Newsletter · 2026-09-29 · Ashutosh Agarwal

Oracle Invokes Force Majeure Over Power Permits While Turbine Makers See No Slowdown - Powering AI Infrastructure - Week of September 29, 2026

The powering-AI-infrastructure weekly for the week of September 29, 2026, synthesizing operator and investor podcasts on the power buildout. Oracle invoked force majeure on its 2.45 GW Project Jupiter campus over power permits, Morgan Stanley said GE Vernova is in talks to contract turbines for 2031 and 2032, and Bloom, Boom Supersonic and PowerSecure pitched on-site power as permanent.

Powering AI Infrastructure

Week of September 29, 2026: Oracle Invokes Force Majeure Over Power Permits While Turbine Makers See No Slowdown

Last Wednesday, Oracle ($ORCL) did something none of the AI-power bulls had on their bingo card. It invoked force majeure on Project Jupiter, its 2.45-gigawatt data-center campus in New Mexico.

Force majeure is the contract clause that says "something outside our control happened, so we're not on the hook yet." In this case the something was permits for power.

Oracle's stock fell about 4% that morning. Bloom Energy ($BE), one of the project's power suppliers, got hit too. Blue Owl ($OWL), the equity behind the project, slid as well.

And yet, the same week, the people who actually sell turbines and engines into this build told a room full of investors that nothing has changed. Not one booking. Not one delivery slot.

That split is the story of the week. The politics are now reaching real contracts. The equipment order books haven't noticed.

TL;DR

Politics hit a real contract. Oracle's force majeure on its 2.45 GW New Mexico campus is the first time the power-permit fight showed up in a hyperscaler's paperwork, not just in headlines. Bloom Energy and Blue Owl took the hit.
The equipment queue got longer, not shorter. Morgan Stanley says GE Vernova ($GEV) is now in talks to contract large gas turbines for 2031 and 2032, with prices expected to keep rising into the 2030s.
"Nobody wants the grid" is becoming a business model. Bloom, Boom Supersonic and Southern Company's PowerSecure all argued that on-site power is permanent, not a stopgap. That matters for engine makers, fuel cells and gas.

What's new

1. Oracle's force majeure: the backlash finally touches a contract

This week on Squawk on the Street (9AM Hour, Sept 24), CNBC's David Faber walked through what happened at Project Jupiter. His read: this is not about money.

"This is delays having to do with permits for power and things of that nature in New Mexico that conceivably are having a delay on the actual build of the powered shell that eventually will house a campus as much as 2.45 gigawatts of power."

His theory is that Oracle doesn't want to start paying rent on a building that can't switch on. Leslie Picker added a source's view that Oracle is "stuck in the lease," and the notice just makes the lease longer. A spokesperson told her the notice "does not change the financial commitments to this multi-year project." Oracle itself said Jupiter "remains on their planned schedule" and noted it is "reimagining Project Jupiter's power plan."

On Stock Market Today With IBD (Sept 24), the hosts connected the dots to Bloom. Their summary of Oracle's position: "if it doesn't get built on time, we don't want to have to pay our suppliers. And one of those suppliers is Bloom Energy." The stock dropped to its 21-day moving average and bounced, which the IBD host read as strength: "for a stock that has really a pretty high ATR, with bad news, this could have been, you know, pretty poor."

Why it matters: For months, the political pushback lived in hearings and permit freezes. Now it has a dollar sign attached. A hyperscaler used a legal clause to push out payments because the power wasn't ready. Every supplier with a data-center contract will re-read their termination and delay terms this week.

2. The turbine queue now runs to 2032, and prices keep climbing

The single most bullish data point of the week came from Morgan Stanley's Thoughts on the Market, "AI Meets the Physical Economy" (Sept 25). Analyst Dave, fresh from the bank's Laguna conference, said management teams across power equipment were strikingly consistent.

"We all see the headlines about data center moratoriums, political pushback, community challenges... But at least with the power equipment companies, they're just not seeing it."

