If you are going new, CCGT costs cluster near $2,000/kW for the post-2027 cohort against roughly $900/kW before 2023. In ERCOT a 1 GW plant at a 6.5 heat rate and 75% utilization cannot be financed above $1,500/kW on conventional terms, so it takes a PPA near $70/MWh to break ground. PJM needs roughly $500/MW-day of capacity, or a 15-year bilateral at the cap. Numbers will vary, but the band is narrow and knowable.

The market to buy is hot. Operating gas M&A doubled from about $0.5 million/MW before 2025 to roughly $1.0 million/MW last year, and against a $2.0 to $2.3 million replacement cost that is fifty cents on the dollar. Expanding at an already-interconnected site lands at $1.3 to $1.6 million/MW. The cheapest megawatt in the market is one that already exists.

Long Ridge showed what that megawatt is worth. MARA’s purchase included the well pads as well as the operating asset and implies a $52/MWh contract. Priced off an $80/MWh behind-the-meter PPA instead, a $15/MWh uplift to LMP and in line with comparable deals, we value the Hannibal plant at $3.76 million/MW against $2.53 million merchant. Capturing the contract means a ~50% uplift for the same turbines, at the same node, burning the same gas.

Not every plant has the same options. When Talen paid $1.33 million/MW blended for three gas plants in January we marked Waterford at $2.22 million/MW and Darby at $0.34 million. Waterford earned that on merit, a 7.0 heat rate running 88% of the year, and the same efficiency is what makes it worth contracting. Darby runs 8% of the year on a 12.4 heat rate, so it is cheap today with no option to contract. Across the fleet, the plant with the best heat rate carries both the highest value per megawatt and the largest PPA upside, roughly 16% of NAV at $60/MWh.

Old or new, the upside sits in PPAs struck above merchant. I would pay for the plant and chase the option, which means the work is finding the plants that carry one. Pick the right heat rate at the right node. Get past a market monitor that already killed Talen’s Susquehanna colocation and wants Long Ridge’s capacity kept in PJM. Sign the contract. Or watch somebody else do it.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Utility physics follows the same pattern. Load growth should predict earnings growth. Utilities are designed to serve load, after all. Growing demand means more generation, transmission and distribution, which should mean more rate base. Per-share growth should follow if the utility grows rate base, earns its allowed return and avoids diluting the equity along the way.

Easy math. And it works, up to a point.

In our Utility Compass, the relationship between our load forecasts and market-implied growth is positive, but the slope is shallow and the dispersion wide. Dividend yield is the big lever, more than I expected, explaining 62% of the variation in forward price-to-earnings multiples across our coverage group. The remaining 38% contains differences in cost of equity and expected growth, and load should explain part of it.

Even if load growth has some explanatory power, the trajectory is noisy. AEP’s deck points to a 25.8% load CAGR through 2030, roughly 69 GW. We model 2.0% and 2.6 GW. The market implies 3.2%.

More broadly, our 2030 peak demand estimates run below disclosed guidance for every utility in the group. WEC is the only name where market-implied growth of 2.8% sits below our own estimate of 3.4%. ETR carries the widest growth premium in the group, roughly 4.1 points against our 0.4% load forecast.

Then there is the load that never arrives. We expect behind-the-meter generation to serve 41% of incremental data center load through 2030, roughly 30 GW of new gas capacity built outside the rate base. Of PJM’s 11.8 GW high-confidence large-load pipeline, 58% is expected to go behind the meter. MISO’s 11.6 GW stays 78% grid-connected, but its proposal would have those customers pay the attributed upgrade costs upfront. One market loses the load. The other keeps it and gives away the return.

Load growth is murky for utilities and investors alike. Converting it into earnings runs into a new limit: who funds and owns the build.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

ERCOT was playing its own summer games last week. The grid served a preliminary record of 91.1 GW on July 22, more than 5.5 GW above the previous official record. Breaching 90 GW is a big deal. How big? Up there with the Berlin Wall and the Apollo moon landing, at least according to our lead ERCOT analyst.

Load was also pulling a Mondo Duplantis, clearing the record without revealing how high it could really go. Summer in ERCOT is 4CP season. Under Four Coincident Peak, transmission costs are allocated based on demand during the highest 15-minute interval in each of the four summer months. Our team issued 4CP alerts ahead of the record, and some large users were likely responding by curtailing operations, discharging batteries or switching to on-site generation.

Did you notice? Likely not. ERCOT made history without the price spikes or public panic. Chalk that up as a win.

But what is good for customers today may not be good for them in the future. Record load produced a weak price signal for the firm capacity ERCOT may need under less favourable conditions. A few summer days clearing around $70/MWh will not finance much merchant capacity. We think new build needs a long-term contract at around that to pencil, not a handful of isolated summer clears.

While ERCOT was writing records, a transmission fault in Northern Virginia caused data centers to transfer more than 3 GW of demand to backup generation within seconds. For PJM, it was a generator outage in reverse: demand vanished, then returned once conditions normalized.

Large Load drops like this are a growing concern for grid operators. We expect 62 GW of new U.S. data center capacity through 2030. The risk of similar multi-gigawatt load swings is growing, and not just in Virginia.

ERCOT and PJM are telling the same story at different speeds. In Texas, cost-allocation incentives shape gigawatts of demand over hours. In Virginia, automated protection systems remove gigawatts of grid demand within seconds. Those lessons extrapolate across markets.

