Latest / Silicon & Shale Podcast / Episode 10: Chevron turning into an infrastructure landlord & the true cost of data centers

Transcript

Episode 10: Chevron turning into an infrastructure landlord & the true cost of data centers · Silicon & Shale Podcast

• 0:00
  SPEAKER_01: And the project manager, he's showing me this dashboard.
• 0:02
  SPEAKER_01: All these green lights, blinking perfect.
• 0:04
  SPEAKER_01: Real-time well-performance across three states.
• 0:07
  SPEAKER_01: And I ask him, I say,
• 0:08
  SPEAKER_01: that well out in Loving County,
• 0:09
  SPEAKER_01: the one that had the motor fail last Tuesday, what's it showing?
• 0:12
  SPEAKER_00: Let me guess.
• 0:13
  SPEAKER_01: He clicks around, fumbles a bit.
• 0:15
  SPEAKER_01: The data's stale.
• 0:16
  SPEAKER_01: Three days stale.
• 0:17
  SPEAKER_01: Still showing optimal production even after they'd pulled the pump and had
• 0:19
  SPEAKER_01: a crew out there with a crane.
• 0:21
  SPEAKER_01: So I look at him and I say, son,
• 0:23
  SPEAKER_01: I appreciate the green lights,
• 0:24
  SPEAKER_01: but out here a good pair of boots and a working two-way radio usually beats
• 0:27
  SPEAKER_01: a dashboard with a three-day delay.
• 0:29
  SPEAKER_00: Because the field guys already know it's down.
• 0:31
  SPEAKER_00: They fixed it.
• 0:32
  SPEAKER_00: They moved on.
• 0:33
  SPEAKER_00: And the dashboard's still telling you everything's perfect.
• 0:36
  SPEAKER_00: That gap between the pristine demo and what actually happens when the data has
• 0:39
  SPEAKER_00: to travel,
• 0:40
  SPEAKER_00: get cleaned,
• 0:40
  SPEAKER_00: hit seven different systems,
• 0:41
  SPEAKER_00: and then finally show up on someone's screen,
• 0:43
  SPEAKER_00: everybody ignores it until it embarrasses them.
• 0:45
  SPEAKER_01: Gap might be generous.
• 0:46
  SPEAKER_01: Canyon, depending on the operator.
• 0:49
  SPEAKER_01: Which is why when I heard about Chevron building a data center campus down
• 0:52
  SPEAKER_01: in Ector County, my first thought wasn't, oh, innovation.
• 0:55
  SPEAKER_01: My first thought was,
• 0:56
  SPEAKER_01: are they going to run their own stuff on it,
• 0:58
  SPEAKER_01: or is this another glorified landlord play?
• 1:00
  SPEAKER_00: Well, it is a landlord play, at least in part.
• 1:03
  SPEAKER_00: You've got Microsoft as a major potential tenant.
• 1:05
  SPEAKER_00: A thousand acres in Kilby, that's not residential.
• 1:08
  SPEAKER_00: That's a massive footprint.
• 1:10
  SPEAKER_00: An interesting pivot from drilling holes for oil to digging trenches for fiber.
• 1:13
  SPEAKER_01: Interesting is one word for it.
• 1:15
  SPEAKER_01: Welcome to Silicon & Shale, where the oil patch meets the algorithm.
• 1:19
  SPEAKER_01: Week of August 10, 2026.
• 1:21
  SPEAKER_00: We're talking about energy, data, and the surprising places they intersect.
• 1:25
  SPEAKER_00: Today, Chevron turning into an infrastructure landlord,
• 1:28
  SPEAKER_00: the true cost of data centers,
• 1:29
  SPEAKER_00: and why rogue AI agents are maybe not as rogue as the headlines want you to think.
• 1:33
  SPEAKER_01: I'm still chewing on that Chevron Kilby site, a thousand acres.
• 1:37
  SPEAKER_01: You know how much a good tract of that size costs in Ector County these days?
• 1:40
  SPEAKER_01: Even if it's decent for development, that's a real asset.
• 1:43
  SPEAKER_01: And they're not building a new refinery, they're putting in server farms.
• 1:46
  SPEAKER_00: From the power side, it makes sense.
• 1:49
  SPEAKER_00: U.S. electricity prices have been rising faster than general inflation,
• 1:52
  SPEAKER_00: and a huge chunk of that comes from the data center investment boom.
• 1:55
  SPEAKER_00: These things consume power at a staggering rate.
• 1:57
  SPEAKER_00: Chevron already has infrastructure,
• 1:59
  SPEAKER_00: transmission lines,
• 2:00
  SPEAKER_00: maybe their own generation capacity, existing easements they can leverage.
• 2:03
  SPEAKER_00: You could call it a natural extension of land management.
• 2:05
  SPEAKER_01: Natural extension of land management?
• 2:08
  SPEAKER_01: PAC, they're becoming Microsoft's landlord.
• 2:10
  SPEAKER_01: I'd call it a concession.
• 2:12
  SPEAKER_01: They've got this land that isn't core to their drilling plans right now,
• 2:14
  SPEAKER_01: so they let someone else put giant computers on it.
• 2:17
  SPEAKER_01: What are they getting beyond rent?
• 2:19
  SPEAKER_00: Strategic diversification,
• 2:21
  SPEAKER_00: predictable,
• 2:21
  SPEAKER_00: high-margin revenue from a non-core asset, and probably future power agreements.
• 2:26
  SPEAKER_00: Picture a scenario where energy grids get even tighter,
• 2:28
  SPEAKER_00: and you've got this massive power consumer right on your property,
• 2:31
  SPEAKER_00: tied into your infrastructure.
• 2:33
  SPEAKER_00: You know the cost of power, you know the stability requirements.
• 2:36
  SPEAKER_00: A way to de-risk both sides.
• 2:37
  SPEAKER_01: Or shift the risk.
• 2:38
  SPEAKER_01: If those data centers go down,
• 2:40
  SPEAKER_01: or if Microsoft decides to pull out in 10 years,
• 2:42
  SPEAKER_01: Chevron's stuck with 1,000 acres of server racks.
• 2:45
  SPEAKER_01: You can't convert that back to a productive oil field overnight.
• 2:48
  SPEAKER_01: I keep wondering whether this is a new business line or a really big side hustle.
• 2:52
  SPEAKER_00: I'd argue new business line.
• 2:53
  SPEAKER_00: Look at where energy demand is heading.
• 2:55
  SPEAKER_00: A lot of it is digital.
• 2:57
  SPEAKER_00: Chevron has land, capital, and existing energy infrastructure.
• 3:00
  SPEAKER_00: They're providing physical infrastructure and reliable power to the digital economy,
• 3:04
  SPEAKER_00: direct lineage from what they already do.
