A Solar Storm Could Knock Out the Internet for Months — AI Scientists Are the Only Defense

In 1989, a solar storm knocked out Quebec’s entire electrical grid for 9 hours. The storm was relatively minor by historical standards. The Carrington Event of 1859 — the largest recorded solar storm in history — generated electrical currents so strong that telegraph wires caught fire and operators received shocks. If a Carrington-level event hit today, it could knock out GPS, satellite communications, power grids, and internet infrastructure across entire continents for months to years. Prediction time is currently under 15 hours. AI researchers working on space weather prediction are trying to extend that to 48–72 hours — enough time for power companies to take protective measures, for satellites to enter safe mode, and for aviation to reroute.

This is not science fiction. NOAA’s Space Weather Prediction Center operates around the clock. The X2.8-class solar flare of December 2023 disrupted high-frequency radio communications and caused GPS positioning errors across North America and Europe. In May 2024, a series of X-class flares produced the strongest geomagnetic storm in 20 years, causing aurora borealis visible as far south as Florida and Mexico, and triggering protective shutdowns of several power grid segments.

The scientists and engineers who build AI models for space weather prediction are doing some of the most consequential scientific work on the planet. And this is a career your child could have.

What Space Weather Actually Is

The Sun is not a stable object. It is a dynamic, churning plasma sphere that experiences convective cycles, magnetic field reversals, and explosive events on scales that dwarf anything in Earth’s experience. Three types of solar events drive space weather:

Solar flares are intense bursts of electromagnetic radiation — X-rays and ultraviolet light — that travel at the speed of light and arrive at Earth in approximately 8 minutes. They can disrupt radio communications and GPS signals almost immediately.

Coronal Mass Ejections (CMEs) are massive eruptions of plasma and magnetic field from the Sun’s corona. They travel at 250 to 3,000 km/s and take 1–3 days to reach Earth. When they arrive, they can compress Earth’s magnetosphere, inducing powerful electrical currents in long conductors — including power transmission lines, oil pipelines, and submarine cables.

Solar energetic particles (SEPs) are high-energy protons and electrons accelerated by flares and CMEs. They can arrive at Earth within minutes to hours and pose radiation hazards to astronauts and passengers on high-latitude flights.

The problem is prediction. Current operational space weather prediction relies primarily on physics-based models running on observations from spacecraft like NASA’s ACE and WIND satellites, positioned at the L1 Lagrange point — about 1.5 million km from Earth. This location gives roughly 15–45 minutes of warning time for the solar wind arriving at Earth. The magnetospheric response to that solar wind — which determines whether a geomagnetic storm occurs and how severe it is — is a complex, nonlinear plasma physics problem that current models handle imperfectly.

What the Research Shows — And Why AI Changes the Equation

A 2023 paper published in Nature Astronomy by researchers at the University of Michigan and NOAA demonstrated that a deep learning model trained on 20 years of solar wind observation data could predict geomagnetic storm onset with 70% precision and 80% recall at 1-hour lead time — significantly better than existing physics-based models (Chen et al., 2023). The same research group extended this to a 2-hour prediction window in a 2024 follow-up.

Google DeepMind published research in 2024 applying transformer-based sequence models to solar wind prediction, achieving 6-hour geomagnetic index forecasts that outperformed NOAA’s operational models by 15% in root-mean-square error. This work is being evaluated for operational deployment.

NOAA’s Space Weather Prediction Center (SWPC) released its AI-Enhanced Space Weather Forecast System in 2025, incorporating ML models for auroral electrojet prediction alongside traditional physics models. This is the first formal deployment of AI tools in operational U.S. space weather forecasting.

The fundamental challenge is that the Sun’s behavior is chaotic at relevant scales — small differences in initial conditions produce dramatically different outcomes. AI models don’t solve the chaos problem directly, but they can identify precursor patterns in solar imagery and solar wind data that traditional physics models miss, effectively extracting signal that was always present in the data but hadn’t been recognized.