Grid operators have traditionally treated load as the score and generation as the players. Load is now on the field. It responds to incentives, anticipates the rules and occasionally makes its own moves. Like the games at our house, understanding what happens next requires knowing who is playing and which version of the rules applies.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

The change is evident in the ERCOT congestion data and our August Congestion Revenue Right (CRR) outlook. If power is not your day job, grid congestion is traffic for electrons, priced as the gap between points on the grid, and CRRs are how you participate in the market.

Our work ran six load and weather scenarios across the state. The wire into the Dallas Fort Worth hyperscale corridor and the line feeding Lancium’s Abilene datacenter bound in all six. Bottlenecks usually need weather, which is the point of running the scenarios. These bind on load alone, because real energized megawatts, Riot’s 400 MW, Lancium’s 500 MW, sit behind a wire that is already full. Data center load is making congestion structural.

Meeting that load also takes new generation. Our supply-side screen scored 95 in-development ERCOT gas plants, about 60 GW, on deliverability and offtake. Only 16 clear the bar into Tiers 1 to 3, roughly 6 GW. The other 54 GW sits in Tier 4, mostly at Facility Study with no contracted buyer, because a gas plant does not get financed until a creditworthy load signs the PPA. A $2,500/kW CCGT needs roughly $70/MWh to pencil, and merchant revenue in an energy-only market does not get there. The buyer signs first. Everyone else waits in the queue.

So who benefits? LCRA owns the top of the table, with both Schneider RICE phases the only names at 77.5% and its Schneider CCGT and Lost Pines Recip close behind. Deliverability does the sorting, not size. Among listed IPPs, NRG’s Cedar Bayou 5 screens highest at 58.4% on a signed interconnection agreement, while the other 5 GW of NRG gas sits in Tier 4 with no offtake. VST’s Texas Gulf Sulphur scores 54.1% and is all of 16 MW. Headline gigawatts are the cheapest thing in this market.

PJM is the same match on a bigger pitch. PJM East spends $3.5 billion a year on transmission, the most in the nation, and congestion still moves the wrong way. The clocks explain why. Load arrives in three years, generation takes four to six, transmission takes ten.

The grid is Messi at 39, living on the magic that’s left, still capable for producing frantic comebacks. But the load wave is only warming up. Beautiful football is patient. It waits for the magic to run out.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Meta’s power arrangement is by now commonplace. The load gets matched with renewables on paper, annually, while the firming comes from the Greenlight Electricity Centre, a 932MW gas plant Pembina (with partners Morgan Stanley Infrastructure Partners and Kineticor Asset Management) is building against a long-term Meta tolling agreement, online by 2030. Clean on the surface, gas underneath.

Looking to the US data center build, the fuel cell is an interesting variation on the theme. Our work has Bloom’s solid oxide stacks penciling around $91/MWh behind the meter, within 1% of a combined-cycle plant, and Oracle stood one up in 55 days. The technology is attractive for the use case, and the stock has run roughly 1,000% in a year on that promise.

The real bet is the order book. Bloom looks reasonable as long as it executes on its backlog, and that is where the whole thesis sits. Some of it is signed and binding. Some is announced, optioned or aspirational. The gap between the headline and what Bloom ultimately installs is the difference between a scaling infrastructure platform and a product that is more hat than cattle. There is nothing new about this. Whether it is the Big E exploration or technology, the potential shows up in the order book and investors are asked to bet on its quality.

The pattern underneath holds either way. Getting to the groundbreaking ceremony is the hard part, and plenty of announced projects get bucked off in the queue and never energize. What keeps one on to eight seconds now is a signed buyer. Post-2024 combined-cycle plants cost around $2.0 million/MW to build while existing gas trades near $1.0 million/MW, so an acquirer pays 50 cents on the replacement-cost dollar. Nobody spends into that math on hope.

Williams proved it in the US yesterday. It pulled $5.34 billion from a Blackstone-led group, with Apollo and KKR, for 49% of its first five behind-the-meter gas projects, keeping control and pointing the capital at a backlog north of 6 GW. Pembina builds gas in Alberta against a Meta toll. Williams builds gas in the US against contracted demand and shared capital. Fuel cell or turbine, the rule holds. The buyer signs first.

Coming up: Join Pat Lynch of CBRE and me tomorrow as we discuss “Bringing the Site to the Power” and the realities of data center development.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

For larger projects, clearing the under-construction bar ahead of the deadline won the prize of “safe harbor”: roughly four more years to reach commercial operation and still collect the credit. Meet that bar, and the investment and production tax credits keep paying out well beyond 2027.

For small and residential systems that build in weeks, the end-2027 deadline is well within reach, so locking in under-construction status this summer mattered little. But for utility-scale projects, first queue date to an interconnection agreement alone runs three to four years, before a shovel moves. Without the time that safe harbor buys, the tax credits are effectively worthless.

That is why the fight over what “begin construction” means got contentious. For years a developer could lock eligibility on paper, spending 5% of project cost, no dirt required. IRS Notice 2025-42 tried to remove that route, leaving physical work as the only path. But on June 6 a federal court vacated the notice and restored the 5% safe harbor, and conceded its own appeal would not resolve before the deadline it was fighting over. At the project level, OBBBA’s impact is messy and uneven. Right after enactment we pegged 30% of queued solar and 57% of onshore wind as economic without credits at all. At EVOLVE in May we mapped some solar clearing below $30/MWh in MISO and PJM and the best Midwest wind under $20/MWh. Pine Gate’s bankruptcy was a reminder of the risk weaker portfolios still carry.