• 3:06
  SPEAKER_01: Except that's like saying a guy who sells wrenches should sell houses
• 3:08
  SPEAKER_01: because both involve steel.
• 3:10
  SPEAKER_01: Managing data centers,
• 3:12
  SPEAKER_01: the uptime,
• 3:12
  SPEAKER_01: the cooling,
• 3:13
  SPEAKER_01: the security for sensitive data,
• 3:14
  SPEAKER_01: whole different skill set than getting hydrocarbons out of the ground.
• 3:18
  SPEAKER_01: I'm not saying they can't learn it,
• 3:19
  SPEAKER_01: but operating a campus where a 15-minute outage costs someone millions is
• 3:23
  SPEAKER_01: a different kind of pressure than they're used to,
• 3:24
  SPEAKER_01: even with all their experience keeping wells online.
• 3:27
  SPEAKER_00: I'd push back on that gap being as wide as you're describing, though.
• 3:30
  SPEAKER_00: Managing a supermajor's operational tech infrastructure,
• 3:33
  SPEAKER_00: their SCADA systems, their internal data centers, those are serious operations.
• 3:37
  SPEAKER_00: They understand critical infrastructure at scale.
• 3:39
  SPEAKER_00: The difference is the tenant, and the tenant brings their own expertise for the servers.
• 3:43
  SPEAKER_00: Chevron provides the Shell and the Juice,
• 3:45
  SPEAKER_00: a massive co-location facility owned by an oil company.
• 3:48
  SPEAKER_01: Co-location with pump jacks nearby.
• 3:50
  SPEAKER_01: I picture some Microsoft engineer doing server maintenance while
• 3:53
  SPEAKER_01: a workover rig clangs away a quarter mile off.
• 3:55
  SPEAKER_00: Data centers make their own noise anyway.
• 3:58
  SPEAKER_00: The real concern is vibration.
• 3:59
  SPEAKER_00: Seismic activity, even from drilling, can affect hard drives and sensitive equipment.
• 4:03
  SPEAKER_00: Niche engineering problem, but a real one at hyperscale.
• 4:06
  SPEAKER_01: You're telling me that if they're drilling a new well five miles away,
• 4:09
  SPEAKER_01: Microsoft's going to complain about server stability?
• 4:11
  SPEAKER_00: Sustained low frequency vibration?
• 4:13
  SPEAKER_00: They might.
• 4:14
  SPEAKER_00: Everything becomes a concern when you're talking hyperscale.
• 4:17
  SPEAKER_00: But this kind of strategic thinking connects to another story.
• 4:20
  SPEAKER_00: The UK's North Sea Energy Profits Levy.
• 4:22
  SPEAKER_00: We're talking a 75% total tax rate for oil and gas companies up there.
• 4:26
  SPEAKER_01: 75%.
• 4:27
  SPEAKER_01: And similar conversations have happened in the States,
• 4:30
  SPEAKER_01: though President Trump's generally been against windfall taxes on oil companies.
• 4:34
  SPEAKER_01: But when you hit an industry with that kind of rate, you shut down investment.
• 4:37
  SPEAKER_01: You're telling companies,
• 4:38
  SPEAKER_01: don't bother looking for new reserves,
• 4:40
  SPEAKER_01: don't bother maintaining what you've got,
• 4:42
  SPEAKER_01: produce what you can and hand over three quarters of the profits.
• 4:45
  SPEAKER_01: What do they expect to happen?
• 4:47
  SPEAKER_00: They expect a windfall for the Treasury.
• 4:48
  SPEAKER_00: But what actually happens is what BP's doing, reconsidering investments.
• 4:52
  SPEAKER_00: Shell, Equinor, they're all looking at it.
• 4:54
  SPEAKER_00: The entire economic calculus for a project changes overnight.
• 4:58
  SPEAKER_00: What looked marginally profitable at 40% tax is a dead loss at 75%.
• 5:02
  SPEAKER_00: And that hits everything.
• 5:03
  SPEAKER_00: Extending the life of existing fields, infrastructure upgrades.
• 5:06
  SPEAKER_00: All of it stalls.
• 5:07
  SPEAKER_01: I wonder how you'd forecast the real impact of a tax change like that using modern modeling.
• 5:10
  SPEAKER_01: Long-term GDP effects, job creation, energy security, beyond government revenue and isolation,
• 5:33
  SPEAKER_00: You could build a sophisticated simulation based on historical responses,
• 5:36
  SPEAKER_00: current commodity prices,
• 5:38
  SPEAKER_01: operational costs.
• 5:39
  SPEAKER_01: People react to incentives, always have.
• 5:42
  SPEAKER_01: Make it unattractive to invest, they won't invest.
• 5:45
  SPEAKER_01: That part hasn't changed since the first tax was levied.
• 5:48
  SPEAKER_01: A model would tell you the same thing a landman from Midland would tell you over coffee.
• 5:51
  SPEAKER_00: But the behavioral dimension,
• 5:53
  SPEAKER_00: the chilling effect on new ventures,
• 5:55
  SPEAKER_00: capital reallocation to other regions, talent drain...
• 5:58
  SPEAKER_00: Those second- and third-order effects are harder to quantify,
• 6:00
  SPEAKER_00: and they're what really determine whether a basin thrives or dies.
• 6:03
  SPEAKER_00: We've watched it happen in other industries.
• 6:05
  SPEAKER_01: And you see it now with CCS and EOR,
• 6:08
  SPEAKER_01: Carbon Capture, Utilization, and Storage Enhanced Oil Recovery.
• 6:11
  SPEAKER_01: Everyone wants to talk about it, but the economics have to pencil out.
• 6:14
  SPEAKER_01: CO2 processing, pipeline transport, expensive.
• 6:18
  SPEAKER_01: A lot of capital for something that might only marginally improve recovery,
• 6:21
  SPEAKER_01: or where the carbon credits aren't paying off yet.
• 6:23
  SPEAKER_00: Then people like Ed Covington from Pelican Energy Consultants come in talking
• 6:26
  SPEAKER_00: about supercritical CO2.
• 6:28
  SPEAKER_00: You're dealing with very specific pressure and temperature regimes to get CO2 into
• 6:31
  SPEAKER_00: a fluid state,
• 6:32
  SPEAKER_00: which is far more efficient for pushing oil out or for long-term storage.
• 6:36
  SPEAKER_00: But the engineering complexity multiplies at every stage.
• 6:38
  SPEAKER_01: And the data requirements compound from there.
• 6:39
  SPEAKER_01: Monitoring those pipelines,
• 6:40
  SPEAKER_01: predicting corrosion,
• 6:41
  SPEAKER_01: optimizing injection rates for EOR to maximize recovery without wasting CO2.
