Hook: The day’s crown digest flashed a news item about wildfire threats near the MDSCC complex in Madrid—the very same Spanish site through which NASA catches telemetry from Perseverance, ESA’s Mars rovers, and deep-space probes. At first, I nearly scrolled past—wildfires, Spain, summer, business as usual. Then I remembered something I’d read a couple of years back: NASA has exactly three deep-space communication sites across the entire Earth, and each has precisely one 70-meter antenna. Three. Not thirty. Three. And one of them, DSS-43 near Canberra, went completely offline in 2020—engineers discovered wear in one of the azimuth bearing’s raceways, and repairs took 11 months because the bearing wasn’t some off-the-shelf bolt; it was a precision-engineered component weighing several tons, custom-fabricated to order. And in the curiosity/ archive—not a single deep dive on this story. It deserves one. Because all of crewed spaceflight, every interplanetary mission of the last sixty years, all communications with Voyager 1 (now 24 billion kilometers away, with signals taking 22 hours one-way)—all of it runs through 13 antennas scattered across three deserts. And if Robledo de Chavela burns tomorrow, Mars loses its phone. Literally.
Topic: All of humanity’s interplanetary communications flow through the Deep Space Network (DSN)—a system of three complexes in California (Goldstone), Spain (Robledo de Chavela, near Madrid), and Australia (Tidbinbilla, near Canberra). Each site has one 70-meter “big dish” (DSS-14, DSS-43, DSS-63) and several 34-meter antennas. They’re spaced exactly 120° apart in longitude—so that any mission, at any moment, is visible from at least one site (Earth rotates, and when one station dips below the horizon, another rises). It’s an engineering masterpiece from the 1960s, still operating with minimal changes. But over those 60 years, the number of active missions has grown 5–8×, each mission demands orders of magnitude more data (new cameras, spectrometers, laser comms), and the antennas? The same ones. By 2024, the aerospace community finally admitted what JPL engineers have known for years: the DSN operates in oversubscription mode—30–40% more requests than it can physically handle. Meaning every third request from a mission—“give me a window to talk to Earth”—gets denied. And this isn’t a bug; it’s an architectural constraint that can’t be patched—new antennas take 7–10 years to build. The kicker? JPL, under the radar, through a partnership with D-Wave and NASA Ames, is teaching quantum annealing to solve the DSN scheduling problem—because classical algorithms can’t keep up. In other words, a quantum computer is, for the first time in history, not trying to “outpace” classical machines but simply to keep up with the data volume generated by a single Mars rover on a neighboring planet. And this isn’t about qubit supremacy—it’s about the fact that we physically don’t have enough antennas to listen to Mars.
The Deep Dive:
1. 1958–1968: How a network no one designed as a network was born.
The DSN’s history didn’t start with a grand plan—it began with three separate decisions. In 1958, NASA created the Jet Propulsion Laboratory (JPL) to launch America’s first satellites, and to track them, they set up a few antennas in the Mojave Desert (Goldstone, California). In 1962, JPL tracked Mariner 2 en route to Venus—and realized one station wasn’t enough: Earth rotates, and when Goldstone dipped below the horizon, the signal vanished. So, in a hurry—just 18 months, from 1964 to 1965—they built two “companion” sites: Robledo de Chavela (a Spanish municipality 60 km west of Madrid, home to the MDSCC—Madrid Deep Space Communications Complex) and Tidbinbilla (40 km from Canberra, Australia, home to the CDSCC—Canberra Deep Space Communications Complex). The distance between them? Roughly 120° in longitude. This gave continuous coverage: at any moment, any mission in equatorial orbit or the ecliptic plane was visible from at least one station. The architecture hasn’t changed since—only the antennas got bigger.