Zoom out and safe harbor looks more like a bridge than a loophole. Solar and wind are what most of the country can build now, and the under-construction wedge, 133.8 GW and rising, has roughly four years to come online, supported by tax credits and climbing PPAs. That capacity carries the system until gas turbine deliveries ramp and firm resources like SMRs and geothermal arrive at scale. The date everyone raced to beat turned out to be a starting gun, not a finish line. Four years to turn safe-harbored paper into steel, and to build a bridge to the next era.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Power markets have their own brackets. Big announcements building queues everywhere. The hard part now is filtering the contenders from the pretenders. That is where our team is spending a lot of their time.

Start with data center capacity, the clearest read. About 96% of the additions we expect in 2028 trace to projects under construction or contracted. By 2030 that share falls to 26%. The bets you can place today fade fast a few years out. Potential is not the issue. We have mapped 272 GW of it in our Sites Unseen work. Naming the winners a couple years out is the murky part, and even marquee projects slip; OpenAI and Oracle scrapped a 600 MW expansion at their Abilene Stargate campus. Sorting that out is the game we are playing.

Emerging technologies and long project queues raise the same problem in sharper form. The market prices the prospect. We try to price the proof. That gap can be the whole valuation. O&G veterans have seen it before. OGX went public in 2008 as Brazil's hottest offering, claiming exploratory success north of 90% and reserves to rival the majors, then the wells came in dry and the stock lost 99% of its value. The resource was prospective. It was never proven. Price a story as if it is solved and you underwrite the promise. Price what has been demonstrated and you underwrite the well.

ERCOT decided to charge for the filter. Batch Zero asks 197 large-load projects to put up cash, $50,000 per MW, $50 million for a gigawatt campus with $40 million at risk if you walk. Announcements are free, but deposits are not. By our count 55 advance to the next round, 21.7 GW, and the rest wait for Batch 1+. Making the bracket does not guarantee the trophy, but you have to get in to have a chance.

The rules keep moving. FERC’s June orders to all six grid operators will reshape how large loads connect. Better stay on top of that.

Canada is through the group and a knockout round. It only gets harder. Power runs the same gauntlet, except the capital commits years before the whistle. Calling which side survives is the game.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Start with location. Chevron (CVX) signed a 20-year deal to power a Microsoft data center in West Texas, building roughly 2.5 GW of dedicated gas generation on 2,000 acres in Reeves County, behind the meter and off the public ERCOT grid. On June 18 FERC pointed the rules the same way, issuing show-cause orders to all six grid operators and tilting toward colocated projects. The good address now sits next door to generation, and EIR puts AEP, Oncor and NEE in the best zip codes to capture it.

Then fuel. The Chevron plant burns gas from the company’s own Permian fields, a short hop from wellhead to turbine to server. Fervo (FRVO) is the other side of that coin, introducing geothermal as alternative fuel. We see it priced to perfection at $35.09 against an EIR fair value of $18.10 even after granting the technology works. Its rock is plenty hot. Converting that heat to steady megawatts is the hard part. The plan calls for about 25 power units a year, four times what the entire global ORC industry ships in a typical year, near 300 MW, or five to six units.

Then delivery, which is where the tenants separate. Chevron has already lined up the hardware, a majority of the generation from GE Vernova (GEV) turbines with the balance from Caterpillar’s (CAT) Solar Turbines, the exact equipment everyone else is waiting in line for. Crusoe is the counterexample. Black Hills confirmed it walked from Project Jade, a 1.8 GW campus in Cheyenne, leaving Black Hills to build directly with a customer that had already sunk more than $200 million into milestones. EIR’s ERCOT Batch Zero work scores all 197 large-load projects in the queue on exactly this, real tenant or tourist.

The grid has turned into a property market. Location, fuel, delivery. You get paid for the address, the megawatts and the ability to show up. The resource and the announcement are table stakes now. The deal closes for the one who shows up with the turbines.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Taste will matter more when the machines can write. Judgment will be the moat.

I believe that is right. I also think most people mean something too soft by it. Taste is not a feel for clean prose or a nose for the elegant chart. Taste is the willingness to take a position and stand behind it long enough for someone to attack it. That is the thing that does not get cheaper. That is the thing organizations will pay more for, not less.

I have been building AI into my own workflow for months. Dictation changed the first mile, from thought to draft. AI-assisted editing changed the second. Together they shifted my production function. I can produce far more written work from the same amount of attention than I could a year ago.

So can everyone else.

If every analyst, strategist, PM, consultant, and executive can generate more drafts and more arguments, the world fills up with output. More memos, more dashboards, more plausible-looking arguments, more things that look finished because they are polished. It does not fill up with better decisions.

The constraint moves. Output bandwidth expanded. Input bandwidth did not. Reading still takes time. Absorbing an argument still takes attention. Deciding whether something is worth acting on still requires a person. AI widens the pipe coming out far faster than it widens the pipe going in, and that asymmetry is where the value relocates. What deserves attention. What should be done with it. Who is willing to sign their name to the recommendation.

This is where taste actually lives, and it is harder than the comfortable version.

A summary can be complete and leave everyone exactly where they started. A position points somewhere. It says, given what we know, do this. Or wait. Or this risk matters more than that one. Or this is the tradeoff we should accept.

That move is what makes analysis useful inside an organization. A position is specific enough to be challenged, grounded enough to be defended, and concrete enough that someone can disagree with it productively. A good position gives other people something to push back on. The pushback is the point. It surfaces the missing context, the hidden constraint, the better option, the place where the argument was weaker than it looked.