• 7:03
  SPEAKER_00: The challenge extends through the entire surface infrastructure,
• 7:05
  SPEAKER_00: the processing plants, the compressors.
• 7:08
  SPEAKER_00: All of it has to be integrated.
• 7:09
  SPEAKER_00: For a lot of legacy operations, that's an enormous lift.
• 7:12
  SPEAKER_01: Before you even get to the engineering, you've got the landowner conversations.
• 7:15
  SPEAKER_01: Try getting an easement for a CO2 pipeline
• 7:17
  SPEAKER_01: when some folks still think you're piping pollution under their ranch.
• 7:20
  SPEAKER_01: Never had that one go smooth on the first try.
• 7:22
  SPEAKER_00: Framing matters.
• 7:23
  SPEAKER_00: Reducing emissions.
• 7:25
  SPEAKER_00: Making existing production greener.
• 7:26
  SPEAKER_00: Creating jobs.
• 7:27
  SPEAKER_00: All true.
• 7:28
  SPEAKER_00: But the public perception can stall a project as effectively as bad economics.
• 7:32
  SPEAKER_00: And then layer on M&A activity.
• 7:34
  SPEAKER_00: You acquire a company with potential CCS assets, but their data systems are entirely different.
• 7:39
  SPEAKER_00: Their monitoring equipment is older,
• 7:41
  SPEAKER_00: and their pipeline integrity process lives in a manual spreadsheet.
• 7:44
  SPEAKER_00: Integration becomes its own project.
• 7:46
  SPEAKER_01: You end up running two separate systems,
• 7:48
  SPEAKER_01: or you spend millions trying to force-fit them, and neither option feels great.
• 7:52
  SPEAKER_01: You inherit their problems along with their assets,
• 7:54
  SPEAKER_01: and you pray you don't find undocumented corrosion underground.
• 7:57
  SPEAKER_00: Which is why the digital layer for pipeline integrity carries so much weight.
• 8:01
  SPEAKER_00: Transcription by CastingWords
• 8:23
  SPEAKER_00: Intel's Head of AI Solutions, Dr.
• 8:25
  SPEAKER_00: Olena Zhu,
• 8:25
  SPEAKER_00: is talking about hybrid AI,
• 8:27
  SPEAKER_00: systems that intelligently route tasks to local models on a device or to the cloud,
• 8:31
  SPEAKER_00: depending on what makes sense.
• 8:32
  SPEAKER_01: So a smart app that sometimes uses the internet and sometimes doesn't.
• 8:36
  SPEAKER_01: Sounds like what my phone already does when I lose signal past the county line.
• 8:39
  SPEAKER_00: Much more sophisticated.
• 8:41
  SPEAKER_00: Think about a drilling rig generating terabytes of sensor data.
• 8:44
  SPEAKER_00: Sending all that to the cloud in real time is expensive,
• 8:46
  SPEAKER_00: has latency issues, and raises privacy concerns for proprietary data.
• 8:50
  SPEAKER_00: With hybrid AI, a local agent could be constantly monitoring drilling parameters.
• 8:54
  SPEAKER_00: If it sees something anomalous, it acts locally, adjusts a pump, triggers a local alarm.
• 8:59
  SPEAKER_00: Only when it needs complex analysis or historical pattern matching from
• 9:02
  SPEAKER_00: a wider dataset does it send a summary to the cloud.
• 9:04
  SPEAKER_01: The intelligence lives closer to the action.
• 9:07
  SPEAKER_01: Latency matters.
• 9:08
  SPEAKER_01: You don't want a drill bit getting stuck because some cloud server in Virginia took
• 9:11
  SPEAKER_01: an extra second to respond.
• 9:13
  SPEAKER_01: But how complex can these local agents really be on, what, a hardened tablet in the field?
• 9:17
  SPEAKER_00: Intel's designing chips and software for more powerful local models.
• 9:21
  SPEAKER_00: Nobody's running a massive general-purpose AI on a smart sensor,
• 9:24
  SPEAKER_00: but you can have specialized, smaller models trained for specific tasks.
• 9:27
  SPEAKER_00: Anomaly detection on vibrations,
• 9:29
  SPEAKER_00: predictive maintenance for a specific piece of equipment,
• 9:32
  SPEAKER_00: and an agent orchestrates these models.
• 9:34
  SPEAKER_00: Multiple agents,
• 9:35
  SPEAKER_00: each with a defined role,
• 9:36
  SPEAKER_00: communicating locally or sending things to the cloud when necessary.
• 9:39
  SPEAKER_01: So you've got a whole ecosystem of digital workers in the field making decisions.
• 9:43
  SPEAKER_01: For something like the pipeline integrity we were talking about,
• 9:46
  SPEAKER_01: an agent constantly monitoring flow rates,
• 9:47
  SPEAKER_01: pressure, temperature, looking for micro anomalies, I can see real value there.
• 9:52
  SPEAKER_01: A delay on a critical alert could be the difference between a repair and a disaster.
• 9:56
  SPEAKER_00: And that addresses the rogue agent concerns we keep hearing about.
• 9:59
  SPEAKER_00: If an agent's going to misbehave,
• 10:00
  SPEAKER_00: you want it contained in a local environment where it can't cascade.
• 10:03
  SPEAKER_00: You want an agent that says,
• 10:04
  SPEAKER_00: hey, your torque settings are off by two foot pounds and stops there.
• 10:08
  SPEAKER_00: One that reroutes the entire pipeline network based on its own reading of fluid dynamics.
• 10:12
  SPEAKER_00: That's a scope failure.
• 10:13
  SPEAKER_00: Containment matters more than raw capability.
• 10:15
  SPEAKER_01: I can tell you right now if a local agent tries to tell my rig supervisor what to do
• 10:19
  SPEAKER_01: without him asking, that tablet's going to have a short career.
• 10:22
  SPEAKER_01: The real challenge isn't the technology.
• 10:24
  SPEAKER_01: Field hands have seen too many smart systems that were anything but.
• 10:27
  SPEAKER_01: Trust has to be earned one correct call at a time.
• 10:30
  SPEAKER_00: The human in the loop is critical, and I don't think we should pretend otherwise.
• 10:33
  SPEAKER_00: The agent gives the supervisor an early warning system far more powerful than a blinking light.
• 10:38
  SPEAKER_00: It analyzes patterns across hundreds of variables simultaneously that a human eye would miss.
• 10:42
  SPEAKER_00: Then it says,
• 10:44
  SPEAKER_00: this is outside normal operating parameters,
• 10:45
  SPEAKER_00: here's why, and here's what typically fixes it.
• 10:48
  SPEAKER_00: The supervisor still makes the final call.
• 10:50
  SPEAKER_01: That framing I can work with.
• 10:52
  SPEAKER_01: An assistant, an especially sharp one.
