Northern Dipper vs the North Star: The Changing Character of War

Part Seven · The deeper difference: how the two systems are built

For twelve days in June 2025, American jamming drove Iran's GPS-guided missiles into open desert. Eight months later the same jamming failed against the same arsenal, because the receivers had changed: Iran had moved onto the encrypted military signal of BeiDou, China's satellite-navigation system. The protections on that signal are one part of the story. The deeper difference between GPS and BeiDou is not in the signal at all, but in where their satellites orbit. GPS keeps to a single orbit; BeiDou spreads across several. That choice decides much of what follows: how fast a receiver can settle on a precise fix, how hard the signal is to jam, and how much the system can do beyond navigation.

Three orbital bands matter: low Earth orbit (LEO) a few hundred kilometers up, medium Earth orbit (MEO) around twenty thousand kilometers, and geostationary or inclined orbits some thirty-six thousand kilometers out.

• Low Earth orbit (LEO), roughly 300 to 1,200 km up. This is where Starlink and most spy satellites crowd. No navigation system operates here yet, though the first test satellites are now going up. A satellite this low crosses the whole sky in about ten minutes.
• Medium Earth orbit (MEO), around 20,000 km. This is where all four navigation systems live: America's GPS, Russia's GLONASS, Europe's Galileo, and most of BeiDou. A satellite up here takes hours to cross the sky.
• Geostationary and inclined orbit (GEO/IGSO), around 36,000 km. These satellites either hang over one fixed spot or trace a slow daily loop over a single region. Among the global systems, only BeiDou uses them.

That single choice splits the two systems. GPS lives at one altitude: a clean, single-shell design that does one job well. BeiDou spans three, with a low fourth on the way: global coverage from the middle layer, dense regional coverage and messaging from the high one, and soon a fast low layer. GPS bets on simplicity and three decades of track record. BeiDou bets on capability at the cost of complexity. Two different bets about what a navigation system should be.

Part Eight · What the layers buy

Orbits are architecture. What matters is what the architecture buys, and the layers pay off twice: in the speed of a precise fix, and in raw signal power against jamming.

Begin with the slow fix. The centimeter-grade position is the slow one: the receiver has to watch satellite geometry change to resolve the carrier-phase ambiguity. MEO satellites barely shift in the sky; that centimeter fix can take tens of minutes to converge today. Fine for a surveyor with a tripod; without base-station support the long convergence makes it impractical for fast-moving applications. Low-orbit satellites race across the sky, so the same fix can lock in seconds. That is the promise of BeiDou's planned LEO layer, due around 2029: centimeter precision, in motion, in real time.

GPS receivers can get centimeter fix today, with caveats. Two paths. A real-time-kinematic (RTK) base station within roughly twenty kilometers of the rover delivers a fix in seconds, the standard surveying technique: free if you have your own equipment or a public reference station nearby, useless if you're operating beyond that twenty-kilometer bubble or anywhere without friendly ground infrastructure. The other path is a paid commercial correction service like Trimble's RTX or Fugro's Marinestar, globally available, perhaps ten or fifteen minutes to converge. Galileo's High-Accuracy Service offers free, public, high-accuracy corrections; what no constellation does today is broadcast cm-grade corrections that lock in seconds, free, anywhere a satellite passes. That is what the LEO layer would do, and what nothing else does.

Then the power arithmetic. A GPS signal crosses twenty thousand kilometers and arrives at the ground below the receiver's own noise floor. The signal is recoverable only because the receiver knows the exact pseudorandom pattern to listen for. Drop the source to low orbit, around six hundred kilometers, and the same signal arrives roughly a thousand times stronger. That gain is altitude, not antenna aim. It lifts the whole signal everywhere the satellite passes, not over one chosen patch.

The thousand-fold gain has a sharp consequence for the jamming math. Jammer denial radius scales with the inverse square root of received signal power, so a thousand-times-stronger signal shrinks the denial radius by about thirty-three times. A single hundred-watt jammer that denies GPS across roughly fifty kilometers reaches barely a kilometer and a half against a LEO signal. But the more useful figure is area, because that is what a defender or attacker is actually trying to cover. Denial area scales with the square of radius, so the thousand-fold power gain shrinks denial area by — a thousand times.