A lot of my year has been spent in rooms where positions get tested. Roadshows where the idea gets challenged and dinners where guests come at power and the energy transition from a different angles. AI helps me prep faster than I could have a year ago. It pulls background, surfaces prior positions, stitches together data I would have spent days gathering. That is the cheap part. What it cannot do is tell me which theme deserves to lead in each room. The same analysis lands differently with a long-duration investor than with someone running a trading book. The position I carry into a meeting is not the model’s output. It is what I am willing to defend, calibrated to who is across the table. The work product is not the deck. It is the set of positions that survive contact with people who have skin in the game.

The position I carry into a meeting is not the model’s output. It is what I am willing to defend, calibrated to who is across the table.

So what does taste actually decide? Three things have to attach to the analysis before it is worth anything. Judgment, context, accountability.

Judgment selects from the expanded possibility space. This is the part people point at when they say taste, and they are half right. Knowing when a sentence sounds good and says very little. Knowing when a model is precise around the wrong question. Knowing when a caveat is real and when it is throat-clearing.

Context shapes the surviving candidate into something that fits a specific room. Some of it is explicit: budget, operator, risk tolerance. Some of it is social: who got burned last cycle, which objection is real, which concern is performative, when the room is ready to hear the answer. AI can produce situated analysis if you hand it the situation. It cannot read the room.

Accountability is the part the soft version of taste leaves out, and the part that matters most. A recommendation can be handed off. A position has a person standing behind the reasoning long enough for it to be tested. That does not require certainty. It requires conviction strong enough to defend and humility honest enough to update. The sentence that matters is, “This is the best answer I can defend, given what we know and what we are trying to do.” AI can produce the reasoning behind it. Accountability still has to attach to a person.

If I am right about this, the analysts and firms that treat AI as an output multiplier are going to drown in their own polished noise. The ones that treat it as a position-generation system are the ones organizations will actually listen to. The work product is the quality of the positions someone is willing to defend, the speed at which they update them under pressure, and the track record of carrying them into rooms where decisions get made.

The work product is the quality of the positions someone is willing to defend, the speed at which they update them under pressure, and the track record of carrying them into rooms where decisions get made.

That is what taste is when output is free. Taste is the willingness to be wrong in public, on the record, with your name attached. The product is a position, because a position is the form analysis takes when it is ready to move.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Enhanced geothermal, Fervo and its peers, fractures hot rock and pulls heat fast by convection. The price is water lost to the formation. Closed loop, Eavor and XGS, seals the wellbore and loses very little. The price is a slow heat rate and higher upfront capital cost. Conventional hydrothermal, 17 GW worldwide, is the proven option, bound to the rare geology it needs. Superhot, Quaise and Mazama and 400 C, promises a step change in heat the engineering cannot yet deliver. Four bets, one pool of capital.

That isolation looks like a problem. Capital is finite, and a sector fighting itself for dollars fragments the story for every allocator trying to underwrite it. Fervo just priced a roughly $1.89 billion IPO at a $7.7 billion valuation. Eavor put 0.5 MW on a German grid in December, a quarter of its 2 MW first phase. Those are different risk profiles reaching for the same check.

But the cost of the tribalism is smaller than it seems. Below the surface the tribes stand on common ground. Fervo’s edge is multistage fracturing and fluid conformance lifted wholesale from the Permian, and SAGD operations in the Oil Sands. Eavor drilled Geretsried to 4.5 km with SLB. Quaise runs Nabors rigs and has raised $120 million. The conversion side leans on off-the-shelf hardware, ORC and flash steam, that predates all of them. The result? While the subsurface dialects diverge, the toolset is shared.

That shared toolset is why the fragmentation does not have to hurt. When a service company shaves a week off a lateral or drops cost per meter, every tribe benefits at once. Fervo cut drilling times 70% between its pilot and Cape Station. That learning curve does not respect tribal borders. It is one curve, and the whole sector rides it, whichever subsurface approach turns out to have the most velocity.

The hyperscalers already sense it. Meta signed 150 MW with Sage and another 150 MW with XGS. Google lined up both Fervo and Ormat through the same Nevada tariff. They are spreading chips across the tribes because the ground underneath is the same.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

If you build on frontier models, the question is no longer theoretical. How exposed are you to a switch you do not control?

The lesson is that rented frontier capability is revocable. Backup models help, but they do not solve the deeper problem. A model you do not own can be recalled, restricted, degraded, or made unavailable through a decision made above you. You may find out at the same time your users do.

That risk always existed in theory. This week it became an operating assumption.

The common read is that the newest model is always the prize. I think that is incomplete. The newest model is also the least understood model. It has the least operating history, the least mapped behavior, and the greatest regulatory surface area. For many commercial workflows, stepping back from the frontier gives up less capability than people assume while reducing exposure to the most switchable part of the stack.

For most companies, the edge was never only in the model. It is in proprietary data, domain judgment, workflow design, and evaluation. A near-frontier or open-weight model, paired with the right data and measured against the work that actually matters, can beat a generic frontier model on the jobs that pay you.

Keep using frontier models. Do not build the company around the assumption that one will always be available.

We have spent years talking about data sovereignty. Inference sovereignty is next. Controlling where your data sits is different from controlling the model that reasons over it. The expertise you layer into a model through tuning, memory, skills, prompts, evaluations, and workflow design becomes an asset you either own or rent.