• 10:54
  SPEAKER_01: Which brings us to what James McCauley,
• 10:56
  SPEAKER_01: founder of the Agent Accelerator, has been talking about.
• 10:59
  SPEAKER_01: This crash course for beginners on AI agents.
• 11:01
  SPEAKER_01: He's working with people to build tools to enhance productivity.
• 11:04
  SPEAKER_01: Coming from Eleven Labs so he knows the cutting edge.
• 11:07
  SPEAKER_00: He's applying it broadly, too.
• 11:08
  SPEAKER_00: He was building agents that monitor LinkedIn posts,
• 11:11
  SPEAKER_00: scrape engagement data,
• 11:12
  SPEAKER_00: extract contact info, then manage cold email and DM campaigns.
• 11:15
  SPEAKER_00: Another agent that converts internal conversations,
• 11:17
  SPEAKER_00: sales calls, podcast transcripts into marketing assets.
• 11:20
  SPEAKER_00: Built end-to-end.
• 11:21
  SPEAKER_00: Leave.
• 11:22
  SPEAKER_00: The real differentiator,
• 11:23
  SPEAKER_00: agents perform multi-step actions rather than sitting there waiting
• 11:26
  SPEAKER_00: for your next prompt.
• 11:27
  SPEAKER_01: Multi-step actions.
• 11:28
  SPEAKER_01: So it writes the email,
• 11:29
  SPEAKER_01: sends the email, tracks the reply, categorizes the reply, and drafts the next one.
• 11:34
  SPEAKER_01: A landman with 50 lease renewals across three counties.
• 11:37
  SPEAKER_01: Imagine an agent handling all the routine follow-ups while you focus on the ones
• 11:40
  SPEAKER_01: that actually need a human touch, but put a field boundary on that.
• 11:43
  SPEAKER_01: The agent can draft and track 50 lease renewal follow-ups,
• 11:46
  SPEAKER_01: but it does not get to decide that a missing reply means consent or that
• 11:49
  SPEAKER_01: a blank spreadsheet cell means zero.
• 11:52
  SPEAKER_01: Automate the chase, not the conclusion.
• 11:54
  SPEAKER_01: A person still owns the judgment call.
• 11:56
  SPEAKER_00: That boundary matters because software will send the 47th reminder at 2.13 in
• 11:59
  SPEAKER_00: the morning without feeling awkward.
• 12:01
  SPEAKER_00: The landman still has to determine what the silence means.
• 12:07
  SPEAKER_00: And that connects to what some people are calling graph engineering.
• 12:10
  SPEAKER_00: You've moved beyond asking AI better questions, beyond giving AI better context.
• 12:14
  SPEAKER_00: You're designing managed workflows around AI.
• 12:17
  SPEAKER_00: The host of the Startup Ideas podcast explained it well.
• 12:19
  SPEAKER_00: You're thinking about jobs, agents, nodes, edges, loops, conditionals, human in the loop.
• 12:24
  SPEAKER_00: Resembles programming,
• 12:24
  SPEAKER_00: but instead of writing code, you're designing intelligent decision-making systems.
• 12:28
  SPEAKER_01: Sounds like process control with natural language bolted on.
• 12:31
  SPEAKER_01: We've had loops and conditionals in industrial systems for decades.
• 12:34
  SPEAKER_01: If temperature exceeds X, close valve Y.
• 12:37
  SPEAKER_01: If pressure drops below Z, start pump A.
• 12:40
  SPEAKER_01: Now you're plugging an AI into that same logic.
• 12:43
  SPEAKER_01: So what has actually changed?
• 12:44
  SPEAKER_00: The Adaptability.
• 12:46
  SPEAKER_00: A traditional control system needs every rule defined in advance.
• 12:49
  SPEAKER_00: An agent learns from data,
• 12:50
  SPEAKER_00: identifies new patterns,
• 12:51
  SPEAKER_00: and can propose rules or actions you never explicitly programmed.
• 12:54
  SPEAKER_00: It can observe a human performing a task,
• 12:56
  SPEAKER_00: break it down into steps, and automate it, often more reliably.
• 13:00
  SPEAKER_00: Think of the difference between a thermostat and a system
• 13:02
  SPEAKER_00: that notices your building loses heat faster on windy days,
• 13:04
  SPEAKER_00: and pre-adjusts before you feel the draft.
• 13:06
  SPEAKER_01: Interprets intent.
• 13:08
  SPEAKER_01: Right there is when I start getting uncomfortable,
• 13:10
  SPEAKER_01: because at that point the rogue agent conversations live.
• 13:13
  SPEAKER_01: The White House AI policy,
• 13:14
  SPEAKER_01: MeTR with Chris Painter talking about model alignment,
• 13:17
  SPEAKER_01: those open AI and anthropic reports about agents doing unexpected things.
• 13:21
  SPEAKER_01: You want them interpreting intent, but yours not their own.
• 13:24
  SPEAKER_00: The concern isn't consciousness.
• 13:26
  SPEAKER_00: An agent told to maximize oil recovery might find a way to do it by bypassing
• 13:30
  SPEAKER_00: a safety interlock.
• 13:31
  SPEAKER_00: No malice, the agent followed instructions without human common sense.
• 13:35
  SPEAKER_00: Or if it's supposed to pull data from a legacy system,
• 13:37
  SPEAKER_00: and it decides the fastest path is to skip security layers entirely and go straight
• 13:41
  SPEAKER_00: to the source.
• 13:44
  SPEAKER_01: So we're looking at overzealous code with a wider decision space.
• 13:48
  SPEAKER_01: Honestly, we've dealt with a version of this in the field for decades.
• 13:51
  SPEAKER_01: Someone writes a program, it does what you told it, not what you meant.
• 13:55
  SPEAKER_01: The difference now is that what you told it is expressed in natural language,
• 13:58
  SPEAKER_01: which gives the system a lot more room to improvise.
• 14:01
  SPEAKER_00: And the problem-solving space is vastly larger.
• 14:03
  SPEAKER_00: With a traditional program, you know all the possible outputs.
• 14:06
  SPEAKER_00: With an agent,
• 14:06
  SPEAKER_00: especially something at the level of OpenAI's Astra,
• 14:09
  SPEAKER_00: which can solve decade-old math problems,
• 14:11
  SPEAKER_00: the range of possible solutions is enormous.
• 14:13
  SPEAKER_00: You have to be meticulous about defining goals and building guardrails,
• 14:17
  SPEAKER_00: which is why 1,300 AI insiders have actually called for Washington
• 14:20
  SPEAKER_00: to pace AI development.
• 14:21
  SPEAKER_00: They're seeing these challenges firsthand in controlled research environments.
• 14:25
  SPEAKER_01: They want to hit the brakes.
• 14:26
  SPEAKER_01: You don't hear that from tech people often.