GPS's answer is to do more from the orbit it already has. Its next block of satellites will carry the Regional Military Protection spot beam, power concentrated up to a hundred times stronger over one chosen region for a chosen time. It is real, and it will close the anti-jam gap at the MEO layer. But it is power squeezed from a single shell, one beam at a time. BeiDou's bet is to add a shell instead. One side keeps refining a fixed design. The other keeps building outward.

None of this is the end of the story. The layers don't just buy precision and resilience. They open the door to functions a positioning system was never designed to do: messaging, sensing, integrated kill chains run from one constellation. To see the full reach of what an integrated PNT platform becomes, look at what BeiDou is doing that GPS is not.

Part Nine · The kill web

Modern force comes from integration: from binding the pillars of combat into one working system rather than fielding each on its own. Positioning, communication, surveillance, decision, weapons. The question is who owns each piece, and how the pieces talk to each other. How the US assembles its kill chain, and how China does, are now visibly different, with operational consequences.

The US runs its kill chain across a constellation of separate programs. GPS (the Global Positioning System, the position-and-timing backbone) is operated by the Space Force. Imagery and signals intelligence come from an entirely different agency, the National Reconnaissance Office, with its own classification regime, its own procurement chain, its own culture. The two have to interoperate, but they were not built to. They were built to do different things, under different authorities, on different timelines.

Communications add another layer of separation. AEHF (Advanced Extremely High Frequency, the jam-resistant strategic comms satellite system) flies on its own constellation. MUOS, the Mobile User Objective System, handles narrowband mobile traffic on its own constellation, with its own terminals. On the ground, Project Maven processes satellite imagery into machine-tagged target data; JADC2 (Joint All-Domain Command and Control) is the umbrella program meant to fuse it all into a single operational picture. JADC2 is still maturing. Every line of communication between every program is a place where data has to cross an organizational boundary, a clearance threshold, a software-stack handoff. Each crossing takes time. Each is also a place where the system can fail.

BeiDou starts from a different premise. The same satellite that gives you a position can carry a message. Inside the navigation system itself, the same constellation, reached through the same chip that fixes the position, sends a short text, up to about a thousand Chinese characters regionally, relayed through geostationary-orbit satellites, twelve million messages an hour. It kept rescuers talking through the 2008 Sichuan earthquake when the ground networks went down, and by 2022 it was built into an ordinary Huawei phone. No other satellite navigation system does this. Comms and navigation share the same constellation.

The next layer is more ambitious: ISR, the work of finding and tracking targets from above. In the public record it surfaces mostly on constellations adjacent to BeiDou, run within the same national space enterprise, on three tiers that differ mainly in who is said to own them. The military tier is Yaogan (遥感), the PLA's reconnaissance series flown since 2006, optical and radar and electronic-intelligence satellites together; its naval sets fly in threes, fixing a ship by the timing of its own emissions. The civil tier is Gaofen (高分), the state's high-resolution earth-observation program, filed under agriculture, weather, and disaster relief, though its later satellites carry almost no public description. The commercial tier is Jilin-1 (吉林一号), more than a hundred sub-meter optical and video craft run by Chang Guang out of Changchun. Some platforms sit in geostationary orbit and stare without pause; of one, Yaogan-41, a CSIS study concluded in early 2024 that China was nearing the power to track any car-sized object across the whole Indo-Pacific, with no place left to hide. By the Pentagon's late-2025 count China flies more than five hundred ISR-capable satellites across these programs.

Because these constellations and firms overlap, coordinated within one national space enterprise, China crosses fewer of the organizational seams a US shooter has to cross to pull the same picture from separate agencies. The fusion is visible on the open market. During the 2026 war the Shanghai firm MizarVision combined satellite imagery, live flight-tracking, and shipping data into AI-read intelligence on US forces, publishing reports that mapped the aircraft parked at Prince Sultan Air Base and the Patriot batteries ringing al-Udeid, and sold the result. That is exactly what the United States indicts. Beijing presents Gaofen as civilian science and Jilin-1 as private enterprise; Washington reads them as one military fleet, under the doctrine of Military-Civil Fusion (军民融合), an explicit Chinese state strategy that Xi Jinping personally oversees. The State Department calls MCF the deliberate "elimination of barriers" between China's civilian and defense sectors, and on that reasoning the Commerce Department has put Chinese satellite firms on its export blacklist, treating a nominally commercial builder as a military supplier by another name.

