Wired to Burn
An interactive wildfire simulation connecting AI energy demand to wildfire risk through power grid infrastructure.
Every prompt you send needs electricity. That electricity travels on transmission lines through wildland. As AI energy demand grows, more lines cross more dry ground. More infrastructure. Less water. The conditions for catastrophic fire.
Six of the ten most destructive California wildfires were started by power equipment.1 The state just approved 31 new transmission projects to meet rising demand.2 By 2030, AI data center energy consumption is projected to reach 945 TWh, the entire electricity demand of Japan.3
Wired to Burn doesn't ask you to feel guilty about your prompts. It asks you to see the system those prompts are part of.
I live in San Francisco. I see AI everywhere, on billboards, in conversations, in the tools I use every day. At some point I stopped seeing it as technology and started feeling it as something heavier: a demand, spreading across the city, across the grid, across the land.
I wanted to understand what that demand costs. Not in dollars. In ground. In water. In the terrain that power lines cross on their way to the data centers feeding all of it.
This piece was born from that question, tracing the line between a prompt and a power line, between a query and a wildland fire. The more I looked, the more that line became undeniable.
What I wanted was something that could carry that understanding beyond a report, beyond a statistic, beyond a policy document nobody outside a regulatory proceeding will ever read. Something a person could stand in front of and feel.
If this piece helps push even one person toward asking harder questions about the energy behind the tools they use every day, toward responsible AI use, toward ethical AI design, then it has done what I made it to do. This is my climate care vision. This is what data humanism means to me in practice: not just making data human. Making it matter enough to protect what we still have, the land, the climate, the earth.
Wired to Burn is built on cellular automata, the same computational framework Daniel Shiffman explores in Nature of Code. Every cell holds a terrain type, a moisture value, and a state: alive, burning, or dead. Fire is not triggered by a button. It emerges from conditions.
Core spread rule
P(spread) = spreadRate × (1 − moisture)
If a cell's moisture is above its ignition threshold, that probability drops by 95%. Wind applies a dot product across spread direction, fire travels 2.5× downwind, 0.2× against it.
Infrastructure ignition
P(ignition) = 0.0003 × demand² × (1 − moisture) × spreadRate × 10
Doubling energy demand quadruples ignition risk. That is not a design choice. That is the physics.
Moisture decay
moisture(t+1) = moisture(t) + (target − moisture(t)) × 0.04
As energy demand rises, the moisture target drops. The landscape dries the way drought happens: gradually, and then catastrophically.
Six terrain types. Six fire signatures.
| Terrain | Spread | Burn | Moisture | Recovery |
|---|---|---|---|---|
| Forest | 4% | 28 frames | 70% | 0.1% per tick |
| Shrubland | 8% | 22 frames | 35% | 0.3% per tick |
| Agriculture | 6% | 16 frames | 50% | 0.8% per tick |
| Grassland | 12% | 12 frames | 45% | 1.2% per tick |
| Wetland | 1% | 35 frames | 90% | 0.6% per tick |
| Developed | 2.5% | 20 frames | 55% | none |
- Forest: The ancient one. Slow to catch. Crown fire energy ten times grass. Keeps spreading after grassland recovers.
- Shrubland: The dragon. Oil-rich chaparral at 2,000°F. Highest spread after grassland, lowest base moisture.
- Agriculture: The open field. Nothing stops the wind. Dry stubble to the horizon.
- Grassland: The sprinter. 12% spread, ~14 mph in the real world. Burns out in 12 frames but carries fire forward.
- Wetland: The guardian. Natural firebreak until water is diverted for cooling infrastructure.
- Developed: The settlement. No recovery. What burns here does not come back in the simulation, because in the real world, most of the time, it does not.
Every rule is displayed inside the piece on the terrain picker screen in mathematical notation. The math is not hidden. Transparency is part of the design.
The design philosophy
The first design question was not what to build. It was what the user should feel the moment fire appears on screen.
Not clarity. Not legibility. Visceral immediacy. The design had to make someone feel the phenomenon before they could intellectually process it. That is the threshold where data stops being information and starts being experience, Giorgia Lupi's argument made spatial.
Most data visualization is optimized for understanding. Wired to Burn is optimized for feeling. The understanding is the aftermath.
Wired to Burn was exhibited at TIAT in San Francisco as part of the Nature of Code Exhibition, a show celebrating Daniel Shiffman's computational thinking and its influence on creative practice. TIAT is a nonprofit gallery at 151 Powell Street in Union Square.
Visitors found the terrain picker first, some chose California immediately, some built a custom mix until the total reached 100%. Then they moved into the simulation: moisture, wind, ignition points, fire mode. Most people started by dragging moisture down.
What I noticed: people slowed down when fire did not do what they expected. Something in a corner would catch, spread diagonally with the wind, cross terrains at different rates, and they would look up and say something to whoever was standing next to them. That moment is the one the piece was made for: when the system behaves emergently and awareness stops being individual and starts moving.
Wired to Burn was developed during the DVS (Data Visualization Society) mentorship program, a framework for thinking rigorously about the argument before touching code: what the piece is saying, who it is saying it to, and whether every design decision serves the thesis.
The DVS community is where I learned to hold the distinction between making something beautiful and making something true. Wired to Burn is trying to be both.
Wired to Burn is one simulation, one screen, one exhibition in San Francisco. But the pattern it models is not local to California, or to one fire season, or to one utility company's liability.
I made this piece because I live in a city where AI is everywhere and its consequences are somewhere else, somewhere less visible, somewhere that gets drier every year. The awareness gap between people building and using these tools and people living closest to the infrastructure that powers them is structural. We do not feel what we do not see.
Data art can close that gap, not by assigning blame, but by making the invisible visible. Data that cares. Design that is accountable. Technology that does not look away from its own footprint.
If this piece reaches one classroom, one policy conversation, one designer who starts asking different questions about the systems they build, it has done what it was made to do. The simulation is open. The code is documented. The argument travels.
- Six of the ten most destructive California wildfires caused by power equipment: CAL FIRE Incident Investigation Reports; CPUC Wildfire Safety Division proceedings, 2019–2021. cpuc.ca.gov
- 31 new California transmission projects: CAISO 2024–2025 Transmission Planning Process; CAISO Board approval, 2024. caiso.com
- 945 TWh AI data center energy projection by 2030: International Energy Agency, Electricity 2024; IEA World Energy Outlook 2024. iea.org
Additional sources: USFS Fire Behavior Field Reference Guide (2021); NLCD 2016; Epoch AI, Trends in Machine Learning Hardware (2024).
- Inspired by Daniel Shiffman, Nature of Code, natureofcode.com
- Design influences Giorgia Lupi, data humanism, Dear Data, Pentagram editorial design thinking
- Exhibited TIAT, San Francisco, tiat.place
- Developed with support from DVS mentorship, datavisualizationsociety.org
This simulation models contributing factors to wildfire risk. It is not a prediction.