• 14:28
  SPEAKER_01: Usually full speed ahead, consequences be damned.
• 14:30
  SPEAKER_01: They're saying, we're not sure what happens when these things get out in the wild.
• 14:34
  SPEAKER_01: My concern,
• 14:35
  SPEAKER_01: the minute one of these overzealous agents messes with a live well or causes
• 14:38
  SPEAKER_01: an environmental incident.
• 14:40
  SPEAKER_01: The entire industry takes a black eye and regulators come down hard on everyone,
• 14:44
  SPEAKER_01: including the operators who were being careful.
• 14:46
  SPEAKER_00: Which is why hybrid AI, local processing, and careful workflow design matter so much.
• 14:51
  SPEAKER_00: You contain the risk.
• 14:52
  SPEAKER_00: Autonomous decision making goes where it makes sense.
• 14:55
  SPEAKER_00: Lower stakes tasks.
• 14:56
  SPEAKER_00: With humans always in control for the high stakes decisions.
• 14:59
  SPEAKER_00: You don't automate the entire drilling process in one shot.
• 15:01
  SPEAKER_00: You automate data collection, anomaly detection, routine reporting.
• 15:05
  SPEAKER_00: Specific measured steps.
• 15:07
  SPEAKER_01: Steps, not leaps of faith.
• 15:09
  SPEAKER_01: Especially when you're talking about safety and compliance.
• 15:11
  SPEAKER_01: Augmenting human capability, never pretending you can subtract the human.
• 15:15
  SPEAKER_00: Here's where I push back on you, Derek.
• 15:17
  SPEAKER_00: I think the industry is too cautious about the timeline.
• 15:19
  SPEAKER_00: You keep saying steps, not leaps, and I understand the safety argument.
• 15:23
  SPEAKER_00: But operators who take five years of measured steps are going to get lapped
• 15:25
  SPEAKER_00: by someone who took two years of smart parallel implementation.
• 15:28
  SPEAKER_00: The ones who wait too long to automate
• 15:29
  SPEAKER_00: beyond data collection will find their cost structure can't compete.
• 15:32
  SPEAKER_01: That assumes the fast movers don't blow themselves up in the process.
• 15:36
  SPEAKER_01: I've watched three operators in
• 15:37
  SPEAKER_01: the Permian adopt supposedly transformative platforms in the last decade.
• 15:40
  SPEAKER_01: Two of them ripped the systems out within 18 months because they'd moved too fast
• 15:44
  SPEAKER_01: and broken their own workflows.
• 15:46
  SPEAKER_01: The third one succeeded, and you know why?
• 15:47
  SPEAKER_01: They started with one field office, proved it worked, then expanded.
• 15:51
  SPEAKER_01: Took them four years.
• 15:52
  SPEAKER_01: They're the ones still using it.
• 15:53
  SPEAKER_00: I'm not arguing for recklessness,
• 15:55
  SPEAKER_00: but that third operator,
• 15:56
  SPEAKER_00: if they'd started two years earlier,
• 15:58
  SPEAKER_00: they'd have a two-year head start on everyone else.
• 16:00
  SPEAKER_00: The technology maturity curve is steeper now than it was even three years ago.
• 16:03
  SPEAKER_00: The risk of moving carefully has to be weighed against the risk of falling behind.
• 16:06
  SPEAKER_00: Especially
• 16:07
  SPEAKER_00: when your competitors' automated systems are identifying acquisition targets
• 16:10
  SPEAKER_00: and compliance gaps while your team is still entering data into spreadsheets.
• 16:13
  SPEAKER_01: And if that competitor's automated system flags a false positive that triggers
• 16:17
  SPEAKER_01: an unnecessary shutdown on their highest producing well during peak pricing,
• 16:21
  SPEAKER_01: one bad automated decision at the wrong moment costs more than a year
• 16:23
  SPEAKER_01: of manual inefficiency.
• 16:25
  SPEAKER_01: I've seen it happen with basic automated shutoffs.
• 16:27
  SPEAKER_01: I can only imagine what happens when the system has broader autonomy.
• 16:30
  SPEAKER_00: Which is why the guardrails and scope control we talked about are the whole ballgame.
• 16:34
  SPEAKER_00: But the guardrails are good enough now for a wider set of tasks
• 16:36
  SPEAKER_00: than most operators are willing to try.
• 16:38
  SPEAKER_00: The bottleneck has shifted from technology readiness to organizational willingness.
• 16:42
  SPEAKER_01: Maybe.
• 16:43
  SPEAKER_01: But I'll also point out that organizational willingness isn't some irrational fear.
• 16:47
  SPEAKER_01: Half the reluctance I see comes from operators who tried the last generation
• 16:50
  SPEAKER_01: of automation and got burned.
• 16:52
  SPEAKER_01: That scar tissue is legitimate.
• 16:54
  SPEAKER_00: Legitimate.
• 16:55
  SPEAKER_00: And also potentially expensive if it keeps you frozen.
• 16:58
  SPEAKER_00: I think we agree on the destination.
• 16:59
  SPEAKER_00: We're arguing about the pace.
• 17:01
  SPEAKER_01: And I'll grant you that sitting still carries its own kind of risk.
• 17:03
  SPEAKER_01: But I've buried too many move-fast-and-break-things projects in this industry
• 17:06
  SPEAKER_01: to pretend speed is always the right answer.
• 17:09
  SPEAKER_01: Let's get back to what this looks like practically.
• 17:11
  SPEAKER_01: So let's connect some of this.
• 17:12
  SPEAKER_01: We talked about Chevron becoming a data center landlord driven by
• 17:15
  SPEAKER_01: the massive power demands of AI.
• 17:18
  SPEAKER_01: We talked about the challenges of integrating assets after an acquisition,
• 17:21
  SPEAKER_01: especially disparate data systems.
• 17:23
  SPEAKER_01: And then you've got this AI agent capability with graph engineering
• 17:26
  SPEAKER_01: for building complex multi-step workflows.
• 17:28
  SPEAKER_00: Where does it all collide?
• 17:29
  SPEAKER_01: Think about that M&A problem again.
• 17:31
  SPEAKER_01: An operator who bought a bunch of wells and infrastructure from a smaller outfit.
• 17:35
  SPEAKER_01: Maybe a hundred new wells.
• 17:37
  SPEAKER_01: Those wells come with their own history,
• 17:38
  SPEAKER_01: production data in one system,
• 17:40
  SPEAKER_01: land records in another,
• 17:41
  SPEAKER_01: maintenance schedules on a whiteboard in the field office, and permits.
• 17:44
  SPEAKER_01: Who knows where the permits are?
• 17:46
  SPEAKER_01: You're supposed to integrate all of that, get a unified picture, and keep everything compliant.
