ChargeWheel
A browser-based simulation that demonstrates the core thesis behind ChargeWheel's UpGrid platform: in AI data centers, the binding constraint is no longer chips, it's grid power.
Core trick
LLM chooses per-interval battery and compute load by weighing grid power cost against compute revenue within a fixed interconnection limit.
Description
A browser-based simulation that demonstrates the core thesis behind ChargeWheel's UpGrid platform: in AI data centers, the binding constraint is no longer chips, it's grid power. Utility interconnection upgrades take years, so the operators who win are the ones who can run more compute behind the power they already have. UpGrid does this by integrating a solid-state transformer and battery storage into one cabinet, then letting an AI agent decide how to spend that scarce power profitably.
README excerpt
# UpGrid — DC Native Compute Orchestrator A live demo showing why owning the power-electronics layer lets an AI data center run more compute than its grid connection allows — with Claude Opus 4.8 making and explaining the economic dispatch decisions in real time. Built for the Claude Fable 5 Build Day. Hosted on AWS Amplify; inference runs through Claude Platform on AWS (Anthropic's native Messages API, accessed via the AWS account). --- ## What this is A browser-based simulation that demonstrates the core thesis behind ChargeWheel's UpGrid platform: in AI data centers, the binding constraint is no longer chips, it's grid power. Utility interconnection upgrades take years, so the operators who win are the ones who can run more compute behind the power they already have. UpGrid does this by integrating a solid-state transformer and battery storage into one cabinet, then letting an AI agent decide how to spend that scarce power profitably. ## Why it's needed The advantage UpGrid claims is physical, and therefore hard to show on a slide. A software-only competitor can also "orchestrate compute and batteries," so the differentiation isn't obvious until you can see two systems respond to the same event and diverge. The demo exists to make that divergence visible and undeniable — and to prove the AI layer is doing real reasoning, not following a script, by letting anyone perturb the scenario live. ## How it works The simulation runs two systems side by side against one sha