The firms that demanded data privacy will increasingly demand inference sovereignty too: control over the model, the reasoning layer, and the institutional know-how embedded inside it.

The practical answer is architecture.

Keep model IDs behind an abstraction layer. Maintain fallback paths across frontier, near-frontier, and open-weight models. Move memory, skills, evaluations, and routing logic outside any single provider. Treat frontier access like an interruptible input: useful, powerful, and not fully under your control.

For energy, the bigger implication is fragmentation.

If more firms decide they need inference sovereignty, they will not all build hyperscale campuses. They will want controlled inference capacity closer to their data, operations, customers, or regulatory boundary. Some will sit in smaller data centres. Some will sit behind the meter. Some will be embedded inside industrial sites, campuses, labs, hospitals, banks, utilities, and defence-adjacent facilities.

The aggregate load may stay the same or rise.

Hyperscale compute is efficient because it pools demand, runs infrastructure hard, optimizes cooling, and shares capacity across many users. Sovereign inference moves in the other direction. It favours control over utilization, proximity over scale, redundancy over sharing, and permission over pure cost minimization.

That sacrifices system efficiency for control. For firms worried that a provider, regulator, or government can interrupt access to their inference layer, the trade may be rational.

For grids, this makes AI demand harder to see and harder to plan around. Large interconnection requests will still matter, but more compute may appear as smaller clusters embedded inside commercial, industrial, institutional, and behind-the-meter load. The forecast becomes less about known hyperscale campuses and more about a sovereignty premium: lower utilization, more redundancy, and more distributed compute.

Inference sovereignty does not stop at the model. It extends through the chip, the site, the interconnection, and the meter. If the goal is control, compute and power become part of the same stack.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

What’s driving the move? Time to connect, growing pressure to cover most or all of the connection costs. It’s a bit like cutting the cord with cable. At first it looked easier, and cheaper. But over time our family found ourselves “needing” Netflix, Prime (free delivery), Disney+ (kids), and even a cable (sports)… a stack of streaming logins that keep us connected to new partners and, still, the cable company.

The data center industry is living the same experience with power. The promise of BTM is simple. Bring your own power, and you can skip the queue, run more DC supply (and capture the 13% gains from NVIDIA’s MW block), and maybe save money. Some will end up truly islanded, self insuring their reliability. But many will keep a tie to the grid. It just will not be for the full load.

The logic is simple. Sizing the wire for full peak means demand charges on the whole load, a network build, and years in the queue. The on-site fleet carries the bulk energy and the rack battery covers the fast swings. The grid tie becomes the expensive insurance layer: useful for startup, maintenance, partial import, future resource integration, or curtailed-load backup, but not necessarily sized to carry the whole campus. So you self-supply the cheap part and rent the expensive part on a thin tie, sized well below peak. The wire size decouples from the operating size. That is the whole trick.

The real deals already look like this. Tallgrass is building its Cheyenne Power Hub to bring its own power to Crusoe’s Project Jade, up to 2.7 GW of dedicated on-site gas on its own pipeline. That is about as close to off-grid as a project gets, and it still designed a grid interconnection into the plan.

This is why FERC and ERCOT are not writing rules for islands. They are writing menus, from PJM’s Non-Firm Contract Demand tier to ERCOT’s controllable-load and bring-your-own-generation elections. The grid is learning to sell a partial connection, and the data centers are learning to buy one.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

For powered compute the builders are assembling fast. SoftBank, DigitalBridge, and ArcLight are stacking demand, digital infrastructure, and power under one $150 billion-plus roof. Blackstone put $5 billion into a TPU venture with Google. KKR and Energy Capital Partners staked $50 billion. Each looks like what a homebuilder sells: one counterparty that delivers the finished product, powered compute, and manages the trades that go into it. xAI already sells it, with Anthropic paying around $1.25 billion a month for all of Colossus 1, roughly 300 MW of powered compute.

But not every buyer is worth building for, and load is not one thing. For example, our electrification forecast adds about 24 GW by 2035 from heat pumps, EV charging, and industry, each with its own price elasticity and none in a hurry. Hyperscalers and model labs are different: large, concentrated, and creditworthy, and they care about far more than the commodity price: time-to-power, reliability, density, cooling, scale, execution certainty all factor. Microsoft is paying Constellation at least $100/MWh to restart Three Mile Island, double the market, for that bundle.

The timing is right. Sourcing powered compute keeps getting harder: more scale to coordinate, a grid clogged with big new loads, generation costs inflating, and local and regulatory pushback. Between the offtake and an energized megawatt sits a chain of sub-scale counterparties, the powered-shell developer, the IPP, the turbine slot, the chip schedule, each handoff a seam where the timeline can tear. The buyer is bankable. The middle is the problem. The platform is the alternative, collapsing the chain into one balance sheet.

Integrated, de-risked cash flows carry a far lower cost of capital than a chain of weak bilaterals could. Scale procurement and fewer coordination failures pile on. The customer’s total cost can fall even as the platform earns a strong return. That is also why the gas gets built. A merchant plant at $2,000 to $3,000/kW cannot clear against a capacity cap of $333.44/MW-day, well short of the ~$500 it needs, but a sponsor holding the hyperscaler’s contract underwrites it off-grid.

This is business model innovation. The margin lives in the cost of capital. Whoever owns the fewest seams and the deepest balance sheet wins.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

The combined entity becomes the third largest energy company in the United States by market value, behind only ExxonMobil and pushing Chevron for second. Adding Dominion adds scale to NextEra, but will adding bulk (like “The Missile” Magnussen) bring performance (Magnussen finished last in both his events)?

