TL;DR
Google's 3,590 MW Constellation deal starts delivering in 2028; its 500 MW Kairos agreement targets 2035. Read as a portfolio, the commitments sort by delivery date. The most substantive reply pushed the logic further: sorted by arrival date and cost, the largest bucket isn't nuclear at all.
A procurement leader described what lateness costs: when a 2030 source slips to 2033, the load still arrives on time, and the gap gets covered with whatever power is left, usually at the worst price. The PJM rule that would have made lateness more expensive was largely rejected by FERC on October 9.
Finland halted clearing at two Google data center sites pending an environmental review. In Memphis, xAI ran turbines for about a year under a temporary-generator exemption before the county issued permits. Paid readers get the financing argument a practitioner built from the pause, and the question it puts to site selection teams.
---
A megawatt that arrives in 2028 and one that arrives in 2035 are not the same purchase, and a site review that happens before clearing and one that happens after are not the same risk. This week's two most-discussed posts were about who carries the cost of a date.
Coverage This Week
Dozens of climate lawsuits wait on one Supreme Court case - Suncor v. Boulder County, argued October 5, could decide whether climate liability gets settled once or argued court by court. Read →
Gas is scarce this winter and may be in surplus by 2030 - Qatar's LNG stall and Europe's low storage send opposite signals to anyone building, financing or buying gas infrastructure. Read →
Ten US senators have $100,000+ riding on data centers - Senate disclosure minimums, and the federal energy and tax law those members help write. No wrongdoing is alleged. Read →
The US embargo on Cuba now has a solar loophole - US solar exports to Cuba are rising from a tiny base through an embargo exception, with US Gulf and Florida ports an increasingly used route. Read →
Finland paused clearing at two Google data center sites - two regulatory systems, two responses to the same industry, and the gap lands on builders, financiers and neighbors. Read →
The weather may make the degrowth decision for us - insurers paid roughly $107 billion for natural disasters in 2025, and that bill reaches everyone through premiums and project risk. Read →
Google is buying nuclear power for 2028 and for 2035 - portfolio allocation logic applied to hyperscaler nuclear procurement, with delivery date as the sorting mechanism. Read →
This Week’s Signals
Each signal below traces the practitioner debate on the week's most-discussed posts and who the post reached. Audience figures come from LinkedIn's viewer demographics for each post; industry comparisons are against earlier posts on the same topic and are stated as upper bounds. Who reads a post is attention, not demand.
Google Bought Arrival Dates, Not Megawatts
Google's new Constellation Energy deal is 3,590 MW, with first upgraded output due in 2028. Its Kairos Power agreement is 500 MW, targeted for 2035, with a first 50 MW plant in 2030. Seven times the power, arriving seven years sooner. Read together, the two deals look less like a technology bet than a portfolio allocation: the largest commitments go to power that can arrive soonest.
The headline number needs unpacking. Of Constellation's 3,590 MW, 890 MW is a 20-year agreement tied to upgrades at 11 existing units in Illinois, Pennsylvania and New Jersey; the other 2,700 MW is not tied to a specific plant.
Across hyperscaler nuclear deals, the sequence appears to hold. Power from plants already running comes first: Meta's 20-year Clinton deal supports the plant's relicensing, with deliveries from June 2027. Restarts come next: Microsoft's offtake from the Crane plant targets 2027, Google's Duane Arnold deal early 2029. Small modular reactors sit at the far end as option stakes, with Amazon's X-energy plan at up to 5 GW by 2039. These are company schedules, and Google hasn't called its deals a portfolio; the sort is a read of the published dates.
The PJM context has moved since the deal. PJM proposed that new large loads bring their own new capacity or be curtailed first under grid stress, and the companies said the Constellation deal responds to that. On October 9, FERC largely rejected that framework, finding it intruded on state jurisdiction over retail customers, and kept only PJM's large-load definition and a registry. The penalty for arriving late is now less formal than the deal anticipated.
The most substantive reply came from a solar-industry CEO who accepted the sorting logic and pushed it past nuclear: "If hyperscalers allocate by arrival date and deliverability, the largest bucket isn't nuclear at all. Wind, solar and batteries arrive in one to three years, at the lowest cost per MWh." The evidence offered was the scale of the same buyers' renewable deals, including Microsoft's 10.5 GW framework with Brookfield. The answer to the post's closing question: "I'd hold most of my power budget in what's already being built, and treat everything with a 2035 delivery date as an option, not a plan." The counterpoint a grid planner would raise is that interconnection gates arrival for every technology, and the lowest cost per MWh is not the lowest cost per firm megawatt for a 24/7 load. About half the 26 comments became a separate argument over Australian generation costs and battery safety, which does not bear on US procurement.
The sharpest procurement point was shorter. A technology leader in procurement described what lateness costs: "If a 2030 source slips to 2033, the load still arrives on time, so the buyer covers the gap with whatever power is left, usually at the worst price. Pricing each option with that bridge cost included shows why the soonest power gets the biggest share." For most hyperscaler deals the gap is financial before it is physical: the utility still serves the load, so a late source shows up as unhedged market and capacity-price exposure and a shortfall against 24/7 clean-power goals. That is the mechanism under the post's read, and why the post expects suppliers who can show a delivery date and financing to win larger allocations.
