransomware shut the largest refined-products pipeline in the country for six days. Fuel ran short on the East Coast, including at military sites. The ransom was about $4.4 million — not billions. The strategic lesson is larger than the payment: civilian monitoring failures become national-security events.
90 seconds
The 1989 Quebec geomagnetic storm collapsed the provincial grid, leaving six million people without power for about nine hours and causing roughly $2 billion in damage. Modern electronics are more exposed, not less. EMP, directed energy, and severe space weather are the same class of simultaneous, wide-area stress.

DoD policy asks installations to show energy resilience — the ability to island from the civilian grid. Most cannot, because the microgrid has no intelligence for autonomous operation. Base maps of power, water, fuel, and comms are as stale as a town’s GIS. When power fails on a base, the response is still walk-down, radio, and phone. In an operational environment that latency is the vulnerability.

What Kyber does here

The same three products, adapted — not a separate catalog. Atlas maps installation power, water, fuel, and comms as one graph. Command prioritizes medical, communications, and command load while the fault is isolated. Sentinel at the edge lets the microgrid act in milliseconds. The path in is the Defense Innovation Unit and installation microgrids, as on the rest of this site.

Continue in the platform →

Section 04 · Oil & gas

Hours of leak before anyone treats it as a rupture.

The United States has on the order of 2.6 to 3 million miles of oil and gas pipeline — enough to circle the Earth more than a hundred times — moving most of the crude and gas the country burns. A significant disruption shows up immediately in fuel, heat, and manufacturing.

17 hours
How long the 2010 Enbridge Line 6B rupture ran before it was treated as a release: 843,000 gallons of diluted bitumen into the Kalamazoo River. Cleanup passed $1.2 billion over four years. The monitoring system read a pressure drop as a normal operational swing.
~300 / yr
PHMSA-significant pipeline incidents in the US in a typical year: property damage, remediation, and lost product in the billions. A recurring root is late or misclassified anomaly detection.
$6B
Estimated annual global cost of pipeline theft. An unauthorized tap has a flow signature. A model trained on that signature can see it in live telemetry.
10–30 min
Typical delay before a third-party strike — the leading cause of incidents — shows up as a SCADA pressure change. By then product and ground are already involved.

Colonial in 2021 is the cyber case; Enbridge is the leak case. Both are the same posture: operators discovering the event after the damage is already priced. Kyber does not need a new story for each medium. The graph, the signature, and the isolate are the same loop.

What Kyber does here

Atlas watches pressure, flow, and acoustic signatures on the real topology and flags leak, tap, and strike patterns in seconds. Command starts shutoff, meets PHMSA notification, and sends inspection to a GPS span — not a county. Sentinel isolates the smallest section the iron allows.

Continue in the platform →

Section 05 · Transportation & smart cities

Failures cascade because the city has no shared twin.

A feeder outage kills traffic signals, which gridlocks the response, which worsens the original event. A main break floods a subway and closes the street above it. These are not exceptions. They are how a city without a unified intelligence layer actually runs.

20–40%
Increase in emergency response time when signals fail in a major weather event. For cardiac arrest, survival drops about 10 percent each minute without treatment.
55 million
People without power in the 2003 Northeast blackout — a cascade that began with a software and awareness failure in Ohio and crossed eight states and two provinces. Losses were on the order of $6 billion. From first fault to regional collapse was minutes, not hours.
$9B
Approximate annual US cost of stormwater flooding. Drain networks are among the least-watched urban systems.
$10B
Estimated annual US transit losses from infrastructure disruptions, many of them power faults to third rail or catenary — the same class Atlas and Command already treat on a feeder.

Urban heat plus aging plant puts peak demand on the oldest laterals. Those events write the longest outage histories in city utilities. The desk still coordinates across departments by phone.

What Kyber does here

Atlas builds one twin from the sensors the city already has — power, water, traffic, transit, comms. Command coordinates the response across departments on that twin instead of a conference bridge. Sentinel holds the physical edge so a feeder event does not take the intersection, the hospital, and the rail at once.

Continue in the platform →

Section 06 · Data centers

Density that did not exist five years ago, watched with tools that did.