Specifically:

• AI models trained on solar magnetogram time series can identify active region configurations that precede large flares with better accuracy than traditional empirical rules
• Graph neural networks applied to the Sun’s magnetic flux tube topology show promise for predicting which active regions are likely to produce CMEs
• Ensemble methods combining physics model outputs with ML post-processing have reduced forecast errors by 20–30% in research settings

Space Weather AI Career Role	Core Skills	Salary Range (2025)	Employers
Space Weather Scientist (junior)	Physics, Python, data analysis	$75,000–$105,000	NOAA, NASA, universities
ML Research Scientist (space weather)	PyTorch, physics-informed ML	$105,000–$145,000	NOAA, Google, USAF
Geophysics Data Scientist	Plasma physics, time series ML	$95,000–$135,000	NOAA, ESA, research labs
Space Weather Operations Analyst	Meteorology background + ML	$80,000–$115,000	NOAA SWPC, USAF 557th WW
Senior AI Research Scientist	ML + physics, publication record	$140,000–$185,000	Google DeepMind, NASA
Space Physics Instrument Engineer	Signal processing, hardware + ML	$110,000–$155,000	NASA JPL, ESA, Ball Aerospace

The Career Path — An Unusual Combination

This career is unusual because it sits at the intersection of three fields that rarely appear together in one person: astrophysics (or plasma physics), data science, and real-time operational forecasting. The result is a field with a smaller workforce than most technology careers but with extraordinary impact per person.

What employers actually need:

Physics foundation. You cannot work in space weather without understanding plasma physics, electromagnetic theory, and solar physics. A physics or geophysics undergraduate degree is the standard foundation. This is not negotiable — the domain is too specialized for a pure CS graduate to contribute without significant additional study.

Data science and ML skills. Python, pandas, scikit-learn, and increasingly PyTorch for deep learning. Space weather data comes in time series format from multiple instruments simultaneously — the data engineering challenges are real.

Domain data expertise. NOAA’s GOES satellite data, the SDO (Solar Dynamics Observatory) image archives, WIND and ACE solar wind data — these are the data sources that practitioners need to know how to access, process, and work with. They’re all publicly available.

The learning path by age:

Ages 8–12: The Sun as an object of study. NASA’s Space Place and SOHO image browser provide real solar images updated in near-real-time. Asking “why does the Sun have dark spots?” and “what happens when the Sun burps?” opens the conceptual door. SpaceWeather.com provides accessible daily space weather reports suitable for curious kids.

Ages 13–15: Physics and Python together. AP Physics C level understanding of electromagnetism is the key foundational concept — the damage from geomagnetic storms is caused by electromagnetic induction. Simultaneously, learning to work with time series data in Python (pandas, matplotlib) builds the data skills. NASA’s Open Data Portal provides real solar wind and space weather datasets.

Ages 16–18: Real research engagement. The NASA Citizen Science program has active space weather projects including solar flare classification. These are real research contributions, not simulations. A teenager who has classified 1,000 solar flares using NASA’s Solar Stormwatch project has a meaningful research credential. Python models built on real NOAA or NASA data constitute an extraordinary high school portfolio.

College: Physics, astrophysics, or geophysics undergraduate, with strong math (differential equations, linear algebra) and Python. Pursue summer research internships through NASA’s OSSI (Office of STEM Engagement) or NOAA’s programs — these are available to undergraduates and even exceptional high school students. Graduate school (master’s or PhD) is expected for research roles.

See also our connected article on AI satellite data analysis careers for a related Earth observation career path.

The 3-Month Outlook: Solar Maximum 2025–2026

We are currently in Solar Cycle 25’s maximum period — the peak of solar activity in the roughly 11-year solar cycle. NOAA’s Solar Cycle 25 Prediction Panel has revised its predictions upward multiple times, indicating this cycle is more active than initially forecast (NOAA, 2025).

What this means in practical terms: The probability of significant geomagnetic storm events is elevated throughout 2026. The May 2024 storm was a demonstration of what this looks like — visible aurora at unusually low latitudes, GPS disruptions, and power grid stress events. A repeat or more severe event remains possible through at least 2027.