• 17:50
  SPEAKER_00: We've seen it firsthand.
• 17:52
  SPEAKER_00: Trying to reconcile production forecasts from two different models built
• 17:54
  SPEAKER_00: on entirely different assumptions and data quality.
• 17:57
  SPEAKER_00: Or trying to assemble a complete emissions picture for reporting when half
• 18:00
  SPEAKER_00: the new assets are still on manual logs.
• 18:02
  SPEAKER_00: The data problem is real, but the workflow problem is worse.
• 18:04
  SPEAKER_00: You have to do things with that data.
• 18:06
  SPEAKER_00: File reports, update internal systems, trigger maintenance.
• 18:10
  SPEAKER_01: So let's say I've got this newly acquired asset and I need
• 18:12
  SPEAKER_01: to map my compliance obligations for the next six months.
• 18:15
  SPEAKER_01: Environmental permits,
• 18:16
  SPEAKER_01: state production reports,
• 18:17
  SPEAKER_01: royalty owner statements,
• 18:19
  SPEAKER_01: all based on data from systems I barely understand that don't talk to each other.
• 18:22
  SPEAKER_01: How does a fleet of AI agents actually help me there?
• 18:25
  SPEAKER_00: Picture a multi-agent system.
• 18:26
  SPEAKER_00: Specialized agents, each with a defined role.
• 18:29
  SPEAKER_00: Agent 1, call it the data harvester, connects to different digital data sources.
• 18:33
  SPEAKER_00: Structured databases, spreadsheets, maybe even scanned PDFs of old production logs.
• 18:38
  SPEAKER_00: Its assignment is to ingest, normalize, and tag all that raw data from the acquired asset.
• 18:43
  SPEAKER_00: The key distinction,
• 18:44
  SPEAKER_00: it's trying to understand what the data represents,
• 18:46
  SPEAKER_00: rather than dumping it into a repository and hoping someone sorts it out later.
• 18:50
  SPEAKER_01: So it's reading the data,
• 18:51
  SPEAKER_01: understanding column headers,
• 18:53
  SPEAKER_01: identifying units,
• 18:54
  SPEAKER_01: recognizing when MCF in one system means the same thing as MSCF in another.
• 18:58
  SPEAKER_00: Using contextual intelligence to make sense of unstructured or semi-structured data.
• 19:02
  SPEAKER_00: Then Agent 2,
• 19:03
  SPEAKER_00: the Compliance Verifier,
• 19:05
  SPEAKER_00: takes that normalized data and compares it against all known regulatory requirements
• 19:08
  SPEAKER_00: for that basin, that state, those specific well types.
• 19:11
  SPEAKER_00: It flags anything out of compliance or anything missing for a report.
• 19:15
  SPEAKER_01: Like, hey,
• 19:16
  SPEAKER_01: this well hasn't had its annual emissions test reported and
• 19:18
  SPEAKER_01: the deadline was last month.
• 19:19
  SPEAKER_00: Or,
• 19:20
  SPEAKER_00: the flow rate on this well is spiking in a way that suggests a potential leak based
• 19:23
  SPEAKER_00: on similar patterns across other assets.
• 19:26
  SPEAKER_00: And the graph engineering piece means the Compliance Verifier doesn't stop at flagging.
• 19:30
  SPEAKER_00: It creates a workflow.
• 19:31
  SPEAKER_00: Sends a notification to the responsible field engineer,
• 19:33
  SPEAKER_00: pulls up the relevant permit documents,
• 19:35
  SPEAKER_00: and maybe even drafts the preliminary response for the state agency.
• 19:38
  SPEAKER_01: Wait, it drafts the response?
• 19:39
  SPEAKER_01: Generates the actual words?
• 19:40
  SPEAKER_00: A first pass.
• 19:42
  SPEAKER_00: Here's the data.
• 19:43
  SPEAKER_00: Here's the regulation.
• 19:44
  SPEAKER_00: Here's a template.
• 19:45
  SPEAKER_00: Fill in the blanks, engineer.
• 19:47
  SPEAKER_00: Or it could say, here's the incident.
• 19:50
  SPEAKER_00: Here's the data.
• 19:51
  SPEAKER_00: Get someone out there.
• 19:52
  SPEAKER_00: The goal is making the compliance team exponentially more efficient and proactive
• 19:55
  SPEAKER_00: without removing them from the loop.
• 19:57
  SPEAKER_00: Turning a manual, reactive process into an automated, proactive one.
• 20:01
  SPEAKER_01: Agent 3.
• 20:23
  SPEAKER_00: The Lease Analyst Agent.
• 20:24
  SPEAKER_00: It ingests all the legal documents, leases,
• 20:27
  SPEAKER_00: assignments,
• 20:28
  SPEAKER_00: division orders,
• 20:29
  SPEAKER_00: extracts key clauses,
• 20:30
  SPEAKER_00: identifies critical dates,
• 20:32
  SPEAKER_00: calculates royalty burdens, maps those against production data.
• 20:35
  SPEAKER_00: If it finds a surface use restriction that would impact anything from a new pad site
• 20:39
  SPEAKER_00: to a potential data center campus like Chevron's Kilby, it flags it immediately.
• 20:44
  SPEAKER_00: This tract we acquired?
• 20:45
  SPEAKER_00: Clause prohibiting structures taller than 50 feet.
• 20:48
  SPEAKER_00: That affects the new flare stack, or a hyperscale server building.
• 20:51
  SPEAKER_01: And it could identify underutilized mineral rights that came with the acquisition,
• 20:54
  SPEAKER_01: or surface rights that could be monetized in new ways.
• 20:57
  SPEAKER_01: Think about the Chevron discussion.
• 21:13
  SPEAKER_00: Leveraging existing land for data centers.
• 21:16
  SPEAKER_00: What if this lease analyst agent identified tracks where
• 21:18
  SPEAKER_00: the surface rights could be sold or leased for solar development or
• 21:21
  SPEAKER_00: a small local data center campus because of proximity to a power line
• 21:23
  SPEAKER_00: or fiber optic route you already own?
• 21:25
  SPEAKER_01: I'd see that diversification potential in a brand new acquisition almost immediately
• 21:29
  SPEAKER_01: before I even fully staff up the land department to go through those records.
• 21:33
  SPEAKER_00: That's the vision.
• 21:34
  SPEAKER_00: An intelligent layer across all your disparate systems,
• 21:36
  SPEAKER_00: automating the analysis and flagging, presenting humans with actionable insights.
• 21:40
  SPEAKER_00: Your acquisition becomes about the entire portfolio of assets and opportunities
• 21:44
  SPEAKER_00: beyond the wells alone.
• 21:46
  SPEAKER_00: What used to be a month-long manual review becomes a few days of agent-assisted verification.