A single AI training cluster can draw 10 to 20 megawatts in one hall — a small city’s load, in a few rooms. That density creates failure modes existing BMS and EPMS stacks were not built to catch.

$300K–$5M
Cost of one hour of hyperscale downtime, depending on facility and workload. Transaction processing can price a minute over a million dollars.
30–50%
Share of planned US data-center builds for the mid-decade window that have been at risk of delay or cancel because the grid cannot interconnect them. Existing supply to live halls is therefore more precious, not less.
30–40%
Facility power spent on cooling. In AI halls, a thermal miss can destroy millions of dollars of hardware in minutes. Most monitoring still alarms after damage has started.
PUE ~1.58
A common US average: 58 percent of facility power is overhead. Live distribution awareness has shown 10 to 20 percent PUE improvement in published efficiency programs. That is the class of gain, not a Kyber field study.

Operators cannot wait years for the next interconnection and also accept blind hours on the feeders they already have.

What Kyber does here

Atlas maps hall power and cooling as a graph and flags the signatures that precede a trip or a thermal run. Command sheds or reroutes before the path overloads. Sentinel at the distribution edge isolates a fault in milliseconds and keeps the rest of the hall up.

Continue in the platform →

Section 07 · Space & satellite

The least-watched infrastructure the modern world cannot do without.

GPS, transaction timestamps, cellular sync, military comms, and missile warning all ride satellite systems whose ground networks are often monitored with tools a decade behind what Kyber builds for a municipal feeder.

1989 / 1859
A Quebec-class geomagnetic storm is moderate in the historical record and still took down a provincial grid. A Carrington-class event would put a large share of today’s on-orbit and ground infrastructure at risk of simultaneous, years-long loss.
Space Force, 2019
Stood up because space is a warfighting domain. Jamming, spoofing, and physical interference with ground stations are established adversary capabilities. Ground resilience is a requirement, not a slogan.

Missed contact windows are often equipment and power events on the ground, not spacecraft events — the same Dark Grid, at the antenna.

What Kyber does here

Atlas twins the ground-station network and the state of the links. Command keeps contact continuous by failing to an alternate station when the primary degrades. That is the same loop as a feeder transfer, applied to Space Force and commercial constellations alike. No separate product name. The same platform.

Continue in the platform →

The rest of the cost is not on a feeder map.

  • Cybersecurity

    Every operational failure can be a security event. NERC CIP exists because the grid is a target. Kyber is built to produce the continuous monitoring those rules already ask for — not a bolt-on after the OMS. The inability to tell a failed insulator from a induced fault is itself a hole. Topology reconstruction closes it.

  • Workforce

    The United States needs hundreds of thousands more people in grid construction, operations, and maintenance this decade. They are not available at that scale. Kyber is not a replacement for a lineworker. It is a force multiplier: less time synthesizing five systems, more time on the span that is actually open.

  • Climate

    The costliest US hurricane seasons cluster in this century. Wildfire liability from grid-caused fires has already produced judgments and a historic utility bankruptcy. Plant built for a 20th-century climate is being run in a 21st-century weather file without the intelligence to adapt. Every event shortened has a rising dollar value.

  • Equity

    Outages last longer in lower-income neighborhoods. Backup generation is rarer. The plant is older. Sentinel Edge and Micro are how the same intelligence reaches a dialysis clinic or an apartment lateral, not only a downtown network. When isolation holds a clinic through a storm, that is a life, not only SAIDI.

  • Insurance and liability

    PG&E’s 2019 bankruptcy — the largest utility bankruptcy in US history — was wildfire liability from grid-caused fire. Insurers want live risk, not a last-year GIS. Atlas telemetry and anomaly scoring are the kind of feed that changes how infrastructure risk is priced. That is a path, not a product we sell today.

The close

Every number on this page is a failure that did not need a new sensor — it needed intelligence.

Kyber is not an incremental patch on systems that have already failed the people who depend on them. It is the layer that should have existed from the first energized feeder: awareness, autonomous response, and healing without waiting for a complaint call.

The telemetry is already there. The crews are already hired. What was missing is the architecture that turns both into action in real time. Atlas sees. Command decides. Sentinel executes.

The infrastructure of civilization does not have to keep operating blind.