Morning Analysis — 2026-09-07

Dek

Money for AI data centers has never been easier to raise; permission to build them has never been harder to get.

The Big Shift

Texas's grid operator, ERCOT, will finish its governor-ordered audit of data center power requests by December — a hard deadline that puts a public number on how much of the "AI load boom" is real signed demand versus speculative placeholders in the interconnection queue (the waiting line to plug into the grid) (¹). Developers routinely file the same project with several utilities at once, so the queue is almost certainly inflated; nobody knows by how much. That matters because the whole power-demand trade rests on the assumption that queued megawatts convert into contracted, paid-for electricity. December is now a scheduled event risk for anyone long Texas generation — Vistra (VST) above all, given its Comanche Peak nuclear plant sits next to an Amazon deal.

Analysis

The bottleneck has moved from equipment to consent. A year ago the scarce thing was transformers and turbines. Overnight the pattern was different: a Polish council in Masovian Voivodeship voted *unanimously* to kill a data center (²), the first data center approved on US federal land is stuck in court in Nevada (³), and five more states are lined up for statewide building pauses (⁴). This *confirms* the core Own-the-Bottleneck claim — scarcity is in power and permission, not chips — while *breaking* the comfortable version of it, where load growth is smooth and linear. Own the chokepoint, yes, but the chokepoint is now a zoning board and a judge's docket, and those don't respond to price signals.

Capital is already routing around the "no." Three projects broke ground in places that say yes: Humain's 1.5GW campus at Neom in Saudi Arabia (⁵), a €360m Milan site that already has 80MW of power locked ("power secured" is now the headline feature, not the footnote) (⁶), and a renewables-powered AI campus in Malaysia. For electrical-equipment names like Eaton (ETN) and Vertiv (VRT), this is the offset to US siting friction: the transformers, switchgear, and cooling get sold regardless of the postcode. The implication is that domestic moratoria hurt US *generators* far more than they hurt the picks-and-shovels suppliers.

Materials stay tight, and the tightness is showing up in odd places. Copper held just under its record on a thin US-holiday tape, with traders still pricing supply risk (⁷); CATL signed new deals for battery copper foil; and two men were arrested for stripping copper in rural Kentucky. That last one is a joke of a headline with a serious point — when a commodity gets valuable enough to steal from active sites, physical security becomes a line item. Meanwhile grid equipment makers are onshoring but transformers still take two to three years to deliver (⁸). For Cleveland-Cliffs (CLF), maker of the specialty electrical steel inside those transformers, the deficit isn't clearing — new US capacity arrives too late to relieve it. Copper strength is already in the Freeport (FCX) and Southern Copper (SCCO) prices; nothing overnight changes that.

The energy-cost side is turning against the on-site power story. Energy Aspects says crude is "poised for a significant leg higher" while disruptions in the Strait of Hormuz — the narrow sea lane carrying roughly a fifth of the world's oil — continue (⁹). Iran-Oman talks are trying to keep shipping moving, and gold slipped as traders read those talks as constructive (¹⁰), but the US Energy Secretary is now saying there may be no nuclear deal at all. Sustained fuel inflation raises the price of the natural gas that Bloom Energy's (BE) on-site fuel cells burn — which is precisely the mechanism that would break the "our cost curve keeps improving" pillar. This is a risk building, not a break, but it's building in the direction that hurts.

One thread nobody is pricing: physical attacks on the grid. Reports of sabotage on extra-high-voltage lines in two German states show how simple methods can trip protective systems and take capacity offline (¹¹). Combine that with Chinese pressure on Taiwan's periphery — a new scholarly claim on the Philippines' Batanes Islands in the Luzon Strait (¹²) and the fight over how UN Resolution 2758 gets read (¹³) — and you get the same lesson from two directions: the AI buildout's assets are long-lived, fixed in place, and increasingly worth attacking. Grid hardening is an underwritten line of spend.

Housekeeping: Vertiv filed two 8-Ks on Sept 2 — one flagging a material agreement plus a Reg FD disclosure (a company voluntarily releasing information to everyone at once), one an unspecified material event. Neither has been read. That's still the single largest unknown on the desk, bigger than anything in today's news flow.

What Would Prove Us Wrong

• ERCOT's December audit finds the Texas queue is mostly real — say, under 20% phantom or duplicate requests, with a high share backed by signed contracts. That would kill the "speculative load" worry, restore the smooth-growth version of Own-the-Bottleneck, and argue for putting VST conviction back up.
• Copper closes below $10,000/tonne for a sustained stretch, or Grasberg output news resolves cleanly in Indonesia. A break below that level hits pillar 1 (structural deficit) for both SCCO and FCX; a Grasberg resolution would remove the execution discount capping FCX.
• Hormuz shipping normalizes and Brent falls back with diesel cracks in tow. That reopens the cheap-gas path for Bloom Energy's pillar 3 and turns today's cost-inflation risk into a non-event. Watch the Iran-Oman talks and whether any nuclear framework survives — the Energy Secretary's "there may not be" comment says don't count on it.

