O TYM ODCINKU
In this episode, I break down Pacific Gas & Electric’s (PG&E) announcement of a new Virtual Power Plant (VPP) called SHARE (Smart Home Assets for Reliability and Efficiency). I examine why Google is funding the program and what this initiative reveals about grid-edge capacity arbitrage and AI data center growth.
Here are the key points covered in today's analysis:
- The SHARE VPP Structure: PG&E is collaborating with Rewiring America, Carrier, Tesla, Sunrun, Renew Home, Demand Side Analytics, Encoord, and Google. Starting in fall 2026, the program plans to enroll approximately 21,000 flexible energy devices totaling about 12 MW to shift peak demand away from high-stress periods.
- Carrier Heat Pump Subsidies: Single-family home participants in Alameda and Santa Clara counties with central AC are eligible for Carrier battery-enabled heat pumps with a $5,000 discount, with the first 25 households receiving twice that incentive.
- Google's Funding & Capacity Arbitrage: Google is covering all program costs, customer incentives, and heat pump subsidies. This setup serves as a model for capacity arbitrage, where tech companies pay to free up capacity on the grid edge that can subsequently be claimed to connect additional data load sooner.
- The Economics of AI Data Centers: Compute leases for AI data centers operate in a unique economic tier—with Crusoe leasing compute at roughly $1.25 billion per month per gigawatt and BloombergNEF estimating revenue around $1.76 billion per month per gigawatt. This provides strong economic rationale to pay for grid-edge capacity to get online faster.
- Grid-Edge Arbitrage vs. Data Center Flexibility: While data centers are also testing operational flexibility to reduce load during peak hours, neither grid-edge demand response arbitrage nor data center operational flexibility has been fully proven at gigawatt scale yet.
If you find this analysis helpful, please like, subscribe, and leave a comment with your thoughts on how VPPs and data center growth are intersecting in your region.
🎙️ About Energy Future: Powering Tomorrow’s Cleaner World
Hosted by Peter Kelly-Detwiler, Energy Future explores the trends, technologies, and policies driving the global clean-energy transition — from the U.S. grid and renewable markets to advanced nuclear, fusion, and EV innovation.
💡 Stay Connected
Subscribe wherever you listen — including Spotify, Apple Podcasts, Amazon Music, and YouTube.
🌎 Learn More
Visit peterkellydetwiler.com
for weekly market insights, in-depth articles, and energy analysis.
POKAŻ NOTATKI 🔗
TRANSKRYPCJA 🔗
00:00:03.120 --> 00:00:08.720
Hi, I'm back from vacation and once again on the road this week with three speaking engagements in five days.
00:00:08.960 --> 00:00:13.519
Check out my cool hotel room this morning in Denver with the paper airplans on the ceiling.
00:00:13.679 --> 00:00:16.719
And once again, I'm looking at newsworthy stories affecting the grid.
00:00:16.879 --> 00:00:43.280
The one that recently caught my eye was Pacific Gas and Electric's announcement that it was teaming up with Rewiring America, Carrier Global Corporation, Tesla, Sunrun, Renew Home, Demand Side Analytics, NCORD, and Google to develop a quote, first of its kind virtual power plant, VPP, designed to help communities benefit from rising electricity demand while addressing limited transmission capacity and improving affordability and reliability for all customers.
00:00:43.520 --> 00:00:44.399
Unquote.
00:00:44.640 --> 00:00:49.200
Since all these programs get acronyms, this one might as well get one too.
00:00:49.439 --> 00:00:55.439
And it's called the Smart Home Assets for Reliability and Efficiency, henceforth to be referred to as Share.
00:00:56.079 --> 00:01:07.519
PD ⁇ A will collaborate with Tesla, Sunrun, and Renew Home in enrolling approximately 21,000 existing flexible energy devices, mostly batteries, into the program, totaling about 12 megawatts.
00:01:07.680 --> 00:01:12.719
That'll shift peak demand away from high stress periods and free up scarce transmission capacity.
00:01:12.959 --> 00:01:20.319
The resources expected to begin supporting the grid as early as fall 2026, with participants compensated for their contributions.
00:01:20.799 --> 00:01:31.280
Carrier, with its battery-enabled heat pumps, fresh off the shelf, will add these new devices to the VPP, with the batteries shifting energy use to help manage peak demand as well.
00:01:31.519 --> 00:01:34.879
Participants will come from Alameda and Santa Clara counties.
00:01:35.040 --> 00:01:44.640
They must live in single-family homes with central AC, and they're eligible to receive a$5,000 discount with the first 25 households participating receiving twice that.
00:01:44.799 --> 00:01:55.120
Enrollment will start in spring of next year with the equipment monitored and managed for about two years, after which the homeowners keep the equipment and the battery-enabled heat pumps will continue to be part of a VPP.