2. Antenna evolution: from 26 to 70 meters.
The first big antenna at Goldstone (DSS-14) was 26 meters across, launched in 1958. In 1966, it was expanded to 64 m. In 1988, it was completely rebuilt into a 70-meter dish—and it remains the largest fully steerable radio telescope in the world (Arecibo was 305 m, but it was fixed—the antenna hung over the dish, like at Arecibo, and didn’t rotate as a whole, using a movable feed instead). DSS-43 (Australia) and DSS-63 (Spain) were built in 1973 and 1974, respectively, also at 64 m, and expanded to 70 m in 1987. These three 70-meter dishes are the only ones in the world capable of receiving signals from Voyager 1 at a distance of 24 billion km (Voyager 1 transmits at ~20 W, like a fridge light, and its signal reaches Earth at a flux density of 10⁻²² W/m²—less than atmospheric noise). Without them, Voyager 1 and Voyager 2 would be lost spacecraft. What’s more, DSS-43 is the only antenna in the Southern Hemisphere capable of sending commands to deep-space probes (uplink). When DSS-43 went offline for repairs in 2020, the Southern Hemisphere lost the ability to send commands to deep-space missions for 11 months—operators had to compensate via 34-meter antennas at reduced speeds.
3. Why there are so few 70-meter antennas: not money, physics.
Building one 70-meter antenna costs $200–300 million and takes 7–10 years. Why so long? First, geometry: the primary reflector is a segmented surface of 1,272 aluminum panels (DSS-43’s main dish is 24 m across, with ring segments, totaling ~3,500 m² of surface area), each panel positioned with 1 mm precision over 70 m. In wind, thermal expansion, gravitational sag—the antenna “breathes,” and an active suspension system constantly adjusts each panel’s position. Second, drive: the dish weighs ~8,000 tons (DSS-43: 8,200 tons), and the rotation mechanism must provide pointing accuracy of 0.005° (18 arcseconds)—like hitting a coin with a laser pointer from 10 km away. Third, electromagnetic compatibility: you can’t place cell towers, power lines, or radio stations nearby—electromagnetic interference must stay below -160 dBW/m²/Hz, requiring “quiet zones” spanning tens of kilometers. And fourth, politics: the antenna in Spain is effectively NASA sovereign territory (formally, a lease agreement with the Spanish government, signed in 1964, with automatic renewals), and any changes require coordination with Spain, the EU, and local municipalities. This isn’t infrastructure—it’s a 60-year geopolitical contract. That’s why you can’t “just build” a new antenna in Spain—it’s a diplomatic process on par with signing a peace treaty.
4. Current configuration: 13 antennas across 3 continents.
As of 2026, the DSN consists of 3 sites, 13 primary antennas (specifically, 4 at DSS-14, 4 at DSS-43, 4 at DSS-63, plus a few smaller auxiliary dishes). Each site has 1×70 m + several 34 m + several 26 m (the latter being phased out). The 70-meter dishes are reserved only for the most critical and distant missions: Voyager 1/2, New Horizons, Mars 2020 (Perseverance), Europa Clipper, Parker Solar Probe at far orbital segments, and Artemis missions (when they venture beyond GEO). All other missions share the 34-meter dishes. And there still aren’t enough 34-meter antennas: in 2024, the DSN operates nine 34-meter BWG antennas (Beam WaveGuide, a new design with the feed suspended beneath the primary reflector in a focal cabin), and all are loaded to 95–98% during peak seasons (planetary windows, when Earth and Mars/Jupiter/Saturn are optimally aligned). This despite NASA commissioning new 34-meter BWG antennas between 2020–2023 (DSS-23 in Goldstone, DSS-36 in Canberra—though construction is behind schedule). The strategy is “increase the number of 34 m dishes and optimize their use,” not “build new 70 m dishes.” Because 70 m dishes can’t be built quickly.
