Latest / Silicon & Shale Podcast / Episode 2: AI Knows Oil — Collide AI, the Data Center Power Grab, and the Future of Energy Workers

Transcript

Episode 2: AI Knows Oil — Collide AI, the Data Center Power Grab, and the Future of Energy Workers · Silicon & Shale Podcast

• 0:00
  SPEAKER_00: So let me get this straight.
• 0:02
  The same hyperscalers that were telling us fossil fuels are
• 0:05
  dead are now signing
• 0:06
  SPEAKER_00: 20-year contracts for natural gas because their AI needs that much power?
• 0:10
  Welcome to Silicon and Shale,
• 0:11
  the weekly reality check on AI and oil and gas.
• 0:14
  I'm Derek,
• 0:15
  your reformed petroleum engineer who traded mud logs for
• 0:18
  machine learning.
• 0:19
  SPEAKER_00: And I'm Packer.
• 0:21
  I run production in Oklahoma where the wells are real,
• 0:24
  and the promises better be too.
• 0:27
  This week,
• 0:27
  we're diving into three stories that show how fast things are
• 0:30
  moving.
• 0:31
  SPEAKER_00: First,
• 0:32
  SPEAKER_00: digital wildcatters just launched Collide AI,
• 0:34
  SPEAKER_00: an AI search tool trained specifically on oil and gas knowledge.
• 0:39
  We'll break down whether it's actually useful or just chat GPT
• 0:42
  with a hard hat.
• 0:43
  Then we're talking about the energy demand explosion from AI
• 0:46
  infrastructure.
• 0:48
  SPEAKER_01: Wood McKenzie's numbers are wild.
• 0:50
  SPEAKER_01: Data centers are going to need more power than entire countries.
• 0:54
  SPEAKER_01: And guess what's going to generate it?
• 0:56
  SPEAKER_00: And we'll wrap with AlphaSchool,
• 0:57
  an AI education platform that claims kids can do two years of
• 1:01
  school in two hours a
• 1:02
  SPEAKER_00: day.
• 1:03
  SPEAKER_00: Because if that's real,
• 1:05
  it changes how we think about training the next generation of
• 1:07
  energy workers.
• 1:09
  SPEAKER_01: Plus Oil hit $95 this week.
• 1:12
  SPEAKER_01: That's 30% up since January.
• 1:14
  SPEAKER_01: And it's not just geopolitics driving it anymore.
• 1:17
  Let's start with Collide AI,
• 1:19
  because I actually got my hands on it this week.
• 1:21
  SPEAKER_01: Wait, you tried it?
• 1:22
  SPEAKER_01: I was going to demo it live on the show.
• 1:25
  SPEAKER_00: Of course I tried it.
• 1:26
  SPEAKER_00: That's what we do.
• 1:28
  We test things so our listeners don't have to waste time on
• 1:30
  science projects.
• 1:32
  SPEAKER_00: And honestly, I'm impressed, but with caveats.
• 1:35
  SPEAKER_01: Okay,
• 1:35
  SPEAKER_01: for people who haven't heard of this,
• 1:38
  SPEAKER_01: Digital Wildcatters,
• 1:39
  SPEAKER_01: which runs the biggest energy community online,
• 1:42
  just launched what they're calling the first AI specifically
• 1:45
  trained for oil and
• 1:46
  SPEAKER_01: gas professionals.
• 1:47
  SPEAKER_01: It's built on rag architecture.
• 1:50
  SPEAKER_00: English, please.
• 1:51
  SPEAKER_01: Right.
• 1:52
  SPEAKER_01: RAG means Retrieval Augmented Generation.
• 1:56
  SPEAKER_01: Instead of just relying on what the base AI model knows,
• 1:59
  SPEAKER_01: it searches through a massive database of oil and gas documents first,
• 2:03
  SPEAKER_01: technical papers,
• 2:04
  SPEAKER_01: regulations,
• 2:05
  SPEAKER_01: contract language,
• 2:07
  SPEAKER_01: then gives you answers with actual citations.
• 2:10
  SPEAKER_00: Which is huge.