Perhaps. There are clear benefits to balance sheet scale and operational breadth when serving data center concentrated markets. And Dominion serves Northern Virginia, the most concentrated data center market on the planet, with 51 GW of disclosed pipeline. NextEra brings balance sheet, NEER’s development engine, and multi-year relationships with MSFT, GOOGL, AMZN, and META. The bundle is the product. Regulated grid service. Nuclear firm power. NEER solar and storage for 24/7 carbon-free matching. A behind-the-meter bridge during queue waits. Hard to compete with that enhanced offering.

But the Enhanced Games also provide a warning. Kristian Gkolomeev broke a world record, but it won’t be recognized because he used banned substances and a banned suit. How much of the benefit of this combination will be allowed? NEE and D pre-emptively offered $2.25 billion in bill credits across Virginia, South Carolina, and North Carolina. The Virginia SCC will still want more. Will affiliate transaction limits prevent compelling bundles of regulated and unregulated services?

Even if you can outperform your competitors, will you have the opportunity to perform? We have discussed at length the inflated signal of PJM’s load queue. Those risks extend to Dominion’s 51 GW pipeline, 10.4 with ESA and 29.5 GW of which sitting at the riskier substation engineering letter stage. The interconnection backlog has loosened with the move to first-ready first-served, but there is still years of work ahead. And then there are PJM’s capacity markets: under the current cap new gas capacity does not get financed.

The strategic case is the easy one. Whether the enhanced offering converts to performance, like the Games themselves, is what we will be watching.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Highlights from the listing: 70 million shares at $27, upsized from 55.5 million, popped 33% on the open. Third largest US energy IPO in 15 years, behind Kinder Morgan in 2011 and Plains GP in 2013, both midstream MLPs. The recent energy reference list is otherwise LNG infrastructure and state-owned privatizations. Fervo’s $10 billion debut as a venture-backed producer is more noteworthy in that company.

Some napkin math on the price tag. Mark the 500 MW under construction at $2 million per MW, roughly where CEG and VST trade. Mark 3 GW of advanced development at $0.5 to $1 million per MW. That covers $2.5 to $4 billion. The market ascribes the remaining $6 to $7.5 billion to 38 GW of early-stage portfolio, around $150 to $200 per kW.

E&Ps pay for undeveloped sticks. Why not for EGS sticks? Depends on whether they are scarce. For EGS, that is a question of land, technology diffusion, and substitution.

BLM geothermal bids are up tenfold since 2023, with top bids hitting $412 per acre in Nevada’s record 2025 sale. Fervo built its 595,900-acre position at roughly $4 per acre between 2019 and 2021. ORA and Buffalo River Minerals are the new public bidders. Our Geothermal Acreage piece ran the numbers.

Geothermal acreage well sited for the grid is a lot like tier one shale acreage. Early grabs are the advantage, if you got the right rock. More than a few E&Ps missed the core (anyone remember Talisman Energy?). Did Fervo get it right?

Technology diffusion will come. Fervo has an exploitation answer key that is working today, with a learning curve compressing LCOEs. That is an advantage for now. But the dependence on drilling and completion service providers means, like with shale, those advantages will not stay Fervo’s forever.

Power markets are not precious about generation. An electron is an electron. The substitute to watch is SMRs, not solar plus storage. X-energy raised $1 billion last month on the same thesis. Fervo has 658 MW of 15-year PPAs signed across SCE, Shell, Google, and two California CCAs. $7.2 billion of revenue backlog, implied near $93 per MWh. SMRs are still chasing site approvals. That is a wide gap.

Most booms echo the past. With EGS, I hear the echoes of shale: the land capture, the rapid improvements (and falling breakevens), and technology diffusion. If those echoes foreshadow another energy renaissance, last week’s IPO is its first signpost. Congratulations to the Fervo team.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Consider power and energy transition, where I spend most of my time. Project economics are now hyper-location specific. Load, land, interconnection, deliverability, fuel, policy, saturation, buyer behavior. Get the combination right and the asset prints. Miss one variable and the site goes nowhere.

The harder problem is the rate of change, and you would be forgiven if it leaves you feeling a little off balance. Environmental attributes are trading more like derivatives of the political environment than the output of supply and demand. IPP valuations have rerated in recent months on the appetite of hyperscalers to sign PPAs against a subset of their assets: see our take on NRG and CEG. Data centers are moving behind the meter, but for how long? The pieces are too interconnected, excuse the pun. Power markets now need scenario-based models that clear energy, capacity, attributes and project economics together.

Zoom out and the old pattern is visible. Value accrues to whatever is scarce, deliverable and contractable. Everything abundant gets discounted to the floor.

Which brings me to the AI sessions, and the perspective that reasoning itself is becoming abundant. Every company will soon have access to capable models trained on public data, which means the model will not be the moat. One speaker pushed it further: AI is an amplifier. It does not change what you are, it amplifies what you already are. Leaders pull further ahead, laggards fall further behind, the middle gets squeezed.

It sounds hyperbolic, but the undertone was there in the coffee chats and dinners. Folks are excited. But there was also quiet conviction that everybody is behind and the hard questions are just beginning. For some, enterprise restrictions limit which models get deployed. For others it was apprenticeship loss. How does a 25-year-old analyst get the reps to know what good looks like when the first draft is always waiting, or when the MD prompts the model instead of asking the analyst?

For everyone, whatever they are doing, it is not enough.