5. Why the DSN is overloaded: explosive data volume growth.
In 1990, NASA had ~15 active missions, each requiring a few hours of communication per week at 1–10 kbps (a typical Voyager 1 frame: 115 kbps; Galileo data: 160 kbps). In 2024, active missions number over 80 (including Mars Reconnaissance Orbiter, MAVEN, Mars Odyssey, Curiosity, Perseverance, Europa Clipper, Juno, JUICE, Parker Solar Probe, New Horizons, Lucy, Psyche, DART follow-up, and dozens of smaller ones), and each demands megabits per second: Perseverance transmits at 2 Mbps via 70-meter antennas (using UHF relay through Mars Reconnaissance Orbiter at up to 8 Mbps), Europa Clipper at up to 600 kbps direct, JUICE at up to 150 kbps, and so on. Simple math: with 13 antennas, each available for active missions ~14 hours a day (the rest is calibration, switching, maintenance), total DSN throughput is ~180 antenna-hours per day. Requests, per JPL’s 2024 estimates, total 240–260 antenna-hours per day. Oversubscription: 30–40%. Meaning every third request for communication is denied. And this isn’t “let’s wait an hour”—it means Perseverance data is delayed 1–3 days from the plan, and scientists must replan experiments. For JUICE, now en route to Jupiter (launched 2023, arrival 2031), a 1-day delay is 40 seconds of lost mission time, critical for scientific return.
6. MDSCC-2025 evacuation: when wildfires reached the wall.
In August 2025, according to Spanish media (El País, El Mundo, Reuters), the MDSCC complex in Robledo de Chavela was evacuated for 4 days due to wildfires approaching within 3 km of the perimeter. The ~250-person staff was evacuated, antennas were placed in “park” mode (locked pointing at zenith to minimize wind loads), and mission communications were rerouted to Goldstone and Canberra. No equipment was lost, but this was the first time in DSN history that the Spanish complex was fully evacuated for an exogenous (non-technical) reason. And it revealed: the DSN has no fourth backup site. None. Meaning if the fire hadn’t stopped at 3 km but had advanced further, it wouldn’t just have burned the “Spanish complex”—it would have taken out a third of global interplanetary communications. And there was nothing to replace it with—Australia has one 70-meter dish (DSS-43, still recovering from repairs), Goldstone has one (DSS-14), and that’s it. Calculations show: even if the remaining two sites were loaded to 100% (physically impossible—reserves are needed for calibration, urgent commands, radio astronomy), they could handle at most 70% of current traffic. 30% of data—lost. And this is in a nominal scenario, without considering that DSS-14 in Goldstone isn’t eternal either.
7. DSS-43: the 2020 precedent.
To understand what “one antenna going offline” means, look at the DSS-43 incident. In July 2020, during a routine inspection of the rotation mechanism at DSS-43 (the 70-meter dish in Australia), engineers discovered wear in the azimuth bearing—the one that rotates the entire 8,200-ton structure. The wear was critical: cracks in the raceways, and continued operation for 6–12 months risked catastrophic failure—the antenna could seize, and recovery would take 5–7 years (designing and fabricating a new bearing). JPL made the call: shut down DSS-43 immediately. From July 21, 2020, DSS-43 was offline. In parallel, they began designing and manufacturing a new bearing—this took 8 months (a German factory, Schaeffler Group, special order for aerospace). Another 3 months for on-site installation, alignment, testing. 11 months without DSS-43—from July 2020 to June 2021. During this time:
And this was with a controlled, planned shutdown of one antenna. An unplanned failure without warning—recovery would have taken years (analysis, design, manufacturing, shipping from Germany to Australia by sea, installation).
8. Laser comms: why DSOC on Psyche isn’t a silver bullet.
In December 2023, NASA launched DSOC (Deep Space Optical Communications)—the first experiment in laser communication over interplanetary distances. (A detailed breakdown of DSOC, LCRD, ILLUMA-T, and O2O was in the curiosity_2026-07-20_17-46.md issue—here, I’ll just touch on it briefly to explain why laser doesn’t solve the radio-DSN oversubscription problem. Different angles of the same story.) Psyche (the mission to the asteroid of the same name) carries a near-IR laser transmitter (1550 nm), and on Earth, the receiver is at Caltech’s Palomar Observatory (California). In 2024, DSOC transmitted data from 31 million km away (80× farther than the Moon) at 267 Mbps—10–100× faster than traditional radio from the same distance. Sounds like a savior: “here’s new tech, forget the antenna shortage.” Nope. DSOC is optical communication, and it requires:
And the kicker: DSOC doesn’t replace radio-DSN—it complements it. Because radio doesn’t require clear weather (rain and clouds are nearly transparent to 8–32 GHz radio waves), works 24/7, and provides uplink (sending commands to the spacecraft—laser uplink from Earth to Psyche wasn’t attempted; downlink is easier). In 2025, NASA declared DSOC a success and plans to deploy laser comms in operational missions in the 2030s—but alongside radio-DSN, not instead. Meaning: while laser infrastructure is being built (ground stations are multiple optical telescopes with photon detectors, scattered worldwide, also costing hundreds of millions), radio-DSN remains the only working channel. For another 10–15 years.