Nuclear Electric Power Generation made up 5% of this post's viewers, more than its share on earlier nuclear posts (at most 2.67%), and viewers at companies with 5,000+ employees were 31% against a 20% average. Reactions ran about double the nuclear-topic average, while the thread (26 comments) was smaller than this topic's typical 47. (Composition: Nuclear Electric Power Generation 5% vs up to 2.67%; 5,000+ employees 31% vs 20.01%; reactions 208 vs 104 topic avg; comments 26 vs 47.)
> Does PJM, or a state such as Ohio or Pennsylvania, bring back a delivery-date penalty for large loads, or did Google buy firm nuclear against a rule that no longer exists?
Finland Stopped the Digging. Memphis Permitted After the Turbines Ran.
Finland's supervisory agency ordered work halted at Google's Muhos and Kajaani data center sites by October 23, pending a mandatory environmental impact assessment. It is investigating whether hundreds of hectares were cleared without one; Google says the felling followed forest rules. The order is a pause, not a ban: felling, topsoil removal, digging and quarrying stop while planning and surveys continue. Google's announced investment of at least 13 billion euros in Finland stands.
The post set it against Memphis, where xAI ran gas turbines for roughly a year under a local exemption for temporary generators before the county permitted 15 of them in July 2025. The two cases involve different reviews, land use in Finland and air permits in Memphis, but in both, review came after work started. The difference is what came next: a regulator with the power to stop irreversible work, and it used it. It is one case in one country, a data point rather than a Nordic pattern.
Behind the paywall: the financing argument one practitioner built from the pause, the Texas trade-off that points the same direction, what it means for US sites, who the post reached, and Field Notes.
An energy sector professional took the post's "assess before irreversible work" point and turned it into a financing argument: "Assessing before irreversible work starts is the cheaper path for everyone, and that includes the developer. A pause on felling and digging is a lot easier to price than a reversal after the ground has moved." The commenter saw the same direction in Texas, where the large-load conversation "feels like it's heading toward front loaded rules too, with curtailability traded for faster interconnection." The conclusion: "My read is that predictable review timing matters more to financing than how strict the review is." The comment ended by asking whether site selection teams are starting to price regulatory timing the way they price power cost, or whether it is still mostly a gut call. A second commenter agreed that timing matters as much as the rules: "uncertainty after work begins can turn a manageable review into a costly schedule and trust problem." Apart from the author's reply, the remaining comments did not engage the timing question.
The argument holds up because a pause has a known shape. Work stops, the assessment runs, and a lender can model the delay. A reversal after clearing has no such shape. The Texas trade applies the same logic to grid access: the developer accepts a defined constraint up front in exchange for a predictable schedule. In the US the gap is structural. A private data center on private land generally triggers no federal environmental review, because NEPA needs a federal nexus such as federal funding or a federal permit, so review timing depends on state and local rules that vary site by site. If site selection teams start pricing regulatory timing the way they price power, jurisdictions with predictable, front-loaded review become a site advantage.
The post traveled further than it was answered. Impressions reached 22,697, more than three times the data center topic average, while reactions (32), comments (8) and reposts (none) all ran below typical. Construction made up 6% of viewers, more than its share on earlier data center posts (at most 3.9%), and viewers at companies with 5,000+ employees were 29% against a 20% average. (Composition: Construction 6% vs up to 3.9%; 5,000+ employees 29% vs 20.01%; impressions 22,697 vs 6,623 topic avg; reactions 32 vs 69; comments 8 vs 29.)
> How long does Finland's assessment of the Muhos and Kajaani sites take, and does clearing resume on the original plan or under new conditions? The answer is the first test of whether a pause is as priceable as the commenter argues.
Field Notes
US solar exports to Cuba are rising from a tiny base: $6.3M of US cells and modules in the first seven months of 2026, through an exception to the embargo. China still ships more: $19.5M so far this year, 83% below its full 2025 total. Most of Cuba's solar imports are still made in China, but Florida, Texas and Louisiana ports are an increasingly important route. Demand is the driver: Bloomberg counts the seventh national blackout of the year, and resilience is being bought behind the meter, from whichever supplier the politics allow. Read →
Ten senators, $100,000+ in data center holdings: Forbes read the Senate disclosures, which report only ranges, so the figures are minimums. Owning stock in a sector is legal and no wrongdoing is alleged. The post's point is that cost allocation for giant new loads is decided at FERC, under federal law Congress writes, and at state commissions, and when voters can see the portfolio and feel the power bill, they assume the two are connected. Projects that disclose who covers the grid upgrade, and who holds a stake, are likely to face fewer fights. Read →
---
The thread running through this week: when delivery is uncertain, the cost lands on whoever holds the schedule. A hyperscaler pays to cover a late megawatt; a developer pays for review that arrives after the ground has moved. If you are placing capital or building infrastructure, which date in your plan are you actually pricing?
Reply if any of this is playing out at your company, or contradicting what you're seeing on the ground. Every reply goes directly to our analyst desk and feeds our intelligence.