The specifics:

GE Vernova is "now in conversations to contract turbines for 2031 and 2032."
Smaller engine makers building on-site units for data centers are taking reservations into 2029 and 2030.
Utilities, the biggest buyers of this gear, are now planning "farther and farther now into the 2030s. That's new."
Pricing "has been rising," and companies signaled it will likely keep rising "into the 2030s." In his words, "we just haven't seen any signs of softening yet."
No change "in bookings or slot reservations for equipment deliveries."

Why it matters: When a factory is sold out five-plus years ahead and still raising prices, the risk isn't demand. It's execution and whether the customers at the far end of that queue still exist in 2031. This is analyst reporting of what management said, not management itself, but the consistency across big and small players is the point.

3. On-site power stops calling itself a "bridge"

Three very different operators made the same argument this week: data centers are going to make their own power, permanently.

Bloom Energy's commercial chief. On Catalyst with Shayle Kann, "The rush for clean, on-site power" (Sept 28), Aman Joshi, Bloom's EVP and Chief Commercial Officer, said "most of the customers are going to Bloom solution as a permanent power. They're not thinking of it as a bridge." (This is a sponsored "partner content" episode, so weigh it as a company pitch.)

His case is about what grid power really costs once it reaches the building. The headline "6 or 8 cent" power is the price at the power plant. Then you add:

Step-up transformers: "$5 to $10 a megawatt hour"
High-voltage transmission: "another $20 to $30 a megawatt hour"
Substations and distribution lines on top of that

He also took a shot at LCOE, the standard "levelized cost of electricity" math that averages a plant's lifetime cost per unit of power. For on-site power, he called it "wrong math," because it ignores the extra capacity you need for reliability, water savings and AI's spiky load. He claimed some rival setups waste "15-20% of the energy just trying to run dummy workloads" to smooth out those spikes.

Boom Supersonic's CEO. On All-In, "Blake Scholl: Why Plane Speed Stalled..." (Sept 22), Blake Scholl explained that his jet-engine core, minus the fan and plus a generator, becomes a 42-megawatt trailer-mounted power plant for data centers. It needs no water. His factory "will scale to making multiple gigawatts per year," and "over the next five years, we aim to add 10 plus gigawatts to the grid." Asked if it prints money:

"It's an incredible money printing business. I mean, I've never seen a demand like this for anything else in my life... There are tens of gigawatts of demand in my inbox."

His first engine runs next month. Then, he said, "we're going to auction it all off."

Southern Company's PowerSecure. On Energy Central, "Why Utilities Are Turning to DERs as Power Demand Surges" (Sept 25), Todd Jackson of PowerSecure gave the utility-side view. DERs, or distributed energy resources, are small generators and batteries spread around the grid instead of one big plant. His numbers:

About 35 to 40 GW of DER capacity installed today, up roughly 33% from last year.
Industry projections that DERs could cover 15% to 20% of US peak demand by 2030.
The bottleneck has moved. Utilities have "done a great job" solving generation capacity; the problem now is transmission, with "two-year, three-year timeframes for transmission upgrades."

PowerSecure's answer is a product called BridgePlus: reciprocating engines that power the site while the wires get built, then become backup power afterward.

Why it matters: Morgan Stanley framed the debate neatly. Utilities and big turbine makers say "all this data center demand is going to the grid eventually." Small-equipment makers "say nobody wants the grid" and sign 15-to-20-year contracts. Morgan Stanley's own view is that on-site power "is going to be an extremely large market as we get toward 2030," and that both camps win for now.

4. Transformers: the five-year wait

On Today in Manufacturing, Ep. 287, "America's Transformer Crisis" (Sept 28), the editors of Manufacturing.net and Industrial Equipment News walked through an essay by Michael Farkas, CEO of NextNRG. The key figures:

Wait times for large, custom transformers have stretched to up to five years.
Transformer demand in 2027 is expected to more than double 2020 levels.
About 75% of distribution transformers are already past their expected service life.

Co-host Jeff Reinke added a cold dose of reality: permits and funding tack "another year and a half to two years" onto that timeline. He also noted that the alternative, solid-state transformers built from ferrite and silicon carbide, mostly depends on Chinese supply.

On Redefining Energy, Ep. 248 (Sept 28), a guest put the trade problem bluntly: the US "only manufactured 20% of its transformers and imported 80%," largely from countries now facing tariffs. "That's a Western story. It's not an Asian story."