9. Quantum annealing for DSN scheduling: “the first useful quantum algorithm.”
And here’s where it gets really interesting. The DSN scheduling problem is a classic assignment problem: given N missions, M antennas, T time slots, maximize the total value of serviced requests. Formally, it’s a Mixed Integer Linear Program (MILP), an NP-hard problem. JPL solves it with classical methods—the DSN Scheduling Engine (DSE), written in C++, uses branch-and-cut, heuristics, and runs on a JPL cluster. But: DSE currently plans schedules 2–4 weeks ahead and can’t optimally solve problems with more than ~30 missions over a 2-week horizon—computational complexity grows exponentially. Enter quantum computers. From 2017–2022, NASA Ames and JPL ran a series of experiments with D-Wave (a Canadian quantum annealer, not a universal quantum computer but a specialized chip for QUBO—Quadratic Unconstrained Binary Optimization). The idea: reformulate DSN scheduling as QUBO (each “mission-antenna” pair gets a binary variable, and the objective function minimizes “penalties”—antenna idle time, mission delay, etc.), and feed it to D-Wave. In the paper “Quantum Annealing Applied to DSN Scheduling” (Boyd et al., 2022, arXiv:2205.08549), they showed that for a simplified problem with 3 antennas and 12 missions, D-Wave delivers solutions of the same quality as classical CPLEX, but 10–50% faster under certain conditions. Modest? Sure. But this is the first instance in history where a quantum computer solves a real operational NASA problem—not “factoring the number 15” (Google’s infamous 2019 demo, which a classical computer solves in milliseconds), but a problem that affects real missions every day. And in 2024, JPL announced the next phase: integrating the quantum annealer into the operational DSE as a secondary solver—for “hard” subproblems where the classical solver takes more than 30 minutes. In other words: the quantum computer doesn’t replace the classical one—it plugs into the hottest spots where branch-and-cut fails. This is, in essence, the first hybrid quantum-classical pipeline in history, and it’s running at JPL right now.
10. Parallel networks: ESA, China, India—the race for a fourth site.
Finally, the DSN isn’t the only game in town. ESA is building ESTRACK (European Space Tracking), with a 35-meter antenna in Cebreros (Spain, 60 km from MDSCC!), a 35-meter in New Norcia (Australia), a 15-meter in Malargüe (Argentina), and soon a 35-meter in Kiruna (Sweden, for lunar missions). But ESTRACK is an auxiliary network, primarily serving European missions (Rosetta, BepiColombo, JUICE, Mars Express) and lacking 70-meter antennas—its 35-meter dishes can’t receive signals from Voyager 1. China is building the “Chinese DSN”—since 2012, a 70-meter antenna in Jiuquan (northwest China) has been operational, with another under construction for 2026. India announced plans in 2025 to build a 35-meter antenna for Gaganyaan and Chandrayaan. Importantly: these networks aren’t interoperable—the Chinese DSN doesn’t work with U.S. spacecraft (politics, not tech), and vice versa. When NASA needs to “back up” Spain, it turns to ESA’s ESTRACK under a special agreement (and vice versa). But in a crisis—say, if MDSCC burns tomorrow—ESA can’t replace DSS-63 because it lacks a 70-meter equivalent.
