Google, Nvidia and Anthropic Back a Startup That Wants to Squeeze 100 More Gigawatts From the Power Grid
A new coalition says smarter software can let data centers pause and shift their electricity use, unlocking grid space that new power plants alone cannot provide fast enough.
Key points
- Emerald AI's AI Energy Management Alliance, announced in September 2026, counts Google, Nvidia and Anthropic among its founding members.
- The coalition says smarter demand-response software could connect an additional 100 gigawatts of data center capacity to existing grids.
- Emerald AI closed a $150 million Series A, led by Energize Capital and DCVC, to fund a wider rollout.
- A Goldman Sachs study found that capping peak grid use at 90 percent of maximum for a few hours could free up 76 gigawatts.
- Chief scientist Ayse Coskun told TechCrunch the approach will reduce, but not eliminate, demand for new generating capacity.
Power is the bottleneck nobody in tech wants to discuss plainly. Data centers get approved faster than the grid can feed them, and building new generation takes years and eye-watering capital. The AI Energy Management Alliance (AEMA), first reported by TechCrunch, is betting software moves quicker than concrete.
Founding members include Emerald AI, Google, Nvidia and Anthropic, joined by utilities AES, Constellation, National Grid and NRG Energy. Their shared focus is demand response: a decades-old arrangement where large electricity users briefly cut consumption during peak hours in exchange for payment from the utility. AEMA wants that to become standard practice across the data center industry.
What does "demand response" actually mean for data centers?
A factory doing demand response might pause a production line for an hour. Emerald AI's software does something cleaner: it shelves non-urgent computing tasks or reroutes workloads to a facility in a region with spare grid capacity, cutting peak draw without touching critical jobs.
Speed sets it apart from the old playbook. Data centers that join today's demand-response programmes typically fire up diesel backup generators to cover the shortfall, burning fuel and producing emissions. Emerald AI's platform connects utilities directly to data centers, so the response is nearly as fast as a battery. No exhaust, no delay.
The Goldman Sachs study, published before this announcement, put 76 gigawatts of freeable capacity on the table just from capping peak use at 90 percent of maximum. AEMA's broader target of 100 gigawatts layers in coordinated load-shifting across many sites simultaneously.
Why does this matter right now?
AI compute loads are unusually peaky: they spike hard, then drop off quickly. That's exactly the profile demand-response systems are designed to exploit. It's also timely given rising resistance to data center expansion. Our 2 September story on communities pushing back against AI data centres showed that pressure reaching from Scotland to India, with governments beginning to act. And our 17 September piece on data center heat management illustrated how the physical limits of these facilities keep multiplying.
Emerald AI's $150 million Series A gives it real firepower for a market rollout. But Ayse Coskun, the company's chief scientist, was candid with TechCrunch: this technology blunts demand for new generation, it doesn't kill it.
That honesty is the thing worth watching here. A coalition of this size, with Anthropic sitting alongside the dominant names in AI hardware and cloud infrastructure, signals that grid access has become an existential constraint for the industry. Demand-response software won't build a single new power plant. What it might do is buy enough breathing room for the grid to catch up, and for communities still fighting over who bears the cost of expansion. Whether AEMA can scale fast enough to matter is the question its own members haven't fully answered.
Common questions
Will this affect my electricity supply or bills?
Not directly. Demand response reduces strain during peak hours, which can prevent blackouts and, over time, lower the cost of emergency power that utilities pass on to consumers.
Is this the same as data centers going green?
No. Shifting loads and pausing tasks cuts peak demand but doesn't change what fuel generates the underlying power. It's a grid-management tool, not a renewable energy strategy, though it can work alongside one.
