---

Superhot rock geothermal lives or dies on drilling, the one thing the oil patch has spent a century getting good at. That is why the most interesting part of Mazama Energy's $135 million raise isn't the 331°C at its pilot site. It's that ConocoPhillips and Shell Ventures are on the cap table.

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; "topic norm" compares a post with the average of earlier posts on the same topic (n = posts in that comparison). Who reads a post is attention, not demand.

Mazama Energy raised $135 million for superhot rock geothermal at Newberry Volcano in Oregon. Centaurus Capital and Doerr Capital led, with ConocoPhillips, Shell Ventures, Khosla Ventures and Gates Frontier in the round.

The hard part of superhot rock isn't the physics. It's drilling deep, hot, hard wells on budget, with similar rigs, the same crews and the same subsurface playbook pointed at a different product. Mazama says it reached 331°C at its pilot site, which it calls the hottest engineered geothermal system in the world. That is very hot, but still short of the roughly 374°C usually used to define superhot conditions. The cap table is the story: if superhot rock works, a driller's core skill becomes a clean-power asset, and Shell and ConocoPhillips hold an early seat. The fine print matters: 331°C at depth isn't power on the grid. The first power demo is planned for 2027, the 200 MW by 2030 target is a roadmap, not a contract, and the water and well-count savings are company claims not yet proven at scale. Fervo is further along, with contracted projects in Utah, and Quaise is chasing the same prize with a completely different way of drilling. Nobody has won this yet.

The thread backed the drilling logic and questioned the rest. A clean-energy advocate made the case directly: "Gas and oil companies are already really good at drilling. Geotherm needs experienced drillers." An exited climate-tech founder agreed that drilling should be "bread and butter" for the majors, but asked whether surface equipment such as heat exchangers is ready for 331°C. A senior project engineering manager was more skeptical: Newberry "has been a research and development site since the 70s," and "at this pace of deployment no company can absorb the development costs." Another reply flagged the commercial gap: technical fit is strong, but incentive structures and deal cycles would be a culture shift for oilfield service teams.

That last point is where the read-through sits. The large oilfield services firms already sell into geothermal. The question isn't whether they can enter; it's whether geothermal ever becomes material to their revenue, and a 200 MW roadmap is a small number of wells. If Mazama's 2027 demo delivers, the next thing to watch is contract scale, not temperature. If it stalls, the drilling-pivot thesis loses its timeline anchor.

Greater Houston viewers were about 3x their usual share at 13%, the top location ahead of Calgary at 6%, and Oil and Gas was the largest viewer industry at 38%, at least 2.2x its average on oil-related posts. That audience shape is consistent with the region and industry that hold the drilling capability the thesis depends on. Reach was narrow, at roughly half the topic norm, and sends per impression ran about 5x the 90-day average, though on a base of 4 sends that is too small to read as a forwarding pattern. (Composition: Greater Houston 13% vs 4.35%; Oil and Gas 38% vs up to 16.63%; Renewable Energy Equipment Manufacturing 7%; Renewable Energy Power Generation 3%; CXO/VP 17% vs 17.55% average; sends 4 (4.95x per-impression rate); impressions -48.3% vs topic norm (n=122).)

> Do superhot rock and enhanced geothermal developers sign multi-rig, multi-year drilling contracts with major oilfield services firms by 2028?

Read the original post →

Strip data centers out of the numbers and US business construction is shrinking. That is the thesis, and it drew the week's sharpest disagreement.

Census data for July shows total US construction spending down 3.8% from a year ago. Private nonresidential ticked up 0.4% for the month, and according to Associated Builders and Contractors, that entire increase came from data centers. Without them, nonresidential spending fell for a second straight month, to its lowest level since September 2023. Data center construction is running above $75 billion a year, close to 60% higher than last July. Power and highway spending are growing too, but in private nonresidential, one customer type is carrying the category. Meanwhile, factory construction is down 22% this year and roughly a third below its 2024 peak, nearly all of it chip and electronics plants from the CHIPS Act boom finishing up with nothing behind them. When a few buyers set your demand, they also set your pricing, your payment terms, and your downside if AI spending slows.

