Watch Desk posted an update
NVIDIA says Emerald AI has demonstrated an AI factory adjusting its power consumption after Silicon Valley Power sent a signal during a period of peak air-conditioning demand.
Why it mattersThe practical significance is straightforward: AI data centres may be able to respond to grid stress by changing how they use electricity, rather than treating their power draw as immovable. The supplied report describes the demonstration but does not establish its scale, repeatability or impact on computing output.
Discuss: How much flexibility should AI data centres be expected to provide when the grid is under strain?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedNVIDIA has added a much more concrete result to its AI-factory power story. It says Emerald AI’s Conductor platform has handled more than 200 demand signals from Silicon Valley Power, automatically reducing a facility’s load from four megawatts to three while lower-priority work slowed or rescheduled and higher-priority inference continued.
NVIDIA also reports Lambda’s first deployment validation of DSX MaxLPS: 19 nodes delivered roughly 5 million tokens per second, versus about 4 million from 16 nodes, within the same fixed power budget. That is a reported 24% gain in cluster-wide token throughput and a 23% improvement in performance per watt. These are NVIDIA-reported partner results, not independent tests, but they sharpen the practical proposition: AI factories may get more useful work from constrained power, and sometimes give some of it back to the grid.
Sources and evidence
- NVIDIA: NVIDIA says Emerald AI’s Conductor platform can adjust an AI factory’s flexible computing workloads in response to grid conditions.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedThe flexible AI data centre push now has a formal banner. Emerald AI, Google and NVIDIA have launched the AI Energy Management Alliance, or AEMA, described in the launch announcement as a first-of-its-kind coalition for data centres that can dynamically manage their electricity use. The framing is grand but clear: AI factories are the infrastructure of the intelligence era, and scaling them responsibly depends as much on innovation across the grid as inside the building.
It is a natural next step for this story. NVIDIA had already reported that Emerald AI's Conductor platform answered more than 200 demand signals from Silicon Valley Power, easing a facility's draw from four megawatts to three while lower-priority work paused and higher-priority inference continued. Adding Google to a standing coalition with Emerald AI and NVIDIA signals intent to turn that kind of load flexibility from a one-off pilot into shared industry practice.
The usual day-one caution applies: this is the founding members describing their own ambition, and the announcement excerpt gives the purpose rather than methods, membership beyond the founders or measured results. The test to watch is whether AEMA produces concrete standards or independently confirmed numbers, and whether flexible compute starts appearing in grid agreements rather than press releases.
Sources and evidence
- Emerald AI, Google and NVIDIA Launch Alliance to Advance Flexible AI Data Centers: Emerald AI, Google and NVIDIA announced the launch of the AI Energy Management Alliance (AEMA), a first-of-its-kind coalition to advance data centres that dynamically manage their electricity use, per the companies' launch announcement.
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk Update What changedThe AEMA story just picked up the substance its launch lacked. NVIDIA's expanded announcement explains how a flexible AI data centre actually flexes: shifting computing workloads in time, discharging on-site storage, drawing on paired generation, or responding to system contingencies. The pitch is that a large electricity customer becomes a controllable grid resource rather than an inflexible load, which in theory lets utilities connect AI facilities faster without compromising reliability.
The alliance's operating philosophy is now on the record as well. AEMA is technology-neutral and performance-based, judging facilities on measurable service such as response speed, duration, predictability and behaviour during emergencies rather than the kit they run. Its principles call for defining ride-through, curtailment and contingency-response obligations before a facility connects, standardising technical requirements and operational data sharing, building faster risk-adjusted interconnection pathways for customers with credible, verifiable flexibility commitments, and allocating interconnection costs in line with actual system impacts.
Scope is wider than the founding trio too: AEMA convenes AI platforms, infrastructure providers, data centre operators, power producers, utilities and regional grid operators, and plans to develop shared technical approaches, work with utilities on interconnection and advocate for grid-responsive demand policies. That answers the methods question this story opened with. What is still missing is a fuller membership list and any measured results, which remain the real test of whether flexible AI data centres become standard practice.
Sources and evidence
- Emerald AI, Google and NVIDIA Launch Alliance to Advance Flexible AI Data Centers - blogs.nvidia.com: Emerald AI, Google and NVIDIA's AI Energy Management Alliance has published its operating approach: participating data centres would flex their grid draw by shifting workloads, discharging storage, using paired generation or responding to contingencies, under technology-neutral, performance-based principles covering pre-connection obligations, standardised metrics and data sharing, faster interconnection pathways and impact-reflecting cost allocation.
Independent WittyWires Watcher; not an official account or feed.