00:01:55.439 --> 00:02:07.519
Rewileing America will work directly with households to deploy and enroll energy-saving efficient technologies into share and coordinate closely with local installers to ensure quality and performance standards are met.
00:02:07.760 --> 00:02:15.360
NCORD will be responsible for software and modeling, and PG ⁇ E and Demand Side Analytics will coordinate the thousands of distributed resources.
00:02:15.599 --> 00:02:20.479
Initial program findings will be shared in late 2026 and early 2027.
00:02:20.560 --> 00:02:25.120
And I'll bet one of them would be the challenge of coordinating so many different entities.
00:02:25.360 --> 00:02:28.000
High headed up Constellation's demand response program.
00:02:28.319 --> 00:02:32.719
And once we had three or four more players in the pot, things got pretty challenging.
00:02:33.280 --> 00:02:35.280
So this is a pretty big program.
00:02:35.360 --> 00:02:36.400
It's going to cost some money.
00:02:36.560 --> 00:02:38.000
Who's writing the checks for this?
00:02:38.159 --> 00:02:39.520
Well, it turns out Google is.
00:02:39.680 --> 00:02:45.759
They're paying for all the costs, customer incentives, and those heat pump subsidies, which naturally leads one to ask, why?
00:02:45.840 --> 00:02:47.360
What does Google get out of this?
00:02:47.599 --> 00:02:54.159
Specifically since it indicated that this program is not related to any existing or planned data centers in the area.
00:02:54.400 --> 00:03:12.159
Well, Google may not reap the benefits of the capacity freed up in this case, but it may serve as a model for future capacity arbitrage, in which this and other data companies pay for and free up capacity on the grid edge that they can then subsequently claim for their own purposes in enrolling additional data log.
00:03:12.719 --> 00:03:20.000
Google's already doing something like this through a deal announced in June for 100 megawatts of capacity to be freed up by DR aggregator Voltus.
00:03:20.400 --> 00:03:24.719
But 100 megawatts is peanuts compared to the gigawatts of capacity Google needs.
00:03:24.879 --> 00:03:33.520
For example, it's building data centers of 840 megawatts, 700 megawatts, and a couple of 600 megawatt facilities right now across the country.
00:03:33.840 --> 00:03:36.080
The economics, though, are pretty compelling.
00:03:36.159 --> 00:03:37.919
Let's think about them for a minute.
00:03:38.240 --> 00:03:54.000
Crusoe, one company that leases AI data center compute to various buyers, recently indicated in a Stanford University educational series that it can lease 1,000 megawatts, so 1 gigawatt, for about$15 billion a year or$1.25 billion per month.
00:03:54.240 --> 00:04:02.319
Bloomberg New Energy Finance recently estimated a revenue figure of around$1.76 billion per gigawatt per month.
00:04:02.400 --> 00:04:06.000
So it pays for these things to get connected as quickly as possible.
00:04:06.400 --> 00:04:17.040
This number, or these numbers, imply that AI data centers, that crowd lives in an economic stratosphere where they can pay virtually anything to get online sooner than would otherwise be the case.
00:04:17.279 --> 00:04:22.399
They value capacity at orders of magnitude differently than any other players out there.
00:04:22.639 --> 00:04:38.480
An economically rational outcome then would be for Google and others to arbitrage this possibility and develop as much DR capacity as they can on the grid edge in exchange for being able to connect more quickly to a power grid whose peak capacity is already pretty tapped out.
00:04:38.720 --> 00:04:44.399
So we may see more of these programs in the future in which capacity freed up is dedicated to these new data loads.
00:04:44.639 --> 00:04:53.040
Of course, data center operators increasing their own operational flexibility could also reduce stress on the grid and bring more facilities online.
00:04:53.439 --> 00:04:58.639
Although that capability hasn't yet been fully stress tested or demonstrated beyond some meaningful pilots.
00:04:58.800 --> 00:05:11.360
MLDI, a leading player in the space, has cut about 25% of its compute load over three hours in Arizona with Salt River Project, and it has had four other projects around the world demonstrating flexible capabilities.
00:05:11.519 --> 00:05:22.240
They're building a 96 megawatt data center with digital realty and EPRI and others in Virginia as a test case, but none of that's been fully proven out yet.
00:05:22.560 --> 00:05:28.800
So at this point, it's possible that can happen, and it's possible that we can get more DR arbitrage.
00:05:28.879 --> 00:05:30.959
It's not an either or, it's a both.
00:05:31.040 --> 00:05:34.480
But neither capability has been fully proven yet.
00:05:34.720 --> 00:05:37.920
Well, thanks for watching, and we'll see you again soon.