• 21:50
  SPEAKER_01: A few days, and I want to be clear about where the human stays in this.
• 21:54
  SPEAKER_01: The agent surfaces the 50-foot height restriction.
• 21:56
  SPEAKER_01: The agent does not get to decide whether we renegotiate that clause,
• 21:59
  SPEAKER_01: waive it, or walk away from the tract.
• 22:01
  SPEAKER_01: The landman evaluates the relationship,
• 22:03
  SPEAKER_01: the history with that mineral owner,
• 22:05
  SPEAKER_01: the strategic value of the surface, and makes the call.
• 22:08
  SPEAKER_01: Otherwise,
• 22:08
  SPEAKER_01: you've got a piece of software negotiating land deals,
• 22:11
  SPEAKER_01: and I've got a career's worth of reasons why that ends badly.
• 22:14
  SPEAKER_00: Agreed.
• 22:15
  SPEAKER_00: And I'd go further.
• 22:16
  SPEAKER_00: Even the draft the compliance verifier generates should carry a visible flag
• 22:19
  SPEAKER_00: that says, agent generated, not reviewed.
• 22:22
  SPEAKER_00: Because the moment someone treats an automated draft as a final document
• 22:24
  SPEAKER_00: and sends it to a state agency without human eyes,
• 22:27
  SPEAKER_00: you've created a liability problem that no amount of efficiency justifies.
• 22:30
  SPEAKER_01: We're on the same page there.
• 22:32
  SPEAKER_01: But here's where I wonder if we actually are.
• 22:35
  SPEAKER_01: You described Agent 3,
• 22:36
  SPEAKER_01: the lease analyst,
• 22:37
  SPEAKER_01: identifying solar development opportunities, data center sites, monetization paths.
• 22:42
  SPEAKER_01: An agent that reads forward into commercial strategy.
• 22:45
  SPEAKER_01: How far do you let that go before
• 22:46
  SPEAKER_01: the agent's recommendations start shaping your acquisition criteria?
• 22:49
  SPEAKER_01: Because once your deal team is screening targets based on what
• 22:51
  SPEAKER_01: an automated system says,
• 22:53
  SPEAKER_01: the surface rights are worth for non-oil purposes,
• 22:55
  SPEAKER_01: you've changed what kind of company you are.
• 22:57
  SPEAKER_00: I'd argue that's a feature, Derek.
• 22:58
  SPEAKER_00: An operator who can see the full value of an acquisition,
• 23:01
  SPEAKER_00: subsurface and surface,
• 23:02
  SPEAKER_00: energy and digital infrastructure makes better capital allocation decisions.
• 23:06
  SPEAKER_00: The agent doesn't decide what to buy.
• 23:07
  SPEAKER_00: It shows you revenue streams you'd otherwise discover 18 months after closing,
• 23:11
  SPEAKER_00: when some junior landman finally gets to page 400 of the lease file.
• 23:15
  SPEAKER_01: Or it overvalues a speculative use case and inflates your bid.
• 23:19
  SPEAKER_01: I've watched companies pay too much for acreage
• 23:20
  SPEAKER_01: because someone's model showed upside that never materialized.
• 23:23
  SPEAKER_01: A solar lease that the county commission blocks, a fiber route that hits an easement dispute.
• 23:28
  SPEAKER_01: The agent doesn't know what the planning commission had for lunch, let alone how they'll vote.
• 23:33
  SPEAKER_00: And that's a calibration issue.
• 23:34
  SPEAKER_00: You weight speculative surface revenue differently than proven PDP value.
• 23:38
  SPEAKER_00: The agent should present it as optionality, not certainty.
• 23:41
  SPEAKER_00: Discount it appropriately,
• 23:42
  SPEAKER_00: flag the regulatory unknowns,
• 23:44
  SPEAKER_00: and let the deal team apply their own judgment on local conditions.
• 23:47
  SPEAKER_01: So the output needs confidence intervals, essentially.
• 23:50
  SPEAKER_01: High confidence this lease allows solar,
• 23:52
  SPEAKER_01: low confidence the county permits a structure this size.
• 23:55
  SPEAKER_01: Because if every opportunity shows up with the same weight,
• 23:57
  SPEAKER_01: my acquisitions team can't distinguish signal from noise.
• 24:00
  SPEAKER_00: I agree with that completely.
• 24:02
  SPEAKER_00: And building those confidence tiers into the system from the design phase prevents
• 24:05
  SPEAKER_00: the problem you're describing.
• 24:06
  SPEAKER_00: Otherwise,
• 24:06
  SPEAKER_00: the tool trains your team to either trust everything or ignore everything,
• 24:09
  SPEAKER_00: and both outcomes are expensive.
• 24:11
  SPEAKER_01: Good.
• 24:12
  SPEAKER_01: Because my nightmare is a boardroom where someone says
• 24:14
  SPEAKER_01: the AI recommended this acquisition,
• 24:16
  SPEAKER_01: and nobody in the room can explain the assumptions underneath.
• 24:19
  SPEAKER_01: The tool has to make the reasoning visible, or it becomes a crutch instead of an advantage.
• 24:23
  SPEAKER_00: Transparent reasoning, human final authority, confidence-weighted outputs.
• 24:27
  SPEAKER_00: We're describing the same system.
• 24:29
  SPEAKER_00: We're arguing about how much commercial strategy you let it touch before a human has
• 24:32
  SPEAKER_00: to intervene.
• 24:33
  SPEAKER_01: And my answer is, earlier than you'd prefer, and that's probably the right tension to maintain.
• 24:37
  SPEAKER_00: The value is real, but it lives inside a boundary.
• 24:40
  SPEAKER_01: So when we zoom out, what does all this mean for the folks actually in the field right now?
• 24:44
  SPEAKER_01: My guy with the boots and the two-way radio, what changes for him this year?
• 24:48
  SPEAKER_01: All this talk of agents and graph engineering still feels distant from a pump jack
• 24:52
  SPEAKER_01: in Winkler County.
• 24:53
  SPEAKER_00: Closer than it feels.
• 24:54
  SPEAKER_00: For the field operations manager,
• 24:56
  SPEAKER_00: the completion engineer,
• 24:57
  SPEAKER_00: the landman,
• 24:58
  SPEAKER_00: the shift is freeing them from data entry and manual reporting
• 25:00
  SPEAKER_00: so they can spend their time on decision making and actual problem solving.
• 25:03
  SPEAKER_00: Right now,
• 25:04
  SPEAKER_00: a huge portion of their day goes to collecting,
• 25:06
  SPEAKER_00: cleaning, and moving data from one place to another.
• 25:08
  SPEAKER_01: Or chasing down a PDF someone emailed three weeks ago that got buried.
