First the Neighbours Objected. Now the Investors Are.

For two years, the resistance to the data centre boom has come from the ground up: planning committees, parish councils, campaign groups, people whose water pressure dropped and whose nights got noisier. This month it arrived from somewhere else entirely. The people who were supposed to pay for all of it started asking for a way out.

The catalyst was Oracle's decision to serve a force majeure notice on the developer of Project Jupiter, the vast AI campus it is leasing in New Mexico. Force majeure is the clause normally reserved for acts of God — the earthquake, the flood, the war. On Prof G Markets, Scott Galloway and Ed Elson put it more sharply than any filing: "when Oracle invokes force majeure, what they're really saying is that the god is credit markets."

The Trigger: Oracle Asks for an Out

The notice, first reported by Bloomberg on 24 September, seeks the right to delay rent payments if the data centre is not operational by its scheduled 2028 completion date. Oracle's contractual responsibility is to secure power for the site, and a gas pipeline permit has slipped; the campus itself is only about a quarter built, with its fuel-cell microgrid far further behind.

The company says the project remains on schedule and that it is committed; Blue Owl, whose STACK Infrastructure unit is developing Jupiter, says the notice changes nothing about its commitment to the multi-year project. The market read it differently.

Oracle shares fell more than 3%, and at one point more than 6%, on the news. The stock is down roughly 30% year-to-date.
Blue Owl Capital shares fell around 3% that day. The firm has roughly $3bn of equity in Jupiter, and its stock has fallen about 45% from its high.
Around $18bn of loans tied to the project have traded below par — as low as 89–91 cents on the dollar — a signal that lenders are pricing in delay.

That is the irony the podcast seized on. A force majeure clause is meant to protect a customer from a catastrophe. Oracle is using it to protect itself from its own supply chain.

The IPO That Didn't Happen

If Oracle was the signal, the initial public offerings were the proof. Several data-centre-adjacent listings that were meant to price this month have stalled:

SB Energy — the SoftBank-backed developer that is building AI campuses, with OpenAI as a flagship tenant — postponed its Nasdaq IPO. Bankers could not fill a book at a valuation reportedly targeting over $50bn; the listing has slid to late October "or later".
Holtec (referred to in some coverage as WholeTech Nuclear) postponed its IPO indefinitely. Founder and CEO Krishna P. Singh said the nuclear business is now "viewed as connected to" data centres — and that the shift in sentiment around AI demand, combined with local opposition, had created a "perfect storm".
A generation company servicing data centres also delayed its listing.

Note the mechanism here. These are not companies that failed a test of profitability. They are companies whose story stopped working. When bankers cannot find buyers, the smart money has quietly stopped buying the pitch.

Why SB Energy Is the Emblem

SB Energy is the clearest illustration of what the market is now refusing to fund. Per the financials cited on the show and in coverage, the business reports roughly $214m of revenue, down 8% year on year — a declining company — and sought a valuation of over $50bn, on the order of 230-odd times sales. It has no operating AI data centres. Only around 9% of its contracted capacity is under construction; the rest has not broken ground. It promotes a multi-hundred-billion-dollar backlog that, on inspection, is not going to be delivered in two years, five, or arguably ten. The solar business it grew out of is worth on the order of $1.2bn.

It is, in the show's phrase, "the canary asking for a $50 billion valuation on a mine that hasn't even been dug yet." NVIDIA has invested about $3bn into the business and has agreed to backstop a large portion of OpenAI's lease on an Ohio campus SB Energy is developing — a structure that keeps the demand signal inside the family.

The Dot-Com Rhyme

The comparison the hosts keep returning to is the fibre build-out of the late 1990s. Telecoms laid tens of millions of miles of dark fibre — cable that was installed and never lit. Money poured into Global Crossing, into Cisco, into the whole idea that bandwidth demand only went one way. Global Crossing went bankrupt; Cisco lost around 90% of its value.

Their line is the one to remember: "It wasn't that the fibre was wrong. It was just early." The corollary matters more for anyone holding a share certificate: "it's the definition of wrong when it goes on a balance sheet." The infrastructure may well be needed. The timing of the cheque is what goes broke.

The Physical Friction Is Real

The financial wobble is landing on top of a physical one. The pushback datacentretruth has been documenting is now measurable in money:

45 projects worth about $68bn were blocked or delayed in the second quarter of 2026 alone — more than half of all large projects newly proposed in that quarter, per the Data Center Watch research covered by the LA Times and Gizmodo.
The first quarter was worse: 75 projects worth roughly $130bn stalled or were delayed, taking the first-half total to nearly $200bn.
Analysts estimate that 30–50% of the data centre capacity expected online in 2026 faces delay or cancellation, on power constraints, equipment shortages and opposition.
Polling is emphatic. A Gallup survey in March found 71% of Americans opposed to an AI data centre in their local area; a Heatmap poll this month put it at 71% again, and NBC News at 69%.
The midterms are becoming a referendum on AI, and politicians are moving against it because the voters are.

Against that backdrop, CoreWeave's CEO told the Q2 earnings call that "we feel like moratoriums are not going to impact the demand for this infrastructure." Moratoriums do not repeal demand — but they do decide where, when and at what cost that demand gets served. The hosts read the line as wilful blindness rather than confidence. It is hard to disagree.