• 2:11
  Because when I ask about Oklahoma Corporation Commission rules
• 2:14
  or completion
• 2:15
  SPEAKER_00: techniques in The Scoop,
• 2:17
  SPEAKER_00: I don't want ChatGPT guessing based on some random blog post from 2019.
• 2:22
  SPEAKER_00: I wanted citing the actual regs or peer-reviewed papers.
• 2:25
  SPEAKER_01: Exactly.
• 2:26
  SPEAKER_01: So what did you test it on?
• 2:27
  SPEAKER_00: I threw some curveballs at it.
• 2:29
  SPEAKER_00: Asked about flare regulations in different states.
• 2:32
  SPEAKER_00: Wanted specifics on ESP failure rates in high GWR wells.
• 2:36
  SPEAKER_00: Even tested it on some contract language around joint interest billing.
• 2:41
  SPEAKER_00: The sourcing was solid.
• 2:43
  It pulled from actual SPE papers,
• 2:45
  state regulatory sites,
• 2:46
  industry reports I recognized.
• 2:49
  SPEAKER_01: That's what gets me excited about this.
• 2:52
  SPEAKER_01: We've been waiting for someone to do this right.
• 2:55
  Generic AI tools are dangerous in our industry because the
• 2:58
  stakes are so high.
• 3:00
  You can't afford to get reservoir pressure calculations wrong
• 3:03
  or misinterpret
• 3:03
  SPEAKER_01: environmental regulations.
• 3:05
  SPEAKER_00: But here's my concern.
• 3:07
  SPEAKER_00: And I tested this, too.
• 3:09
  I asked it about a specific formation that I know has weird
• 3:12
  pressure gradients,
• 3:14
  SPEAKER_00: something you'd only know if you'd actually worked those wells.
• 3:17
  The AI gave me textbook answers that sounded right,
• 3:20
  but missed the local knowledge that matters.
• 3:22
  SPEAKER_01: That's fair.
• 3:24
  SPEAKER_01: But think about what problem this solves day to day.
• 3:27
  How much time do your people spend digging through regulations
• 3:30
  or trying to
• 3:31
  remember the details of some completion technique they used two
• 3:34
  years ago?
• 3:35
  SPEAKER_00: More than I'd like to admit.
• 3:37
  SPEAKER_00: especially newer folks.
• 3:38
  I've got engineers who spend half their morning Googling basic
• 3:41
  stuff that an
• 3:41
  SPEAKER_00: old-timer would just know.
• 3:43
  Right,
• 3:43
  so imagine this scenario,
• 3:46
  a two-person LAN team trying to manage a 500-lease portfolio.
• 3:50
  SPEAKER_01: Right now,
• 3:50
  SPEAKER_01: they're drowning in paperwork,
• 3:52
  SPEAKER_01: trying to track different state regulations,
• 3:54
  SPEAKER_01: lease terms,
• 3:55
  SPEAKER_01: operator requirements.
• 3:57
  SPEAKER_01: With something like Collide AI,
• 3:59
  SPEAKER_01: they could be asking questions in plain English and getting specific,
• 4:02
  SPEAKER_01: sourced answers in seconds.
• 4:04
  SPEAKER_00: Okay.
• 4:05
  SPEAKER_00: But who's liable when it's wrong?
• 4:08
  SPEAKER_00: That's what nobody wants to talk about.
• 4:10
  SPEAKER_00: If the AI misinterprets a regulation and we make the wrong call,
• 4:14
  SPEAKER_00: who's getting the phone call from the state commission?
• 4:16
  SPEAKER_01: That's exactly why the citation system matters.
• 4:19
  SPEAKER_01: You're not trusting the AI's interpretation.
• 4:22
  You're using it to find the right sources faster than making
• 4:25
  your own judgment call.
• 4:27
  SPEAKER_00: Fair point.
• 4:29
  SPEAKER_00: And honestly, that's how I ended up using it.
• 4:31
  SPEAKER_00: Less, give me the answer.
• 4:33
  SPEAKER_00: And more, help me find the right documents to read.
• 4:37
  SPEAKER_00: Which is actually pretty valuable.
• 4:39
  SPEAKER_01: The integration piece is interesting, too.
• 4:42
  They're building it into the Digital Wildcatters community
• 4:44
  platform so you can ask
• 4:46
  questions and get both AI responses and input from actual
• 4:50
  people in the industry.
