Three Companies, One Admission
On September 16, 2026, NVIDIA, Google and Emerald AI launched the AI Energy Management Alliance, a coalition built around a blunt admission: the AI industry has been asking the power grid for a fixed, permanent connection when it could often settle for a flexible one. Emerald AI founder and chief executive Varun Sivaram put the idea in one line, that AI factories are too valuable to be treated as either passive loads or permanent islands. The alliance says its founding members will be joined by launch partners across the AI and energy sectors in the coming months.
The premise is narrow and specific. A data center that can shift its computing workload, draw on batteries or on-site generation or briefly reduce its draw during a grid emergency is a fundamentally cheaper customer to connect than one that demands the same fixed amount of power around the clock. The alliance's own demonstrations report cutting power draw by a third in under 60 seconds during emergency conditions, without stopping the underlying workload.
What Flexibility Is Actually Worth
The alliance backs its pitch with a specific number: as much as 100 gigawatts of capacity could be unlocked from the existing US power system through moderate data center flexibility, without a single new power plant. That figure only holds if data centers behave differently once connected, which is the part the alliance is actually trying to standardize.
| Metric | Figure |
|---|---|
| Capacity unlocked from the existing grid | Up to 100 GW |
| Rate reduction per 10 percent utilization gain | 3.4 percent |
| Avoided power-system cost per GW of flexible AI capacity | 733 million dollars |
| Typical interconnection queue in key US markets | 5 to 7 or more years |
The alliance lists four concrete levers behind those numbers: shifting computing jobs to a different time or place, discharging on-site storage, drawing on paired generation and responding to grid contingencies on request. None of those require new hardware most modern data centers do not already have; what they require is a contract that lets a utility actually call on them.
The Rules Before the Wires
Four principles anchor the alliance's technical framework, and each one targets a specific stalling point in how data centers get connected today. Ride-through and curtailment obligations get defined before a facility connects, not negotiated after a dispute. Technical requirements and performance metrics get standardized across operators, so a flexibility commitment means the same thing in Ohio as in Texas. Interconnection pathways get built specifically for developers who can show a credible flexibility commitment. And interconnection costs get allocated by a project's actual impact on the system rather than by a flat formula that ignores whether a customer will ever draw its full contracted load at once.
None of this is regulation. It is an industry group proposing a standard it hopes utilities and regulators will adopt, which means it carries no legal force until state commissions or federal rules pick it up.
The Same Crunch, From the Supply Side
The alliance is a voluntary answer to the same pressure that pushed the US House to pass the Ratepayer Protection Act one day earlier: American data centers are adding load faster than the grid can connect it, and someone has to absorb the cost of that mismatch. Where that law forces data centers to pay for the infrastructure they need, this alliance is a bet that some of that infrastructure never has to get built at all, if the data center agrees to flex instead of demanding a fixed, always-on connection.
For a buyer of AI compute anywhere, including outside the United States, the practical signal is that firm capacity and flexible capacity are becoming two different products with two different price and speed profiles. A flexible-load data center should, on this logic, connect to the grid faster and face a smaller share of infrastructure cost than a fixed-load one demanding the same peak power. A contract or a site-selection decision made without asking which category a facility falls into is missing the number about to show up on the bill.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: Google Hedged Its Gas Bet With a Geothermal One | AWS's Next 2 Million GPUs Skip Naming Europe
