GM’s Big EV Incentives, Military Geothermal Plans & AI’s Energy Shakeup!

The energy transition is accelerating across multiple fronts, as demonstrated by a flurry of recent announcements from major players. General Motors has partnered with Pacific Gas & Electric for a groundbreaking vehicle-to-everything pilot offering customers up to $4,500 in incentives when using compatible EVs like the Chevrolet Silverado and Cadillac Lyric. While pilots generate buzz, what matters more is whether GM and PG&E will share results that help other utilities replicate successful strategies.

Military installations aren't being left behind, with the US Air Force and Department of Defense teaming up with GE Vernova and Sage Geosystems to explore utility-scale geothermal for renewable microgrids at bases nationwide. This effort parallels innovations in the residential sector, where BASE is working with Texas utility Bandera Electric Cooperative to deploy residential batteries through a zero-upfront-cost subscription model that benefits both homeowners and the grid.

The cybersecurity battlefield remains tense, as revealed in a Dragos case study detailing how Chinese Volt Typhoon hackers penetrated a US utility for nine months before detection. While utilities maintain stronger protections than many sectors, the sophistication of threats targeting critical infrastructure continues to grow, requiring constant vigilance from grid operators.

Perhaps most intriguing is Baidu's announcement of new AI models claiming to match GPT-4.5 performance at just 1% of the cost. As data centers drive unprecedented electricity demand — with one utility reporting applications for 40,000 megawatts — the energy efficiency of AI systems takes on strategic importance. While Baidu hasn't disclosed specific energy metrics, their technical approaches suggest significant efficiency gains that could reshape both computing and electricity consumption patterns in this rapidly evolving field.

Support the show

🎙️ 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.

Episode Transcript

Speaker 1 (00:00):
I've got your energy stories for this, the third week
of March 2025, and in the firstone, gemini Motors has teamed
up with Pacific Gas and Electricin its residential
vehicle-to-everything pilot,with eligible customers
receiving up to $4,500 inincentives for GM Energy home
products like the GM Energy V2HBundle or Home System.

(00:20):
Additional incentives areavailable for customers who
participate in planned backuppower events.
Designated EVs include the 2024Chevrolet Silverado, equinox
Blazer, gmc Sierra and CadillacLyric EV, with the 2025 model
years to be added soon.
I hope the two entities, gm andPG&E, can provide results of

(00:42):
the program so utility plannersand grid operators can see what
works and what doesn't.
Pilots intrigue, but resultsand replication create lasting
change.
Well, the US Air Force andDepartment of Defense have
designated a team including GEVrnova, sage Geosystems, the
Energy and Geoscience Instituteand the University of Utah to

(01:03):
explore avenues for deploymentof utility-scale geothermal for
future renewable and hydrogenenergy microgrids at military
bases.
This is part of a broadereffort to accelerate procurement
and deployment ofmission-critical technologies.
Sage Geosystems will provideits geothermal technology, while
GE Vrnova will offercapabilities related to power

(01:25):
conversion, energy storage andmicrogrids.
Distributed energy storagecompany BASE will work with
Texas Bandera ElectricCooperative in its residential
battery program that offershomeowners battery backup
systems.
Bandera is planning on anetwork of distributed energy
storage systems that it can useto provide grid services,

(01:51):
increase resilience and provideeconomic value to its members.
Bandera's battery storagesubscription program will have
BASE providing members withbatteries for a monthly
subscription fee and no upfrontcost.
The utility will then operatethese distributed batteries
through a real-time managementand optimization platform.
Base is known for its buy-homeget-backup power joint program

(02:15):
with national homebuilder Lenar.
Cybersecurity company Dragosreleased a case study recently
outlining a cyber attack fromChinese Volt Typhoon hackers.
Onal Utility Littleton Electric, light and Water Department.
The utility was able toidentify and eliminate the
threat, but it highlights yetagain the security challenges
faced by utilities insafeguarding their operational

(02:36):
technology.
The hackers apparentlyinfiltrated the utility about
nine months prior to beingexposed.
This cyber battlefield isconstantly changing.
In its 2025 annual review,dragos identified nine so-called
advanced persistent threatactors, with an increased focus
on the interface betweeninterconnected IT and operating

(02:58):
technology, such as SCADA.
The good news Utilities arebetter protected than many other
industrial sectors, but the spyversus spy game continues every
day and hypervigilance in theutility sector is a must.
And for our last story of theday, chinese tech giant Baidu
just came out with its newest AImodels, named Ernie X1 and

(03:20):
Ernie 4.5.
Where do these names come fromand how does BERT feel about
being scorned?
The company claims these newmodels can compete with OpenAI
and DeepSeek in terms ofperformance and cost, with Ernie
1x at about half the price ofDeepSeek.
Baidu claims its multimodalfoundational model, ernie 4.5,
quote, outperforms GPT 4.5 inmultiple benchmarks, while

(03:45):
priced at just 1% of GPT 4.5,unquote.
To put some clarity around theterms here, the large language
models you frequently hear aboutare specifically set up to
process and generate data basedon text, so they're good at text
generation, translation and thelike.
Gpt 4.0 does that Multi-modelfoundational models go the next

(04:08):
step, building in capabilitiesto deal with other sources of
data, such as images, audio andvideo.
Thus they can facilitate morecomplex and various types of
applications.
So why does this all matter tous in the energy space?
Well, here you have a USindustry hell-bent at throwing
chips and money at the challengeof training AI models, where

(04:29):
Dominion, for example, justannounced applications of up to
40,000 megawatts of data centers.
Then others come in, someChinese open source models like
DeepSeek and Baidu's Ernie, andthey suggest there may be a
better and more cost-effectiveway to get this done.
What about the energy piece,though?

(04:50):
Well, I asked Perplexity AI tosearch for information
concerning Ernie's energyintensity, and it told me
there's nothing publiclyavailable.
But quote several technologiespowering Baidu's Ernie 4.5
contribute to its operationalefficiency, which likely
correlates with lower energyconsumption.
Thought that was the case.
And then it follows on and saysthings like dynamic attention

(05:13):
masking that's a strange termfor masking irrelevant parts of
the data sets so the focus is onwhat matters, plus cutting
redundant processing andboosting efficiencies.
Or how about this oneheterogeneous, multimodal mix of
experts?
That's another way of sayingyeah, different parts of the
model are specialized to look atdifferent types of data, like
audio images or text.

(05:35):
Other approaches were mentionedas well, but I think you get
the point.
So does perplexity, summarizingas follows quote these
innovations collectively enableErnie 4.5 to outperform
competitors like GPT 4.5 at 1%of the cost, which strongly
suggests energy efficiency gains.
However, explicit energyconsumption metrics remain

(05:58):
undisclosed in the providedsources.
That's one worth watching,because it has enormous
implications for dataconsumption, with all of these
training models out there rightnow.
Well, thanks for watching thisweek and we'll see you again
soon.