• 4:52
  SPEAKER_01: It's not replacing human expertise.
• 4:54
  SPEAKER_01: It's augmenting it.
• 4:55
  SPEAKER_00: That I like.
• 4:57
  SPEAKER_00: Because there's still stuff that only comes from experience.
• 5:00
  Like knowing that certain service companies always lowball
• 5:04
  their time estimates,
• 5:05
  or that Formation X always has water issues that don't show up
• 5:08
  in the logs.
• 5:09
  SPEAKER_01: Exactly.
• 5:10
  SPEAKER_01: But for everything else,
• 5:12
  SPEAKER_01: regulations,
• 5:13
  SPEAKER_01: technical calculations,
• 5:14
  SPEAKER_01: contract language,
• 5:16
  having an AI that actually understands oil and gas terminology
• 5:19
  and can cite
• 5:20
  SPEAKER_01: reliable sources,
• 5:22
  SPEAKER_01: that's a productivity boost we've been waiting for.
• 5:25
  SPEAKER_00: Bottom line for listeners...
• 5:27
  SPEAKER_00: If you're in Digital Wildcatters community, try it out.
• 5:30
  SPEAKER_00: Start with stuff you already know the answers to.
• 5:33
  SPEAKER_00: See how good the sources are.
• 5:35
  SPEAKER_00: Don't trust it blindly, but it might save you some serious research time.
• 5:39
  SPEAKER_01: And watch how they expand the knowledge base.
• 5:42
  SPEAKER_01: If they can keep adding recent papers,
• 5:44
  SPEAKER_01: new regulations,
• 5:45
  SPEAKER_01: evolving best practices,
• 5:47
  SPEAKER_01: this could become indispensable pretty quickly.
• 5:50
  SPEAKER_00: All right,
• 5:50
  SPEAKER_00: speaking of indispensable,
• 5:52
  SPEAKER_00: let's talk about why Microsoft,
• 5:54
  SPEAKER_00: Google,
• 5:54
  SPEAKER_00: and Amazon are suddenly very interested in buying our natural gas.
• 5:59
  SPEAKER_01: Laughs.
• 6:00
  SPEAKER_01: Yeah, the irony is thick.
• 6:02
  These are the same companies that were funding studies about
• 6:05
  stranded assets and
• 6:06
  SPEAKER_01: the death of fossil fuels.
• 6:08
  Now they're signing 20-year power purchase agreements for gas
• 6:11
  plants because their
• 6:12
  SPEAKER_01: AI data centers are energy hogs.
• 6:15
  SPEAKER_00: Would McKenzie just publish numbers that blew my mind?
• 6:18
  AI infrastructure is going to need more electricity than entire
• 6:21
  countries.
• 6:23
  We're talking about facilities that consume as much power as
• 6:25
  mid-sized cities,
• 6:27
  SPEAKER_00: and they need that power 24-7.
• 6:29
  SPEAKER_01: The numbers are staggering.
• 6:30
  A single large language model training run can use as much
• 6:34
  electricity as 17,000
• 6:36
  SPEAKER_01: homes use in a year.
• 6:38
  SPEAKER_01: And that's just training.
• 6:39
  SPEAKER_01: Once these models are deployed and serving millions of users,
• 6:43
  SPEAKER_01: the power requirements are constant.
• 6:45
  SPEAKER_00: So here's what's happening in real time.
• 6:48
  SPEAKER_00: Google just announced they're investing in new natural gas plants.
• 6:51
  SPEAKER_00: Microsoft is looking at nuclear.
• 6:53
  SPEAKER_00: Amazon is signing long-term contracts for reliable baseload power.
• 6:57
  SPEAKER_00: Solar and wind can't handle this demand profile.
• 7:00
  SPEAKER_01: Because AI doesn't care if the wind isn't blowing.
• 7:04
  SPEAKER_01: When someone asks ChatGPT a question,
• 7:06
  SPEAKER_01: those servers need power right now,
• 7:08
  SPEAKER_01: not when the weather cooperates.
• 7:10
  SPEAKER_00: Exactly.
• 7:11
  SPEAKER_00: And this is where our industry gets interesting again.