The charge is not wrong. It is only incomplete, because the United States runs the same fusion and files it under procurement. The National Reconnaissance Office pays Maxar, Planet, and BlackSky billions for commercial imagery to serve the warfighter, and Project Maven turns that imagery into the targeting data that picks what to strike. There is no clean line between civilian and military on the American side either.

The difference is not kind but degree of command, and beneath it a difference of creed. The American system hands technology to private enterprise by design, trusting markets to build what the state merely buys. That faith produced SpaceX and the frontier labs, and it exacts a standing price: cost-plus contracts that reward overruns, a defense industry entrenched enough to lobby its own perpetuation, and firms the state does not own and cannot simply command. China has the arrangement Washington lacks: Military-Civil Fusion binds its firms to the state, and it's already largely built. Washington condemns it and covets it: its JADC2 program is the attempt to build it, wiring every sensor and shooter into one network. What it cannot do is command its firms into line the way Beijing can, so it coerces them. Elon Musk's private control of Starlink, shown when he declined to enable a strike off Crimea and threatened to pull Ukraine's funding, is the independence the state moved to fold into its own hands, paying for the service and buying a version it commands, Starshield. When Anthropic barred the government from turning its models on mass surveillance or autonomous weapons, it branded the company a national-security risk and cancelled its contracts. The charge against Beijing describes the very stack Washington is straining to assemble, firm by coerced firm, against the grain of the free-enterprise creed it will not abandon.

The 2026 war showed the parts coming together in the open, and the striking thing was how much of it Iran could simply buy. Through a Chinese firm's "in-orbit delivery" service, Iran's Revolutionary Guard acquired its own half-meter resolution spy satellite (the TEE-01B, supplied by Earth Eye Company for a reported $36.6 million), which photographed Prince Sultan Air Base in Saudi Arabia in the days before US aircraft flying out of it were hit. Chinese commercial AI imagery firms, some posting publicly, reportedly fed Iranian targeting against US bases across the Gulf, though the chain has not been independently confirmed. A sensor finds the target, software names it, navigation pins it, and a link passes it to a shooter. The old sequence of find, fix, track, target, engage stops being a chain run across organizational seams and becomes a single networked loop. The kill chain has not gone away. The handoffs have shrunk — fewer, faster, and closer to invisible.

None of which means the world abandons GPS. GPS is open, documented, monitored by independent bodies, with thirty years of aviation, finance, and telecom built on top of it. The United States switched off Selective Availability (the deliberate blurring of the civilian signal) in 2000, and built its later satellites unable to do it again. China has made no such commitment for BeiDou. Around six to seven billion devices run on GPS, against roughly two billion on BeiDou. GPS carries thirty years of public scrutiny and audit; BeiDou's resilience and transparency commitments are less established. Multi-GNSS receivers, increasingly the norm, often consume both. The interesting question is not "which is better?" but what each is built for, and who is on it.

Part Ten · What war is built on

Two countries are answering the same problem in opposite ways. America is hardening GPS: better anti-jamming, the regional spot beam, the long-delayed military signal finally reaching the troops. The architecture stays and the defenses improve. China is changing what the system is: a low layer for instant precision and a constellation that navigates, communicates, and watches at once. One side fortifies a finished design. The other keeps redrawing it.

The world is not choosing between them so much as taking both. Most phones already listen to all four global systems (GPS, BeiDou, Europe's Galileo, Russia's GLONASS) at once. In roughly 165 of 195 countries there are more BeiDou satellites overhead at any moment than GPS satellites, a function of constellation size and the extra layers China flies over Asia. BeiDou has reached the ground inside Chinese chips, cars, and ships, often as the default satellite-navigation source with GPS as the backup. Britain, France, and South Korea are funding terrestrial backup systems against the day the satellites fail. The United States has discussed one and built none.

But the satellites are only one visible piece of the system. Sustained attrition consumes stocks fast: Israeli and US interceptor magazines over Iran drained in days, Stinger missiles in Ukraine in months, artillery shells by the hundred thousand, drones spent like ammunition. The question that decides the next round is not what platforms a country fielded but how many it can keep making, and how fast. That capacity begins upstream of the assembly line, in the refineries and foundries that process the rare earths, gallium, germanium, and graphite that modern weapons run on. China dominates much of that processing today, often by wide margins. The exquisite platform on the runway is the end of a chain. The chain itself is the strategic position.