Why it matters: Transformers are the quiet reason Bloom's "skip the grid" pitch lands. Every step-up and step-down on the grid needs one.

5. Uranium: a supply gap that mines can't close quickly

On Sprott Radio, "WNS 2026 and the New Buyer" (Sept 28), recorded around the World Nuclear Symposium, the Sprott team laid out the math:

The world produces about 175 million pounds of uranium a year.
Demand is about 200 million pounds and rising, with a base-case future of about 350 million pounds. That means "essentially doubling production."
Some mines now under construction were discovered last cycle and have "been stuck in the ground for 15 years."
The price "has essentially tripled in the last five years."

Costs are rising too. Sulfuric acid, a key input for in-situ uranium mining (pumping solution underground to dissolve the ore), has jumped because of disruption in the Strait of Hormuz. Kazatomprom has flagged the impact, and "Russia banned the export of sulfuric acid due to shortages."

One interesting detail: some utilities have "so much to do" that they are pushing out specific fuel purchase requests, spreading conversion and enrichment buying over time. The conversation at the symposium has shifted from fuel to "how we're going to finance these reactors."

Why it matters: A rising cost floor plus a delayed utility buying wave is a setup for term prices, not just spot. For Cameco ($CCJ) and enrichers like Centrus, the message is: more demand later, higher costs now.

The debate

The bull case: a multi-year supercycle, and everyone in the chain gets paid

Turbines are booked into 2032, with prices rising (Thoughts on the Market).
Transformers face up to five-year waits (Today in Manufacturing).
Uranium needs to roughly double supply (Sprott Radio).
Bloom's Aman Joshi on why the build isn't speculative: "this built out is based on that underlying layer of individual customers like you and I who are starting to use compute in the forms of tokens" (Catalyst).
Oracle's Mary Sprayregan even argued data centers can lower bills by spreading grid costs over more usage. She pointed to Alliant Energy CEO Lisa Barton celebrating two Cedar Rapids data-center developments that helped secure "a five-year freeze on electric rates" (Electric Perspectives).

The bear case: the demand is less real than it looks, and the timelines are too long to invest in

On Better Offline, "Monologue: The AI Data Center Overbuild" (Sept 25), Ed Zitron made the sharpest version of it. His claims:

Actual AI capacity is coming online at "maybe tens or hundreds of megawatts a quarter," far below the headlines. On estimates of 12 to 15 gigawatts coming online: "bollocks."
"Anywhere from half to 75%" of hyperscaler revenue backlogs come from Anthropic and OpenAI, "neither of whom can actually pay for them."
OpenAI made up 70% of Microsoft's AI revenue in fiscal 2026, by his reckoning.

On Frictionless, "AI spending can't grow forever with P Equity Research" (Sept 25), the guest called power "probably not a field that is investible in right now," precisely because the lead times are so long:

"Gas turbines were exciting to invest in maybe 2 years ago. Now it's probably not because all the companies Mitsubishi, Siemens, GE Vernova... has like backlogs extending beyond 2030."

He also noted hyperscaler free cash flow is "trending towards zero or kind of borrowing money," and argued that every technology eventually matures and capex flattens.

Commodity trader Fred Seguy, on Money of Mine (Sept 23), went after the core demand story. US power demand "certainly... is not going to double," he said. By his math, growth of maybe 6% a year is "more than covered by renewables additions."

He has some support. On Climate One (Sept 25), a guest cited EIA projections of 86 GW of new utility-scale capacity in 2026, "nearly double" 2025. On Energy Gang (Sept 23), a panelist cited IEA chief Fatih Birol calling air conditioning, not AI, the single biggest driver of global power demand growth.

My take: The bears are right about one thing that the Oracle news made concrete. The weak link is not equipment supply. It's the customer at the end of a 2031 delivery slot, and whether their power permit, their financing and their AI tenant all show up on time.