• 7:14
  Natural gas is the only fuel that can scale fast enough to meet
• 7:17
  this demand.
• 7:19
  SPEAKER_00: You can't build nuclear plants in two years.
• 7:22
  You can't store enough battery power to run these facilities
• 7:25
  when renewables aren't producing.
• 7:27
  SPEAKER_01: I've been tracking the permitting data in Texas.
• 7:30
  We're seeing new gas-fired power plants getting fast-tracked
• 7:33
  specifically to serve data centers.
• 7:36
  SPEAKER_01: The Permian pipeline infrastructure that everyone said was overbuilt?
• 7:40
  SPEAKER_01: Suddenly it's looking pretty smart.
• 7:42
  SPEAKER_00: What's really wild is the geography.
• 7:44
  These hyperscalers are starting to factor in energy costs when
• 7:48
  they choose data
• 7:48
  SPEAKER_00: center locations.
• 7:50
  Places like Oklahoma and Texas are becoming more attractive not
• 7:53
  just because of
• 7:54
  SPEAKER_00: cheap land,
• 7:55
  but because we have reliable gas supplies and existing grid
• 7:58
  infrastructure.
• 7:59
  SPEAKER_01: And the contracts they're signing are different, too.
• 8:02
  SPEAKER_01: These aren't spot market purchases.
• 8:05
  SPEAKER_01: Microsoft just signed a 20-year deal.
• 8:08
  That's the kind of long-term commitment that makes investors
• 8:10
  comfortable funding new production.
• 8:12
  SPEAKER_00: Which brings us back to prices.
• 8:15
  SPEAKER_00: WTI hit 95 this week, up 30% since January.
• 8:19
  SPEAKER_00: Some of that is geopolitics,
• 8:21
  but a lot of it is this new demand pattern that nobody was
• 8:24
  forecasting two years
• 8:25
  SPEAKER_00: ago.
• 8:26
  The really interesting part is what this means for associated
• 8:29
  gas in the Permian.
• 8:31
  All that gas we were flaring because there wasn't enough
• 8:33
  takeaway capacity?
• 8:35
  SPEAKER_01: Suddenly, every cubic foot has a buyer.
• 8:38
  SPEAKER_00: But let's be realistic here.
• 8:40
  SPEAKER_00: This is good for producers, but it's also creating new challenges.
• 8:44
  SPEAKER_00: Grid reliability, environmental permitting, pipeline capacity.
• 8:49
  SPEAKER_00: We're going to hit bottlenecks fast if this demand keeps growing.
• 8:52
  SPEAKER_01: True.
• 8:53
  But compare this to what we were facing three years ago when
• 8:56
  everyone was talking
• 8:57
  SPEAKER_01: about peak demand and stranded assets.
• 9:00
  Now we've got the biggest tech companies in the world
• 9:02
  essentially becoming energy
• 9:04
  SPEAKER_01: companies because they need so much power.
• 9:06
  The irony is that AI might end up being the thing that saves
• 9:09
  the oil and gas
• 9:10
  SPEAKER_00: industry's long-term prospects.
• 9:13
  SPEAKER_00: Not because we're using AI better,
• 9:15
  SPEAKER_00: though that helps,
• 9:17
  SPEAKER_00: but because AI itself needs so much energy.
• 9:20
  SPEAKER_01: And they needed to be reliable.
• 9:22
  You can't run a data center on hopes and prayers about
• 9:24
  renewable intermittency.
• 9:26
  SPEAKER_01: You need baseload power that works 24-7.
• 9:29
  SPEAKER_01: And that means natural gas in the near term.
• 9:32
  SPEAKER_00: So, what should listeners be watching?
• 9:34
  SPEAKER_00: I'd say keep an eye on power purchase agreements.
• 9:38
  When you see Microsoft or Amazon announce new data centers,
• 9:41
  look at the power contracts.
• 9:43
  SPEAKER_00: That tells you where energy demand is heading.
• 9:45
  SPEAKER_01: And watch the permitting.
• 9:48
  If states start fast-tracking gas plant permits specifically
• 9:51
  for data center power,
• 9:52
  SPEAKER_01: that's a signal that this trend has legs.
• 9:55
  SPEAKER_00: All right, let's shift gears completely.