Private Markets Lie. Public Markets Don't.

The sharpest framing in the episode is this: "Private markets lie and public markets don't." Private valuations can carry a stupid mark for a long time, because no one is forced to print a price. Public markets mark granularly, in public, every day — and when they refuse to buy, there is nowhere to hide the number. The IPO window closing is not an opinion. It is a refusal.

This is why the AI trade looks K-shaped. On one side are businesses with real, audited cash flow — NVIDIA, the hyperscalers, the semiconductor supply chain. On the other are businesses selling potential: SB Energy, Holtec, the neoclouds, the special-purpose developers. The second group is, as the show puts it, "about to get very familiar with the term cost of capital."

The counterpoint matters. This is not a blanket verdict on AI. NVIDIA's forward price-to-earnings multiple has compressed to roughly 24x as of late September — near its lowest in years against a ten-year average nearer 53x. That is not bubble pricing; it is a market that has already de-rated the profitable end of the trade. The scepticism is concentrated in the speculative tail, not the whole industry.

Meanwhile, the Concrete Keeps Pouring

And yet the build itself has not slowed. The point is the decoupling: the financing is wobbling while the steel keeps going up.

Hyperscaler capital expenditure is running at roughly $750bn for 2026, and consensus forecasts put it above $1 trillion in 2027.
US data centre construction spending hit a record $81.5bn by June 2026 — already more than all of 2025 — and ran at an annualised pace above $75bn in July, up nearly 60% year on year.
North American absorption is at record levels with vacancy below 2% in the major markets, much of it pre-leased before completion.

So both things are true at once. Tenants are still signing. Lenders are getting pickier. The two can coexist for a while — and the gap between them is the story.

What This Means in the UK

British readers have every reason to watch this closely, because the UK is where the physical and financial friction meet hardest.

Grid. As of July, 315 data centres were queued for a grid connection, representing some 73GW of demand — close to half again the UK's entire peak electricity demand, with some projects waiting until 2037. Ofgem has proposed a "commitment fee" of 2.5–7.5% of project cost and hard milestones to shake speculative applications out of the queue.
Water. Industry body Water UK has warned there will not be enough water for the planned boom, and a report found 84% of proposed water-intensive data centre developments sit in areas already water-stressed or projected to be by 2040.
Planning. Edinburgh refused a 210MW data centre in February over climate and local-plan grounds. The City of London has imposed a two-year moratorium on new data centres. The Scottish Parliament rejected a full national moratorium in September but froze planning decisions pending national guidance due by Christmas.

Read together with Oracle and the failed IPOs, the pattern is a single argument arriving from two directions at once: the same projects that communities say should not be built are the projects the capital markets are increasingly unwilling to fund. When both sides of that equation agree, the moratorium stops looking radical.

What Communities Should Do

Ask who is paying, and how. For any proposal, require the financing structure in plain terms: who holds the debt, at what cost, and what happens to the site if the money stops. A project that cannot be financed cannot be built, however many hectares are optioned.
Force the schedule question. Demand the construction timetable against contracted capacity — how much is actually under construction, and how much is only announced. "Contracted" is not "built", as SB Energy's numbers show.
Demand the grid and water commitments in writing. A connection-queue position is not a connection. Ask for the date, the capacity and the water source — and press the council to refuse speculative applications that are queuing only to hold a place.
Use the financial wobble as leverage. Delay is now a market-visible risk. A well-argued objection that pushes a timeline out by 18 months can make the finance stack unworkable — that is the pressure point a planning committee actually controls.
Connect local fights to the national picture. Individual refusals raise the perceived cost of capital for the whole speculative tail. Coordinating across campaigns is how a village objection becomes a market signal.

A caution for campaigners: falling valuations are not a win on their own. A stalled project can leave a half-built shell, a stranded grid connection, or an operator undercutting on price to fill capacity. The goal is better projects and honest answers, not just cheaper share prices.

What's Next

Three things to watch. First, whether SB Energy's IPO prices in October or slides again — that is the cleanest read on whether investors will fund the speculative end at all. Second, whether Oracle's notice becomes a template: if other hyperscalers reach for force majeure clauses when power slips, the "guaranteed" demand that underwrites the whole build-out starts to look optional. Third, the UK planning guidance due before Christmas, which will decide whether the freeze in Scotland becomes a national framework or thaws.

The neighbours objected first, and for concrete reasons: water, power, noise, land. It turns out the people with the deepest pockets are asking the same questions. "It's the definition of wrong when it goes on a balance sheet."

Sources: Prof G Markets podcast, "Investors Are Turning Against Data Centers (Here's Why)"; Bloomberg and subsequent reporting on Oracle's Project Jupiter force majeure notice (24 September 2026); LA Times / Gizmodo / Data Center Watch on Q2 2026 blocked projects; Gallup (March 2026), Heatmap and NBC News (September 2026) polling; CreditSights and Goldman Sachs hyperscaler capex estimates; US Census construction spending data via CoStar; Ofgem grid connection reforms, Water UK, and Scottish Government planning statements. Figures from the podcast's discussion are attributed as its interpretation, not as established fact. This is journalism, not investment advice. Part of the Brightfull Tech network. Contact: hello@brightfull.tech