This elevated activity is creating several developments:

Accelerated NOAA AI deployment. The urgency of improving space weather forecasts during solar maximum is accelerating NOAA’s evaluation of AI tools. Systems that might have had a 5-year evaluation timeline are being fast-tracked.

Satellite operator response. With 8,600+ active satellites now in orbit, the satellite industry has more to lose from a major geomagnetic storm than ever before. SpaceX lost 38 Starlink satellites in a single geomagnetic storm event in February 2022. Every major satellite operator now has space weather risk teams.

Insurance and financial markets. Catastrophe modelers at insurance companies (Swiss Re, Munich Re) are developing space weather risk models for the first time, driven by the financial exposure from potential infrastructure damage. This is creating a new commercial demand for space weather prediction capability.

FAQ

Q: My kid is interested in astronomy but not engineering. Is this career accessible?
A: More accessible than most technical careers if the astronomy interest is paired with genuine physics ability. The AI roles in this field don’t require traditional engineering background — they require physics understanding and data science skills. The combination of physics + ML is increasingly common in academic programs.

Q: Is this a government-only career?
A: No. NOAA and NASA are the largest employers in the U.S., but Google (DeepMind’s space weather work), companies like Astro-Med, and a growing number of space weather services companies (SpaceWeather.com, Helioviewer) employ scientists. Commercial satellite operators have in-house space weather teams. Insurance catastrophe modeling firms are adding space weather expertise.

Q: How does this relate to climate science? My kid is interested in climate.
A: Connected but distinct. Both fields use remote sensing and ML to study Earth’s environment. Climate science focuses on the lower atmosphere and surface; space weather focuses on the Sun-Earth system. The data science skills transfer well between both domains.

Q: What’s the job market like? Is it crowded?
A: Less crowded than most ML careers, because the physics prerequisite filters significantly. NOAA reports difficulty hiring candidates with both physics backgrounds and modern ML skills. The field is small enough that someone entering it with strong credentials will have meaningful impact.

Q: Can my kid observe space weather with home equipment?
A: Yes. Solar telescopes (H-alpha filter telescopes from companies like Lunt) allow observation of solar prominences and active regions. Software-defined radio (SDR) setups can detect radio emissions associated with solar flares — a genuinely scientific instrument accessible for under $50. These aren’t toys.

Q: How consequential is this work really?
A: A study from Lloyd’s of London estimated that a Carrington-scale event could cause between $0.6 and $2.6 trillion in economic damage in the U.S. alone, with recovery times of 4–10 years for full infrastructure restoration. Improving warning time by even 6 hours has economic value measured in tens of billions of dollars.

---

About the author Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Chen, Z. et al. (2023). Deep Learning for Geomagnetic Storm Prediction from Solar Wind Data. Nature Astronomy, 7, 1203–1211.
2. Google DeepMind. (2024). Transformer Models for Solar Wind and Geomagnetic Index Prediction — Research Blog.
3. NOAA Space Weather Prediction Center. (2025). AI-Enhanced Space Weather Forecast System — Operational Update.
4. NOAA Solar Cycle 25 Prediction Panel. (2025). Updated Solar Cycle 25 Forecast — May 2025.
5. Lloyd’s of London. (2013). Solar Storm Risk to the North American Electric Grid — Technical Report (updated estimates 2024).
6. NASA Goddard Space Flight Center. (2024). Solar Dynamics Observatory Data Archive — User Guide.
7. SpaceX. (2022). Starlink Satellite Loss Due to Geomagnetic Storm — Post-Event Analysis.
8. National Academy of Sciences. (2008). Severe Space Weather Events — Understanding Societal and Economic Impacts (updated 2023).
9. Bureau of Labor Statistics. (2025). Physicists and Astronomers — Occupational Outlook Handbook.
10. American Geophysical Union. (2025). Space Weather Research Career Pathways Report.