• 9:58
  SPEAKER_00: AlphaSchool,
• 9:59
  an AI education platform that claims kids can learn two years'
• 10:03
  worth of material in
• 10:04
  SPEAKER_00: two hours a day.
• 10:06
  SPEAKER_00: Derek thinks this changes everything about how we train energy workers.
• 10:09
  SPEAKER_00: I think it sounds too good to be true.
• 10:11
  SPEAKER_01: Look, I know it sounds like Silicon Valley hype, but hear me out.
• 10:16
  Alpha School is using AI tutoring that adapts in real time to
• 10:19
  how each student learns.
• 10:22
  They're reporting students achieving top 1% SAT scores with
• 10:26
  dramatically less time investment.
• 10:28
  SPEAKER_00: Okay, but what does this have to do with oil and gas?
• 10:31
  SPEAKER_01: Everything.
• 10:33
  SPEAKER_01: Think about how we train people in this industry.
• 10:36
  SPEAKER_01: Petroleum engineering programs, geology degrees, certification courses.
• 10:40
  SPEAKER_01: It's all based on industrial age assumptions about how learning works.
• 10:44
  SPEAKER_01: Sit in lectures, memorize formulas, take tests.
• 10:48
  SPEAKER_00: Because that's how you learn complex technical material.
• 10:51
  SPEAKER_00: You can't shortcut understanding reservoir mechanics or well completions.
• 10:55
  SPEAKER_01: But what if you could personalize the learning?
• 10:58
  What if an AI tutor could figure out that you learn
• 11:00
  pressure-volume relationships
• 11:02
  SPEAKER_01: better with visual simulations,
• 11:04
  SPEAKER_01: while someone else needs to work through calculations step-by-step?
• 11:08
  SPEAKER_00: I'll give you that.
• 11:09
  I've trained dozens of engineers over the years,
• 11:11
  and they all learn differently.
• 11:13
  SPEAKER_00: Some need to see the data first, others need the theory.
• 11:17
  SPEAKER_00: But two hours a day to learn two years of material?
• 11:20
  SPEAKER_00: Come on.
• 11:21
  SPEAKER_01: The time savings might be real, though.
• 11:24
  SPEAKER_01: How much of traditional education is just inefficient?
• 11:28
  Sitting through lectures covering stuff you already know or
• 11:31
  moving on before you
• 11:32
  SPEAKER_01: actually understand the basics?
• 11:34
  SPEAKER_00: That's fair.
• 11:35
  I've seen people struggle through fluid flow calculations
• 11:38
  because they never really
• 11:39
  SPEAKER_00: got the fundamentals solid.
• 11:41
  And others who could skip half the coursework because they
• 11:43
  picked it up immediately.
• 11:45
  SPEAKER_01: Exactly.
• 11:46
  SPEAKER_01: Now imagine if we could apply this to industry training.
• 11:49
  SPEAKER_01: Instead of sending everyone through the same six-month rotation program,
• 11:53
  what if AI could create personalized learning paths based on
• 11:56
  their background and
• 11:57
  SPEAKER_01: the specific role they're heading into?
• 12:00
  SPEAKER_00: Okay, that actually interests me.
• 12:02
  SPEAKER_00: Because right now, training new hires is expensive and slow.
• 12:06
  SPEAKER_00: Takes months to get someone productive,
• 12:08
  SPEAKER_00: and half of what we teach them isn't relevant to their actual job.
• 12:11
  SPEAKER_01: Right.
• 12:12
  A completion engineer doesn't need the same geology background
• 12:15
  as someone going
• 12:16
  SPEAKER_01: into reservoir engineering,
• 12:18
  SPEAKER_01: but our current training programs treat everyone the same.
• 12:21
  SPEAKER_00: But here's my concern.
• 12:23
  SPEAKER_00: And this applies to Alpha School, too.
• 12:26
  SPEAKER_00: Technical knowledge isn't just about knowing facts.
• 12:29
  SPEAKER_00: It's about developing judgment.
• 12:31
  Understanding when the textbook answer doesn't apply because
• 12:34
  real-world conditions are messier.
• 12:36
  That's where I think AI tutoring could actually be better than
• 12:39
  traditional education.
