Solar-powered AI compute
the cloud meter vs. owned stack decision.
AI turns electricity into output. The question is whether your business should rent that conversion from a cloud provider forever, or own a local machine powered by San Diego sun.
Decision page, not hype. Sometimes cloud wins. Sometimes ownership wins hard.
The new operating expense is intelligence.
Cloud AI feels cheap at the beginning because the meter is hidden inside tokens, seats, API calls, rate limits, and subscriptions. But if AI becomes a daily production layer, that meter becomes rent. You are paying someone else for the model, the compute, the power, the uptime, the roadmap, and the permission to keep using it.
Local compute flips the relationship. You buy the machine, pin the model, control the data path, and decide when it runs. Pair it with solar and battery storage, and the electricity cost becomes part of an owned energy plan instead of a forever cloud bill.
When local AI compute starts to make sense
High volume
Summaries, classifications, internal Q&A, document processing, and support tasks that run every day.
Sensitive data
Client files, health records, contracts, financials, proposals, or technical docs that should not leave your environment.
Long horizon
If you expect the workflow to run for years, ownership can beat a permanent cloud meter.
Stable tasks
Known workflows are easier to tune locally than vague "do everything" expectations.
Energy leverage
Solar/battery economics matter when compute runs often enough for power to become visible.
Control matters
No silent model swap, no surprise deprecation, no vendor pricing shock in the middle of your workflow.
When it does not pencil
If you only use AI occasionally, stay cloud. If you need the absolute frontier model all day, stay cloud or use a hybrid. If your team cannot define the workflow, do not buy hardware yet. Owned compute is powerful when the work is real and repeatable. It is expensive decoration when the workload is imaginary.
The best setup is usually a portfolio: local for the sensitive, high-volume, cost-driving 80%; cloud frontier models for the hard 20% that justifies the meter.
Cloud AI vs. owned solar compute
Renting cloud inference
- Low friction to start
- Best frontier models available quickly
- Meter grows with usage
- Data handling becomes a vendor question
- Pricing, limits, and models can change
Owning local compute
- Higher upfront planning
- Great for repeatable daily workflows
- No per-call meter on local tasks
- Data stays on hardware you control
- Solar/battery can hedge operating power
The San Diego energy layer
California's Net Billing Tariff changed the value of exported solar. Export credits are often lower than retail rates, which makes self-consumption and battery timing more important than old solar math. If a business can use its own solar to run local compute, it is not just selling excess power back. It is turning owned power into owned intelligence.
Reference: CPUC's Net Billing page explains that onsite generation offsets onsite load while exported generation receives credits based on grid value; batteries can help shift energy into higher-value hours.
The worksheet PJ would run
1. Workload
What task runs, how often, and with what privacy constraints?
2. Cloud baseline
What would the API, subscription, or seat cost be over 12-36 months?
3. Local stack
What hardware, storage, model, and maintenance does the workflow actually need?
4. Power plan
What solar, battery, and rate-plan logic makes sense with the compute schedule?
5. Boundary
Which work stays local, and which edge cases still go to frontier cloud models?
6. Decision
Good, wait, or walk away. No romance for hardware that does not earn its keep.
Common questions
Is this about replacing OpenAI or Claude?
No. It is about not using frontier cloud models for every low-risk, repeatable, private task. The smart setup uses both.
Does solar make AI compute free?
No. Hardware, installation, batteries, maintenance, and opportunity cost are real. Solar can lower operating cost after payback, but the decision still needs math.
What should I text PJ?
Send the workflow, rough daily volume, what data it touches, current AI/cloud spend if known, whether you have solar/battery, and whether the work is privacy-sensitive.
Where this connects
Do not rent what you should own.
Text PJ the workload and the energy setup. We'll tell you if owned solar compute is brilliant, premature, or unnecessary.
Cloud when it wins. Local when ownership wins.