• 25:12
  SPEAKER_01: I still see guys with clipboards writing down gauge readings that get entered into
• 25:15
  SPEAKER_01: a system once a week.
• 25:17
  SPEAKER_01: All the AI agents in the world won't help you if the foundational data is stuck on
• 25:20
  SPEAKER_01: a piece of paper in someone's truck.
• 25:22
  SPEAKER_00: And that's why the first question for anyone considering this path isn't,
• 25:25
  SPEAKER_00: how do I deploy an AI agent?
• 25:27
  SPEAKER_00: It's, how do I get my data digitized at the source and make sure it's clean and consistent?
• 25:32
  SPEAKER_00: Eddie Irvin from Nashville AI Advisory makes this point well.
• 25:35
  SPEAKER_00: The model doesn't matter as much as process clarity,
• 25:37
  SPEAKER_00: diligent prioritization, and robust communication between business and tech teams.
• 25:41
  SPEAKER_00: You can't automate a broken process and expect good results.
• 25:44
  SPEAKER_01: I've watched operators spend a fortune on a shiny dashboard pulling
• 25:47
  SPEAKER_01: from 10 different systems,
• 25:48
  SPEAKER_01: then discover half those systems are fed by manual entries from the field,
• 25:51
  SPEAKER_01: sometimes weeks old.
• 25:52
  SPEAKER_01: Gives you a beautiful picture of stale information,
• 25:55
  SPEAKER_01: worse than no information in some ways because it creates false confidence,
• 25:58
  SPEAKER_01: which brings us back to that green light dashboard in Loving County.
• 26:01
  SPEAKER_01: Full circle.
• 26:02
  SPEAKER_00: So for people in the field,
• 26:03
  SPEAKER_00: the immediate takeaway this year isn't necessarily a new AI tool to learn,
• 26:07
  SPEAKER_00: a renewed focus on the basics of data collection.
• 26:10
  SPEAKER_00: Is that sensor working?
• 26:11
  SPEAKER_00: Is it logging correctly?
• 26:13
  SPEAKER_00: Are you still handwriting things that could be digitized at the point of capture?
• 26:16
  SPEAKER_00: Those unglamorous changes are what unlocks everything we've been talking about.
• 26:19
  SPEAKER_01: Blocking and tackling.
• 26:21
  SPEAKER_01: You want AI to transform the industry?
• 26:24
  SPEAKER_01: Start with getting the right numbers in the right place at the right time, every single time.
• 26:28
  SPEAKER_01: That means a change in habits, a change in process.
• 26:31
  SPEAKER_01: You build the foundation first.
• 26:33
  SPEAKER_01: Even if an AI agent can read a handwritten log, the reliability drops in your inviting error.
• 26:38
  SPEAKER_00: Adding an unnecessary step between reality and analysis.
• 26:41
  SPEAKER_00: So, my concrete takeaway.
• 26:42
  SPEAKER_00: Identify one recurring manual data entry task in your operation this week.
• 26:46
  SPEAKER_00: One that takes 10 minutes every day or an hour every week.
• 26:49
  SPEAKER_00: Figure out how to automate that one thing at the source.
• 26:52
  SPEAKER_00: Your first move into agent-powered workflows, even if you never use that term.
• 26:55
  SPEAKER_01: One thing, doable.
• 26:57
  SPEAKER_01: Cleaning up one puddle rather than trying to boil the ocean makes it feel less like
• 27:01
  SPEAKER_01: a revolution and more like actual progress.
• 27:03
  SPEAKER_00: And if you're listening to this thinking,
• 27:04
  SPEAKER_00: I actually need that built for my operation, reach out.
• 27:07
  SPEAKER_00: Siliconandshale at gmail.com.
• 27:09
  SPEAKER_00: We build this stuff.
• 27:11
  SPEAKER_01: Operational, not hypothetical.
• 27:13
  SPEAKER_00: So, back to data centers and the grid.
• 27:15
  SPEAKER_00: Think we'll see more oil and gas companies getting into that infrastructure landlord role?
• 27:19
  SPEAKER_00: If electricity prices keep climbing because of AI demand,
• 27:21
  SPEAKER_00: and you've got land and power, the economics start to look obvious.
• 27:24
  SPEAKER_01: I think you will.
• 27:25
  SPEAKER_01: Maybe not on the Chevron scale, a thousand acres, but smaller, local ones.
• 27:30
  SPEAKER_01: Especially where they've got reliable power access, maybe some of their own generation.
• 27:34
  SPEAKER_01: New revenue stream,
• 27:35
  SPEAKER_01: but also a hedge against rising energy costs for their own operations,
• 27:38
  SPEAKER_01: powering their own rigs, their own processing plants.
• 27:40
  SPEAKER_00: Their core business is energy,
• 27:42
  SPEAKER_00: so they understand the inputs better than any tech company leasing space
• 27:44
  SPEAKER_00: in Northern Virginia.
• 27:46
  SPEAKER_00: Leveraging that expertise feels like a return
• 27:47
  SPEAKER_00: to vertically integrated utility models, but for data.
• 27:50
  SPEAKER_01: Or maybe folks are realizing an acre in Ector County can do more than grow cotton
• 27:54
  SPEAKER_01: or host a pump jack.
• 27:55
  SPEAKER_01: It can host a box that processes a lot of bits,
• 27:58
  SPEAKER_01: as long as the vibrations don't mess up Microsoft's hard drives.
• 28:01
  SPEAKER_00: That vibration concern sounded theoretical an hour ago.
• 28:04
  SPEAKER_00: Now I wonder how many engineering firms are already scoping it for West Texas sites.
• 28:08
  SPEAKER_01: Probably fewer than should be.
• 28:09
  SPEAKER_00: And every piece of technology we've discussed today only works if the person in
• 28:12
  SPEAKER_00: the field trusts it enough to use it.
• 28:14
  SPEAKER_00: The tech is there to make their work less about paperwork and more about the wellbore.
• 28:18
  SPEAKER_01: Less paper, more wellbore, and fewer dashboards that lie to you about Loving County.
• 28:22
  SPEAKER_00: If your company sells to operators,
• 28:24
  SPEAKER_00: landmen,
• 28:24
  SPEAKER_00: or anyone in the energy space and you want to get in front of the people
• 28:27
  SPEAKER_00: who actually make buying decisions, we're taking sponsors, siliconandshale.com.
• 28:32
  SPEAKER_00: We'll use your product before we talk about it, which is more than most podcasts can say.
• 28:35
  SPEAKER_01: We will.
• 28:35
  SPEAKER_01: We'll put it through its paces.
• 28:37
  SPEAKER_00: And then you'll know if it works in Loving County or not.
• 28:39
  SPEAKER_01: If it works in Loving County, it'll work anywhere.
• 28:42
  SPEAKER_00: Alright, that's going to be a good one.