I feel it too. But that may be the point. If reasoning is abundant and AI is an amplifier, advantage moves underneath the model or above it. Underneath, sits proprietary data, controls and tech. Above, sits the judgment to make decisions before the model has the answer. Both are hard, but the kind of hard AI will reward.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note published through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

I have been going down the rabbit hole of AI tools, fully convinced of their power and the probability they will be transformative for the world of work I engage in. On the surface it feels like software, where I engage with agents in natural language and the agents are the enabler of my work. This is incomplete. As I engage further with this technology I can feel my own role shifting, and with it, my relationship with AI.

A core element of my role is to harness my team, a mountain of data, and our capacity to produce ideas and insights and make them consumable and actionable for our clients. This is an intensely human experience. So many decisions are dictated by the comfort and conviction of human beings, based on the trust and credibility of other human beings. Decisions of significance invariably happen in person, where the whites of one’s eyes matter as much as the quality of the case being presented.

What changes as I and our team adopt AI is that my team now includes an army of agents and rapidly growing capacity to generate ideas and content. Set the question of quality aside for a moment. Extrapolate forward, where the models and agents become increasingly competent, and I can imagine the army of agents becoming the dominant source of ideas and insights. Those ideas need a curator and an evangelist in the rooms where they can influence real people making real decisions. I am still needed, if only as an interface for my AI.

That is an oversimplification, but carries several implications relevant now.

Implication 1: The bottleneck is my ability to consume ideas.

For most of my career the scarce resource was production. Finding the data, doing the analysis, writing it up cleanly. AI has flattened that curve. I dictated the first version of this post at about 150 words per minute, 4-5x faster than I could type it, and Wispr Flow ensured the first draft already absorbed my verbal corrections and removed the filler words. Claude turned that draft into something readable and helped refine it into something serviceable before my coffee got cold. It is not perfect, but then, this is also the worst the AI will ever be.

Consider what just happened. My bandwidth from mind to paper expanded with AI-assisted dictation. Those ideas were refined and expanded with at least the same acceleration versus what would have happened in the recent past. Combined, that is 20x the output bandwidth.

I am not alone. Every analyst, PM, and consultant with an API key has the same superpower. The result is my overflowing inbox. The problem is that my input bandwidth has not expanded at the same rate. My ability to read, absorb, and hold something in working memory long enough to do anything with it is similar to five years ago. Call it the read-write asymmetry, where output bandwidth is expanding rapidly and input bandwidth is comparatively fixed.

Implication 2: Choose errors of exclusion over errors of inclusion.

If the flow of inputs coming at me is growing faster than my consumption bandwidth, my instinct is to become stricter about what I consume. That means avoiding errors of inclusion (losing time on the weaker content) and accepting errors of exclusion (missing great pieces to preserve the quality of the corpus I do consume). In practice that is a short list of writers, analysts, and primary documents that consistently deliver more ideas per word than anyone else, and a much harder no on everything else.

But strict filters kill the accidental find. The piece I never would have searched for that turned out to reframe a problem or create a connection I had not considered. There is serendipity in foraging that I do not want to lose. For now I am keeping fifteen or twenty percent of my read budget in deliberately adjacent territory. That allocation still needs a plan.

Implication 3: The bar for output has moved up.

When the cost of producing okay content drops toward zero, the things that make output great are exactly the things AI does not do by default: falsifiable claims, non-obvious reframes, and specific numbers anchoring an abstract argument. These tools allow the author to commit to a point of view rather than hedge, and to make a call rather than balance the debate. When supported with great arguments and analysis this is what I want to consume and I expect others do too.

Implication 4: The real work becomes developing taste.

Taste used to be built by making things. Years of writing badly, getting feedback, trying again, watching what works in front of an audience and what does not. If most of the making is now done by a system, where does that judgment come from?

I think the answer has two parts that have to be held in tension.

The first part is that you still have to do some of the work. It is inefficient, and the throughput pressure of AI-assisted speed pushes against it constantly. But there is a kind of understanding that only comes from sitting with a hard problem long enough to feel its actual shape: build the model from raw data once in a while, write the hard paragraph by hand, and run the analysis without the agent. There are lessons for evaluating work that you only earn by producing it.

The second part takes advantage of the technology and runs more cycles than were previously possible. Each attempt is a chance to make a call, see the result, ask what worked and what fell flat, and update. That also can be how taste gets built. The volume is the new training data for my judgment, if I actually use it that way. The risky default mode of AI-assisted work is to ship faster. Hit go, the output is good enough, move on. That improves throughput. Combine cycle count with reflection, grounded in the wisdom of having done related work, and you get an accelerated apprenticeship.

Where does this leave me?

I can feel my role evolving. Knowing the data and the operators and the history, the input-side expertise that used to be most of the value, has not gone to zero. But that responsibility is shifting elsewhere. The job is to shape what the system, my team of people and agents, produces into something a person can use in the room where the decision gets made. The judgment is still mine. The compression is the product. The taste that drives both is the thing I have to keep working on, using the same tools that threaten to erode it.

It feels like I am becoming the UX for my AI. The question is what it takes to be a good one.

I show up to events like this with a few themes and questions in mind. Top of mind walking in is power as the forcing function this year. Last week the four big hyperscalers told the market they will spend $650 to $725 billion on AI infrastructure in 2026, roughly double 2025. Capex now runs at 47% of sales for MSFT and 54% for META. Wild changes from just a couple years ago. Unpacking what is required to generate returns on that spend is a discussion for another day (it is starting to feel very shale circa 2014 to me). For now, it is enough to know that any returns are gated by speed and access to power.