Several of the 26 commenters rejected the premise outright ("not true," "fake news"), and others called it overstated, a point the post itself conceded about the viral line. A data center infrastructure executive was blunt: "Hottest sector yes, but not true." Others pointed to stadiums, agricultural projects and growth "only in select states." The practitioners who engaged with the mechanism went further than the post. The CEO of a consulting firm agreed the concentration risk is real but pushed back on the margin story: "When one buyer type sets the demand, they also set the spec. A hyperscale operator locking in a 15-year contract for 50 megawatts of cooling means every mechanical contractor in that region is now quoting to that standard, which kills half-measured approaches. The margin exists only if you do it right the first time. Rush the commissioning, and you're eating SLA violations."

Two others described how concentration actually surfaces. A senior practitioner in the energy sector: "Concentration rarely shows up in volume first, it shows up in terms. The anchor customer doesn't cancel. They pay slower and negotiate harder, and the backlog looks healthy right up until it doesn't." They floated the share of backlog tied to projects with a signed tenant lease behind the construction financing as a candidate for the earliest honest signal. A senior executive in data center markets described the sequence: "the order book is the last place you see it. First the inquiries dry up. Then the forecasts get cut. Only then do the signed jobs run out, often two or three quarters after the customer actually changed their mind." An energy founder, citing ABC backlog data, put roughly one in six of its member contractors under contract for data center work.

The post's exposure list is equipment makers and engineering firms: transformers, switchgear, turbines and engineering hours. Who feels a slowdown first is less obvious. With widely reported shortages and long lead times in transformers and turbines, their makers hold leverage today. Contractors and developers building without a signed tenant sit closer to the edge. The factory pipeline already shows how quietly a boom ends. Projects finish. Nothing new shows up.

Founders and startup operators made up 7 of the 26 commenters, about 1.5x their usual share, and Construction was the largest viewer industry at 16%, at least 4.0x its average on data center posts. Impressions ran at more than 3.6x the data center topic norm, which is consistent with the post traveling well beyond its usual audience. (Composition: Construction 16% vs up to 3.96%; Oil and Gas 6%; Utilities 5%; IT Services 4%; companies with 201-5,000 employees 31% vs 22.71% average; CXO/VP 21% vs 17.55% average; impressions +262.6% vs topic norm (n=195).)

> Does the share of data center backlog without a signed tenant lease start rising, or do equipment reservation slots start getting deferred?

Read the original post →

Panels hit 12 cents a watt on the global spot market. US rooftops still cost about 25 times that. A typical US residential system runs about $2.50 to $3.50 a watt installed, and the panel is 10 to 20 percent of that. The post argued that the next decade of US solar belongs to whoever cracks the 80-plus percent of the bill that isn't the panel: permitting, interconnection, labor, customer acquisition, financing and margin. And US buyers don't even get the 12-cent panel. US-assembled modules run around 30 cents, with tariffs and foreign-content rules doing the lifting, and those rules tighten again in January.

Almost nobody in the thread defended the panel as the problem. The fight was over which part of the rest to attack first.

Behind the paywall: the four camps the thread split into on cost, whether the post's 30-cent figure holds up, the AI construction-cost signal and what practitioners said about it, and Field Notes.

On cost, the thread split four ways. Permitting: an architect and energy manager pointed to Australia, where, by his account, installation costs run less than a sixth of the US level, and a research lead argued that rooftop interconnection should be treated "like an appliance rather than a conventional generating station." Interconnection: an energy founder named it the hardest bottleneck and noted that systems already built have paid the permit, interconnection and acquisition costs, which makes keeping them producing about the cheapest solar there is. Labor and sales: one commenter reported materials under $5,000 against $17,000 for a one-day install, and a solar founder said 40% of their cost goes to sales, which they blamed on gatekeeping rather than on finding customers. Hardware redesign: a solar hardware founder challenged the line between hardware and soft costs entirely: "Better hardware architecture can remove downstream soft costs," by cutting parts, engineering, sourcing and installation steps, so solar systems become more like factory products and less like field projects. A smaller group skipped the debate and pointed to plug-in and balcony solar, which sidesteps much of the installed bill.

