Solar power
no grid, no limits
On-site solar arrays and storage. Capacity expands with land, not with permits.
solar AI · energy & light
AI compute that runs
on the sun.
Solar-powered data centers in deserts. Cheap, renewable, simple, and expandable — built for long-term model training, beyond the electrical grid.
what we build
Simple, Solar AI.
Brightpalm builds the energy and the light-based hardware that turns sunlight into computation — for AI models that grow larger every year.
Light hardware
photons over electrons
Optical compute and the supporting energy systems for training and inference at scale.
Jing Compute Block
owned or leased
A self-contained T/G/NPU + solar + storage block, sited on land you control.
the problems we work on
Earth, people, biology — and the models that read them.
🌍 Earth's changing environment is one of the largest problems we face. Human relations and economics shaped our world; biology shapes our health. For each of these, machine learning has become a path to finding solutions — and the results from those models now affect us, and our world.
Training those models takes power. We make the power, and the place to put it.
today's training reality
What model training needs — and what it costs.
- Large compute on GPUs / TPUs / NPUs.
- High electricity consumption — and rising.
- High cooling requirements — heat dominates the build.
- Capacity expansion is limited by the grid and by site selection.
- Data sits close to the internet — but training does not need internet.
forward from physics
The grid is the bottleneck. The sun is not.
- High electricity consumption → solar arrays.
- High cooling requirements → larger surface area for passive cooling.
- Training does not need internet → site close to the energy, not close to people.
Models will grow larger. We will make and need models of everything. Model designs will compete. The compute they need has to come from somewhere — we think it should come from the sun.
long haul computation
The Jing Compute Block.
A solar-powered data center module — T/G/NPU compute, solar generation, energy storage, and data storage — sited on land you own or lease.
Owned or leased
physical control of data
Hardware compute, solar power, and storage all in one block. Blocks instance across the array.
Long-term training
years, not weeks
The land is leased or owned. Data storage sits physically on the block. Train models for years powered by sun.
Off the grid
grid independent
No grid interconnect, no transmission losses. The energy and the compute share an address.
why deserts
Solar AI in the desert.
Land is cheap. Sun is reliable. Heat dissipation is solved by distance and surface area. The grid is no longer the limit.
- Inexpensive
low capital + operating cost - Renewable
no limit on energy supply - Simple
simpler than grid power - Expandable
fewer obstacles to growth
market size
Estimated AI budgets of city governments.
A coarse top-down: world cities, planning budgets at 2 %, ML modeling at 2 % of that. The ML compute spend already runs in hundreds of millions of dollars — at the city level alone.
$394B
World cities, non-salary budget
$7.88B
Planning budget @ 2 %
$157M
City-gov ML modeling @ 2 % of planning
| City | Non-salary budget ($B) | Planning @ 2 % | ML modeling @ 2 % |
|---|---|---|---|
| New York | 44 | 0.88 | 0.0176 |
| Tokyo | 42 | 0.84 | 0.0168 |
| Los Angeles | 30 | 0.6 | 0.012 |
| Seoul | 25 | 0.5 | 0.01 |
| Chicago | 20 | 0.4 | 0.008 |
| London | 20 | 0.4 | 0.008 |
| Paris | 19 | 0.38 | 0.0076 |
| San Francisco | 18 | 0.36 | 0.0072 |
| Dallas | 17 | 0.34 | 0.0068 |
| Washington | 16 | 0.32 | 0.0064 |
| Houston | 15 | 0.3 | 0.006 |
| Osaka | 14 | 0.28 | 0.0056 |
| Shanghai | 14 | 0.28 | 0.0056 |
| Boston | 13 | 0.26 | 0.0052 |
| Seattle | 13 | 0.26 | 0.0052 |
| Beijing | 13 | 0.26 | 0.0052 |
| Atlanta | 11 | 0.22 | 0.0044 |
| Philadelphia | 11 | 0.22 | 0.0044 |
| Singapore | 10 | 0.2 | 0.004 |
| Moscow | 10 | 0.2 | 0.004 |
| Miami | 10 | 0.2 | 0.004 |
| Shenzhen | 9 | 0.18 | 0.0036 |
our network