Constraints combined with capital lead to innovation. And the latest innovation is more like circumvention. Skip the grid. There are real reasons most loads choose to interconnect, but when local pushback grows, when the queue wears you down, when you are already self-insuring reliability beyond what the grid offers, and when power is a small (but still multi-billion) share of a $50 to $100 billion plus project, the calculus shifts.

The calculus of constraints is showing up elsewhere. Energy security is visible now in a way it has not been in years. Hormuz keeps oil bid every day it stays contested. The industry is openly discussing Big-E exploration, new unconventional plays beyond North America, infills, refracs and EOR. I did not hear a consensus favorite at our EnergyEdge event in February. But that was before the Iran conflict.

As for questions, I am most curious about the AI vibe check. I believe AI is here, the hype is justified, and the gap between lab bench and scaled monetization is where the real work is. But where is the room at? Who is bought in. Who is using it. Who is creating value with it. Enverus ONE is our answer. I will be watching everyone else’s.

If you are at Evolve this week let me know. I would love to say hi.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note I publish through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

Gas generation in 2026 is having a Ferris kind of day. GE Vernova’s gas turbine order book crossed 100 gigawatts at the end of Q1 2026, up from 83 GW just one quarter earlier. Siemens is similarly stuffed. Hyperscalers are signing direct. EIR has grid-connected gas builds averaging 10 GW per year, double the prior two-year pace. Pricing on new turbine bids is up 10% to 20% above the 4Q’25 backlog average. The day is bright.

I also remember Cameron, in the garage, listening to the odometer click. Here too, there is tension building. A tension we described two weeks ago in The Sibling Squabble: grid-connected gas burn for power stays roughly flat through 2030 even as load grows 10% and capacity grows by tens of gigawatts. Solar, wind, and batteries conspire to compress gas utilization across more hours of the year. New capacity does not translate into meaningfully more energy delivered.

The note left a few nuances understated. The demand is real, but bottlenecked. Heavy-duty turbine slots are sold into 2029. The chip supply chain (EUV throughput, memory, global allocation) is growing but still gates how fast data centers can consume the power gas is meant to serve. Supply chain is the binding constraint.

Against that constraint, gas demand can grow anyway. An increasing share of incremental data center load is being met by dedicated behind-the-meter gas generation and distributed generation hardware (aeros, recips, fuel cells) with more slack in their queue. Those molecules (around 1.8 Bcf/d by EIR’s count) show up in the gas balance, just not in grid-connected power burn.

But there are more miles still to reverse off the odometer. Coal retirements keep slipping, almost every quarter, deferring the substitution trade gas has been waiting on for a decade. SMRs and geothermal are no longer slideware. GEV expects an NRC license to construct at Clinch River as soon as 2H’26. Fervo just filed its S-1. Both arrive in the early 2030s on cost curves still grinding down, ready to compete for capacity and dispatch in that decade.

The movie ends differently for each of the players. IPPs do not want load moving off grid. Gas producers do not want more capacity that runs less often. Turbine OEMs do not want optionality on the order book to evaporate before it converts. Each of them gets a Ferris headline today and a Cameron question on the drive home.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note I publish through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.

For most of the industry’s history, load was the variable you forecast and generation was the variable you dispatched. Dial the turbines up or down to match the demand, and job done. Ok, it’s a bit harder than that… but you get the idea. Our weekly ISO lookback showed load forecasts were solid across PJM, MISO, and NYISO and beat ISO metrics in ERCOT, CAISO, and SPP. Last week load was defeated, and prices still surprised. Vanquish one challenge, two new ones emerge.

It’s shoulder season so those surprises are revealing. Spring generation in the Lower 48 runs roughly 20% below summer peak. Temperatures are mild, cooling load is minimal, and the grid should be the furthest thing from strained. It is also exactly when generators schedule maintenance, because the system has slack to absorb it. And yet, strip load out as the hard variable, and the forecasting problem does not get easier.

The modern supply curve now moves on its own clock, and the moves are large. ERCOT wind swung from roughly 12 GW at the weekly trough to nearly 28 GW at peak, often in a few hours, sometimes early, and sometimes late. SPP looked quiet all week and produced a single Friday spike when wind shape collided with a binding constraint. This week, NYISO wind is expected to drop nearly 90% from early week to Thursday or Friday. Plenty of capacity to surprise.

Then there is the head that does the most damage. Generation has not just changed type, it has changed location. Wind sits in West Texas and the Plains, solar in the desert Southwest, hydro upstream of population. The wires were built for a fleet that sat closer to load. Supply that looks abundant on a system map is frequently stranded on the way to where it is needed.

Place this against a backdrop where load is rising at the same time: data centers, electrification, and industrial reshoring are pushing the baseline higher even as the supply stack gets lumpier. When weather or congestion pulls renewables off the curve, the market climbs up the merit order, and the top of the stack is steep. A few MW of missed supply or unexpected demand can clear hundreds of dollars per MWh above where the average day settles.

This is today’s market, one that pays for reading between the lines a model cannot absorb. I am glad we have a team doing that work every day.

Comments, questions or things I missed? Send me a note (or hit reply) - I would love to hear from you. Thanks for reading!

Morning Energy is a syndicated note I publish through Enverus Intelligence. My contributions will also be distributed here. Please note that links frequently lead to content available only to subscribers of Enverus solutions. Please reach out if you have any questions. Thanks! - Ian.