The names in play

GE Vernova ($GEV). The bull point is visibility: slots now being discussed for 2031 and 2032, with pricing still rising. The bear point is the same fact seen from the other side. If you can't get a turbine until 2030 or later, buyers go elsewhere (see Bloom and Boom). Next test: whether new orders keep coming at higher prices even as delivery dates stretch.
Bloom Energy ($BE). The best on-site pitch of the week and the most direct exposure to the Oracle force majeure, in the same five days. IBD's host framed the chart as a possible "bullish shakeout," or, in his words, "maybe we fall 15% tomorrow and it all ends in tears." Watch for any update from Oracle on Jupiter's "reimagined" power plan.
Quanta Services ($PWR). On Full Signal (Sept 23), portfolio manager Shanali Jain said the stock is up about 50% this year and "the backlog is huge." She noted Quanta has "delivered five triple plays in a row" (beating on earnings and revenue while raising guidance). This is a fund manager's opinion, not company guidance.
Oracle ($ORCL) and Blue Owl ($OWL). Oracle says it's committed to New Mexico. The market's question is whether a power-permit delay becomes a template other tenants use to delay payments.

Read-throughs

Gensets and on-site engines (Cummins $CMI, Caterpillar $CAT). Morgan Stanley said smaller engine makers are taking reservations into 2029 to 2030. PowerSecure's recip-engine bridge product and Boom's 42 MW turbine trailers point the same way. The longer large turbines take, the bigger the market for smaller units.
Gas producers. On The Allegheny Front (Sept 25), a Boilermakers union leader pointed to the $10 billion Homer City data-center project, which is reportedly building the largest gas-fired power plant in the country. His local has taken in its 230th apprentice in about six months after five years of taking none. EQT ($EQT) CEO Toby Rice was on the same conference panel. Every Bloom fuel cell and Boom turbine also burns gas.
Utilities and grid wires (Southern $SO, Duquesne Light). The same episode reported a Pennsylvania PUC study projecting regional power demand could roughly double by 2040, almost entirely from data centers, with a worst case of 13 blackouts a year. Johns Hopkins' Abe Silverman called that a warning about what happens if we "keep blindly signing up new data centers without requiring them to bring their own new capacity." Duquesne Light said it will invest $2.7 billion by 2030 to expand its grid. "Bring your own power" rules are good news for on-site suppliers and a mixed bag for utilities.
Load flexibility as a pressure valve. Oracle's Mary Sprayregan cited Duke University research that managing just 0.5% of new loads' peak hours could let the grid add "nearly 100 gigawatts of large load" with minimal impact. Oracle cut power use 25% during Phoenix's two peak hours using Emerald AI, while keeping compute running (Electric Perspectives). If flexibility scales, it's a partial cap on how much new generation is truly needed.
Transformers and electrical steel. The 80% import share and five-year waits support pricing for anyone with domestic transformer capacity. Tariffs make the imported alternative more expensive, not cheaper.
Uranium, conversion and enrichment (Cameco $CCJ, Centrus $LEU, physical-uranium funds). Higher acid and diesel costs raise the floor under production costs. Utilities spreading out their conversion and enrichment buying suggests demand that is delayed, not cancelled.
Hyperscaler funders. On The Investor's Podcast, TIP848 (Sept 24), the hosts cited Meta's 2026 capex guidance of $130 to $145 billion, with 2027 projected near $200 billion, and $27 billion of bonds behind its Louisiana data center. On Squawk, Faber warned the true debt load goes beyond reported capex: "residual value guarantees... SPVs... the leases."

What changed

Turbine dates moved out again. Last issue, BloombergNEF said a gas turbine ordered today comes online in 2031 at the earliest. This week Morgan Stanley said GE Vernova is discussing 2031 and 2032 contracts. The queue is still growing.
Politics went from permits to contracts. Last issue, it was Texas pausing data-center permits and the House passing the Ratepayer Protection Act 417-3. This week a hyperscaler cited power permits to delay payments. That's a step up.
The bear case got a new argument. Last issue's bears (Paul Kedrosky's 2029 debt wall, reports that a third to half of planned 2026 data centers were delayed) focused on financing and timing. This week Ed Zitron questioned whether the capacity and demand exist at all.
Behind-the-meter got louder. Last issue, the headline was Generac's $2.4 billion backup-generator deal with Amazon. This week Bloom, Boom and PowerSecure all pitched on-site power as permanent, not backup.