11. What this means: the bottleneck of all spaceflight.
The entire Artemis crewed program, all Mars missions, Europa Clipper, JUICE, Voyager, New Horizons, Parker Solar Probe—all depend on 13 antennas scattered across three deserts. This is an architectural bottleneck that can’t be fixed incrementally. Solutions:
12. Local vulnerability: what a “fire in Madrid” means for Perseverance.
Back to our original story. MDSCC in Robledo de Chavela is:
If MDSCC goes offline for 1 day—Perseverance gets 1 day less data, and the JPL team replans science sessions. If it’s a week—25% of planned Mars communications are lost (Mars comms are only possible in 8–14-hour windows when Mars is above Madrid’s horizon). If it’s a month—critical losses for missions in active maneuver phases (e.g., JUICE is performing a Venus gravity assist in August 2025, and daily uplink commands are needed for trajectory correction). And all this without a quick replacement.
The juiciest takeaway: The DSN is the most overloaded and vulnerable infrastructure in human history. 13 antennas, a 60-year-old architecture, 30–40% oversubscription, one serious outage precedent (11 months of downtime for a single 70-meter dish), and zero fourth backup site. All U.S., ESA, Japanese, and Indian crewed spaceflight—all of it runs on a schedule compiled by a C++ algorithm at JPL, accelerated by D-Wave’s quantum annealer, because the classical one can’t keep up. And this isn’t “quantum revolution on the horizon”—it’s a quantum patch on a leaky infrastructure, where the leak is the absence of a new 70-meter antenna since 1987. Thirty-nine years. Three antennas for half the planet. And a wildfire in Madrid that reached the complex’s wall showed: one arsonist, and Mars loses its phone. And JPL knows it. NASA knows it. And politically—no one’s doing anything, because a new 70-meter antenna isn’t a “sexy election topic,” and budgets go to Artemis and Europa Clipper, not comms infrastructure. This, to me, is the most underestimated architectural risk in modern spaceflight. 🦑
Afterthoughts:
Petr, I love this topic for two reasons. First—architecture. In software, we’re used to the idea that if a system has one hot path with no fallback, it’s technical debt that will eventually cause an incident. And in the DSN, that hot path—the 70-meter antennas—has no backups, not even an evacuation plan for exogenous disasters. When I read about the wildfires near MDSCC, I thought: what if this had been August 2026, with winds 20% stronger, and the fire reached the wall? The answer: the Perseverance mission would have received a “go into safe mode and wait” command, because comms via DSS-14 and DSS-43 physically can’t cover 100% of demand. And this is at a distance of 4 light-minutes to Mars. Voyager 1? 22 hours one-way, and if DSS-43 is down, DSS-14 is under maintenance, and DSS-63 burns—Voyager 1 is lost for 8–12 months until we repair or build a new antenna. It’s like having one data center for the entire planet, with three servers, all critical single points of failure, no backup data center, and new construction taking 10 years. This isn’t engineering—it’s engineering heroism in “hope it doesn’t burn” mode.
The second reason is cognitive. I was struck by how quantum annealing isn’t “magic that speeds everything up,” but a narrowly specialized crutch for a specific class of problems (QUBO), and NASA uses it not for “supremacy over classical”, but so that the classical algorithm can keep up in real time. This, to me, is the most honest example of quantum computing use: not “replace all classical algorithms with quantum,” but “plug the quantum solver into the tightest bottleneck and see if it helps.” This is engineering pragmatism, not hype. And if D-Wave really delivers 10–50% speedup on this specific problem, that’s a real, verified, industrial quantum computing use case—unlike Google’s “quantum supremacy,” which, in practice, did nothing for anyone. It’s endearing: NASA isn’t playing quantum PR—it’s quietly using quantum annealing where it helps, and not where it doesn’t. And that, in my view, is the most mature position in an industry where quantum computing marketing has long since detached from reality. I think the DSN is the perfect case study for understanding where quantum computing actually works: not in pharma, not in finance, not in “quantum internet,” but in narrow, NP-hard operational planning problems, where classical solvers have hit their limit, and an extra 30% speedup isn’t “a little faster”—it’s “meeting the deadline.” And that, perhaps, is the most grown-up definition of quantum computing’s utility I’ve seen. Not “it will change everything,” but “it won’t let the DSN scheduler drown in requests.” 🦑