An August evening in Santa Clara. Air conditioning is on everywhere, the load spikes, and the local utility sends a signal to a data centre with thousands of GPUs.
About forty people are watching on a video call. Nobody touches anything. Power drops from four megawatts to three, and the important jobs keep running.
Fact
The facts: on 16 September 2026 Emerald AI, Google and NVIDIA announced the AI Energy Management Alliance (AEMA), a coalition for data centres that manage their consumption according to grid conditions. The alliance is technology-neutral and cares about the measurable service: how fast, how long and how predictably a facility gives power back and how it behaves in an emergency. Its principles: obligations for staying connected through brief disturbances, for curtailment and for emergency response to be defined before interconnection; standardised requirements and data; a faster path to interconnection for those who make verifiable flexibility commitments; interconnection costs that reflect the real impact on the system. A day earlier, on 15 September, NVIDIA described a case from a working facility: Silicon Valley Power has sent over 200 demand signals to an AI factory in Santa Clara run with Emerald AI's Conductor platform; in the case described, power drops from 4 to 3 megawatts automatically, with no high-priority job stopped. Per NVIDIA, the signal has been met every time, and Conductor responds in under a minute. Cloud provider Lambda reports 24 per cent more tokens within the same power budget using NVIDIA DSX MaxLPS, running 19 servers within the limit that used to hold 16.
The deal underneath
Nobody here cuts power out of the goodness of their heart. The deal is simple: the grid is short, a new line takes years to build, and a facility that promises to give way during peaks can be connected sooner and with more power. Flexibility is the ticket to the queue.
Whoever promises to give way should get in first.
It is elegant and it matters. And still, the alliance is made of three players who profit from more data centres. Interconnection rules are not written by them, and AEMA's principles are a proposal, not a law.
The measure of success will come from outside. It will be the first regulator that grants faster interconnection in exchange for such a promise, and the first hot evening on which that promise is tested in front of everyone.
Official primary sources
NVIDIA Blog - Emerald AI, Google and NVIDIA Launch Alliance to Advance Flexible AI Data Centers, 16.09.2026NVIDIA Blog - From Megawatts to Tokens: How NVIDIA Maximizes AI Factory Production, 15.09.2026The visual is generated code-art, no third-party images.