The same shift is showing up at the front end of the chain. Capabilities that used to require a superpower's defense complex (precision strike, standing surveillance, the sensor-to-shooter plumbing between them) are now being bolted together from civilian industry. Iran did not build a reconnaissance program — it assembled one from purchased pieces. It bought a spy satellite from a Chinese firm the way a company leases a server, and pulled targeting imagery from commercial outfits running AI over the take. Even an assembled program still required Iranian tasking, integration, and dissemination, but the most expensive components arrived bought rather than built. The drones that fill the skies over Ukraine and the Gulf roll off the same lines as consumer quadcopters and electric cars. Military power is rebalancing away from the exquisite platform and the exclusive program, and toward the depth of a country's industry: how much of a war it can build, and keep rebuilding, out of things it already makes.

Satellites are not the only window onto this contest. The same pattern shows up in places that have nothing to do with what is broadcast from space.

The American military continues to procure Ford-class supercarriers at well over $10 billion per hull. The Kennedy is in sea trials; the Enterprise and the Doris Miller are under construction; two more hulls are under review but not cancelled. The Chinese counter is layered: anti-ship ballistic missiles in the DF-21D and DF-26 families, fielded and exercised against ship-shaped targets in the desert, alongside a growing surface and submarine fleet and the maritime aviation around it. The clearest public demonstration of where the cost-exchange math can go arrived in April 2022, when the Russian flagship Moskva was sunk in the Black Sea by two Ukrainian Neptune missiles, on the order of a million dollars between them. The $750 million cruiser was lost. The figure omits targeting, ISR, training, and the damage-control failures that prevented salvage, but the order of magnitude survives the caveats. Ukrainian forces had been operating the Neptune for little more than a year. The Ford-class procurement continues.

The AI contest runs on the same lines. U.S. hyperscalers and frontier labs committed tens of billions of dollars to compute infrastructure in 2024, and many of the leading models remained closed weights behind paid APIs and subject to export controls. In December 2024 the Chinese lab DeepSeek released V3 with open weights under an MIT license, reporting roughly $6 million in marginal training compute. The figure is not a total program cost; it is the order-of-magnitude marker that mattered. On major AI developer platforms, Chinese open-weight models grew from about 1 percent of token traffic in late 2024 to roughly 30 percent by early 2026. On public benchmarks the gap between U.S. and Chinese frontier models was three to six months and narrowing. On deployment economics, Chinese APIs ran at a small fraction of U.S. prices per million tokens. The U.S. frontier still held the capability gap. China was winning the diffusion gap, and the diffusion gap was the one that scaled.

The conventional framing of the Taiwan question runs that Chinese anti-access denial economics make American intervention prohibitively expensive, which is meant to deter China. The denial economics run both directions. The same cost-imposition logic that raises the price of a Chinese assault on Taiwan also raises the price of an American defense of it, even with Japan, the Philippines, Australia, Guam, and the Taiwanese themselves in the line. The cost ledger then loads asymmetrically. American intervention is priced against a strategic interest. Chinese action is priced against what Beijing reads as existential. This is Schelling's dilemma: in the chicken game, the side that cannot back down wins. The two prices are not in the same currency.

American wargames model logistics, magazine depth, and escalation when they are run carefully. The public readouts that filter into doctrine and budget tend to compress: a strike package succeeds or fails, aircraft losses are tallied, long-range anti-ship missile inventories run out in days, the exercise produces a score. CSIS's own First Battle of the Next War shows that even a successful defense produces casualties and losses on a scale that the institutional response has not absorbed.

The Chinese strategic frame does not run on scores. It runs on time horizons that a clock-bound exercise cannot model, and on a tradition far older than the technology: Sun Tzu, formalized for the modern era by Mao in On Protracted War, now extended by PLA writing on system-destruction warfare. The American tradition (Mahan on sea power, the doctrine of decisive battle, the joint force closing on the enemy) runs the other way. Both are coherent. The shift in the character of war happens to favor the first.

The Thucydides trap is usually framed as a single conflict: a rising power, China, against an established hegemon, the United States. There is always a second trap at play, though: the one inside the hegemon. The American model of war and its various interests, like the military-industrial complex, are hegemonic, entrenched, and built on a notion of exceptionalism that has survived three decades of increasingly contrary evidence. It is in unavoidable conflict with the era the country is now entering. The United States has two contests in front of it. The first cannot be deferred, and as a result, the second can be deferred no longer.