Schedule compute. Schedule power.

We spend a lot of time finding power for AI. This week, I asked a different question: Can we also schedule computing around the power available? At KiWi, we are exploring an R&D concept that brings computing workloads and energy dispatch into the same planning process, potentially across data centers. This is still an early technical and partnership discussion. It is not a deployed capability. But working through the constraints has made the opportunity more interesting to me. Most AI infrastructure discussions start with megawatts, PPA volumes and battery capacity. Once we look at the workloads, the picture changes. Real-time inference has latency commitments. Training jobs may have scheduling flexibility, but moving them involves deadlines, checkpoint overhead and GPU availability. Moving work between sites adds data-transfer and security constraints. UPS capacity reserved for emergencies is not automatically available for trading. An optimizer that ignores these conditions might reduce the electricity bill while creating a much larger problem elsewhere. If the customer’s job finishes late, we have not necessarily saved money. We may simply have moved the cost. That is why I want KiWi’s AI-EDC work to start with service commitments, not a target for flexible megawatts. First, establish what cannot move. Then determine what can move, for how long and under whose authority. The energy side needs to understand interval load data, renewable supply, PPAs and battery state of charge. The computing side needs to understand deadlines, available capacity and service quality. Both sides have to agree on a feasible schedule. The hard work ahead is practical: access to data, control permissions, command execution, fallback behavior and verification. Those details will determine whether this becomes a useful operating capability or remains a good idea on a slide. For me, the industry opportunity is to build flexibility into operations, rather than assume it exists because a facility consumes a lot of electricity. That question matters wherever compute and energy partners work together, not just in Taiwan. After years of building a company, I have become more interested in the interfaces between systems. A useful business can come from making two systems work together reliably, without either one breaking its promises. Schedule compute. Schedule power. First protect the computing service. Then earn the flexibility. Making Every kWh Smarter. Steve Huang Founder & CEO, KiWi New Energy #KiWiNewEnergy #AIInfrastructure #AIDataCenters #EnergyStorage #PPA #EnergyTech #DemandFlexibility #EnergyTransition