Two challenges deserve a straight answer. An adviser to infrastructure and energy boards called the framing obvious: equipment is always only a fraction of landed cost. True, but the point of the post is the size of the fraction and that it didn't fall with the panel. And a solar company co-founder asked which US panel is selling at 30 cents a watt. Fair test, and the number holds: Anza's July pricing data put the median US-assembled module at about 30 cents a watt, with modules using US-made cells higher still.

In NREL's residential cost benchmarks, customer acquisition has typically run well above permitting, which is why the sales-cost argument deserves more attention than it got. The commercial read: the firms positioned to take share are the ones selling a reduction in one of these line items, with evidence, not another cheaper module.

Reactions ran +122% vs the 346-post solar topic norm, and founders or startup operators made up 9 of the 25 commenters, about twice their usual share on solar posts, which is consistent with a thread led by people building around the soft-cost problem. (Composition: Renewable Energy Equipment Manufacturing 14%; Construction 4%; Utilities 4%; CXO/VP 16% vs 17.55% average; reactions +121.6% vs topic norm (n=346).)

> Does any US state or utility measurably cut residential permit or interconnection time in 2027? One commenter cited Virginia's online solar permitting tool, due July 2027, as the first test.

Read the original post →

Suffolk Construction and MIT took a finished $180 million San Francisco apartment project and ran a counterfactual: what if it had been built with today's AI tools? Their answer was 17 to 20% cheaper, 22 to 25% faster, and roughly five to six extra points of IRR for the owner. The authors are upfront that it's a single project, built from interviews and roundtables, and that the results are "not causal." The savings came from six levers stacked together: design automation, offsite manufacturing, permitting, scheduling, labor planning and procurement. The post argued that most firms will buy a single scheduling tool, wait for a fifth of the budget to disappear, and call it hype a year later.

Practitioners on the thread were the most skeptical of the headline, and their reasons sharpen the post. A senior procurement leader put the mechanism in two sentences: "Most buying is set up to buy one lever from one vendor. The 20% lives in the handoffs nobody owns." A commissioning manager went to the contracts: "everyone acts from a standpoint of shedding risk onto others," and "it's trust you will need for these 'AI levers' to work as advertised." The principal of a construction firm was blunter: "the people who are telling you that are just trying to sell you software."

The commercial read: the 20% is a contracting problem before it is a software problem. A vendor selling one lever into a risk-shedding contract structure is selling into the exact gap the study says swallows the gain. The firms with a real claim on the number are the ones who can get owner, designer and builder onto one shared outcome.

Saves were the highest multiple in this issue at 2.31x the 90-day average save-to-reaction rate (15 saves on 50 reactions), which may indicate readers keeping it for reference, though the reaction base was small. Construction was the largest viewer industry at 30%, followed by Real Estate at 11% and Architecture and Planning at 9%, an audience concentration that matches the post's subject. (Composition: Construction 30%; Real Estate 11%; Architecture and Planning 9%; Civil Engineering 8%; CXO/VP 21% vs 17.55% average; saves 2.31x (per reaction); impressions +260.1% vs topic norm (n=665).)

> Do owners start tying AI tool purchases to shared-outcome contracts across designer and builder, or keep buying single tools?

Read the original post →

---

The thread running through this week: capital is rarely the constraint. Execution capacity and buyer concentration are. Australia's data centers wait on electricians and grid connections. Superhot rock waits on whether drilling contracts ever reach real scale. Solar's savings sit in permits, interconnection and sales. The AI construction gain leaks out in the handoffs. Nuclear waits on its first units, and US business construction waits on a few buyers continuing to build.

If you sell into any of these markets, the question to pressure-test before next year's budgets lock isn't whether the demand is real. It's whose execution, and whose buying decisions, your revenue depends on.

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.