• 12:41
  SPEAKER_01: Instead of just memorizing formulas,
• 12:43
  it could present realistic scenarios and walk students through
• 12:46
  the decision-making
• 12:47
  SPEAKER_01: process.
• 12:48
  SPEAKER_00: Maybe.
• 12:49
  SPEAKER_00: But there's no substitute for experience.
• 12:52
  SPEAKER_00: You don't really understand well control until you've dealt with a kick.
• 12:55
  You don't understand equipment failures until you've
• 12:57
  troubleshot them in the field.
• 12:59
  SPEAKER_01: Agreed.
• 13:00
  But what if we could compress the knowledge transfer part so
• 13:03
  people get to the real
• 13:04
  SPEAKER_01: experience faster?
• 13:06
  SPEAKER_01: Instead of spending two years in classrooms learning basic concepts,
• 13:10
  what if they could master that in six months and spend more
• 13:13
  time on hands-on
• 13:14
  SPEAKER_01: training?
• 13:15
  SPEAKER_00: Now you're talking my language.
• 13:17
  SPEAKER_00: Less theory, more practical application.
• 13:19
  But I'd want to see proof this actually works before we bet our
• 13:22
  training programs on it.
• 13:24
  SPEAKER_01: Agreed,
• 13:24
  but what if we could compress the knowledge transfer part so
• 13:27
  people get to the real
• 13:28
  SPEAKER_01: experience faster?
• 13:30
  SPEAKER_01: Instead of spending two years in classrooms learning basic concepts,
• 13:34
  what if they could master that in six months and spend more
• 13:37
  time on hands-on
• 13:38
  SPEAKER_01: training?
• 13:38
  SPEAKER_00: Now you're talking my language.
• 13:40
  SPEAKER_00: Less theory, more practical application.
• 13:44
  SPEAKER_00: Here's what I need to see.
• 13:46
  SPEAKER_00: Alpha School publishes their SAT data by December.
• 13:50
  SPEAKER_00: If those results hold up under scrutiny,
• 13:52
  SPEAKER_00: then by Q2 next year we should see pilot programs in technical education.
• 13:57
  I'd start with basic engineering fundamentals and track
• 14:00
  completion rates versus
• 14:01
  SPEAKER_00: traditional programs.
• 14:02
  SPEAKER_01: That's a fair timeline.
• 14:04
  SPEAKER_01: And honestly,
• 14:05
  SPEAKER_01: even if the time compression is half of what they claim,
• 14:08
  SPEAKER_01: it's still game-changing for an industry that can't hire fast enough.
• 14:12
  SPEAKER_00: So here's your homework this week.
• 14:14
  SPEAKER_00: If you're in the digital wildcatters community,
• 14:16
  SPEAKER_00: test Collide AI on regulations you actually deal with.
• 14:20
  SPEAKER_00: Don't just play with it.
• 14:21
  SPEAKER_00: Put it through real scenarios.
• 14:23
  For energy demand,
• 14:24
  track the power purchase announcements coming out of Microsoft
• 14:27
  and Amazon.
• 14:28
  SPEAKER_00: Those 20-year contracts are the real signal.
• 14:31
  SPEAKER_01: And on the education front,
• 14:32
  SPEAKER_01: if you run training programs,
• 14:34
  SPEAKER_01: start thinking about what personalized learning could look like in your
• 14:37
  SPEAKER_01: organization.
• 14:39
  SPEAKER_01: Not because Alpha School is proven yet,
• 14:41
  but because someone's going to crack this problem and you want
• 14:43
  to be ready.
• 14:44
  SPEAKER_00: That's our show for this week.
• 14:46
  SPEAKER_00: Follow us on LinkedIn for more discussion between episodes.
• 14:50
  SPEAKER_00: Next week,
• 14:50
  we're diving into the latest drilling automation pilots and why
• 14:53
  the service sector
• 14:54
  SPEAKER_00: is finally starting to embrace AI for real operations.
• 14:57
  SPEAKER_01: Until then, keep one foot in Silicon Valley and one foot in the field.
• 15:01
  SPEAKER_01: The intersection is where the opportunities are.
• 15:04
  SPEAKER_00: We'll see you next week on Silicon and Shale.