Thesis Impact

• VST | Conviction: DOWN | Surprise: MED | ERCOT will finish the governor-ordered Texas data center audit by December — a hard date for regulatory scrutiny of large-load interconnection in VST's home market, where the Amazon/Comanche Peak co-location sits. Not thesis-breaking, but it CONTRADICTS the assumption that ERCOT DC load converts to contracted power without friction; the audit could expose speculative queue volume. Watch for scope disclosure ahead of December. | ¹
• Own-the-Bottleneck | Conviction: HOLD | Surprise: MED | The disconfirming thread is stacking: ERCOT audit (NEW), a Polish council voting unanimously to block a data center (NEW), the Nevada federal-land project stuck in court, and five states lined up for moratoria. This CONTRADICTS the smooth-load-growth version of pillar 1 while CONFIRMING the core claim that scarcity sits in permission and power, not chips. The binding constraint is migrating from equipment to siting consent — same thesis, different chokepoint. | ²
• BE | Conviction: DOWN | Surprise: MED | Energy Aspects calls crude "poised for a significant leg higher" with Hormuz disruptions ongoing. Sustained global fuel inflation pressures the gas feedstock that on-site solid-oxide economics depend on — directly adjacent to the thesis-breaking trigger. CONTRADICTS pillar 3 (cost curve improving toward broad economics). Offsetting note: signal 19 shows Iran-Oman talks trying to keep shipping open, so this is a risk building, not a break. | ⁹
• FCX | Conviction: HOLD | Surprise: LOW | Copper held just under its all-time high on a thin US-holiday tape, with traders still pricing supply tightness; separately CATL signed battery copper-foil supply-chain deals. CONFIRMS pillar 1 (structural deficit), but price strength is already in the prior — what's capping FCX at 0.60 is Grasberg/Indonesia execution, and there was nothing new there overnight. | ⁷
• SCCO | Conviction: HOLD | Surprise: LOW | Same copper print, nowhere near the <$10,000/t break that would threaten the thesis. CONFIRMS pillars 1 and 2; fully reflected in a 0.86 prior. No Peru/Mexico policy news overnight — that remains the only thing that moves this name. | ⁷
• ETN | Conviction: HOLD | Surprise: LOW | Three new builds broke overnight: Humain's 1.5GW Neom Oxagon groundbreaking, a €360m/80MW Milan project with power already secured, and a Malaysian renewable-powered AI campus. CONFIRMS pillar 1 (electrical content in every data center, geographically diversified) and cuts against any capex-cut trigger. Exactly the expected news flow — no re-rate. | ⁵
• CLF | Conviction: HOLD | Surprise: LOW | Recurring print: grid OEMs are onshoring, but transformer lead times are still 2–3 years. CONFIRMS pillar 1 — the deficit is not clearing, and new US capacity is arriving too slowly to relieve it. The residual to watch (a second GOES producer, tariff softening) did not appear. | ⁸
• VRT | Conviction: HOLD | Surprise: LOW | Siting friction is the one thread that could bite order growth — Nevada's court block, the Polish rejection, and the moratorium watchlist all slow deployments VRT equips. CONTRADICTS pillar 2 at the margin, but nothing here is a hyperscaler capex cut, and the Neom/Milan groundbreakings offset it. Note the two unexplained VRT 8-Ks from Sept 2 (material agreement + Reg FD) are still the bigger unread item. | ³

Inflection Radar

[emergent] Physiological Digital Biomarkers | Combining cardiac, neural, and speech data via Transformers to create comprehensive physiological state models, moving beyond single-sensor tracking. | Touches: NEW | ¹⁴

[emergent] Semantic Geolocation | Using LLMs to fuse semantic understanding with IP data for city-level geolocation, enabling highly contextual digital service targeting and compliance enforcement. | Touches: NEW | ¹⁵

[emergent] Human-Aware Robotics | Developing Transformer models (H2INT) to enable safe robot navigation by explicitly modeling how human movement changes in response to the robot's presence. | Touches: NEW | ¹⁶

QA & Caveats

VST: The claim that the audit contradicts friction-less conversion is an interpretation of regulatory scrutiny rather than a direct contradiction of the load assumption.

BE: The assertion that fuel inflation pressures solid-oxide economics is not directly proven by the crude price outlook; it relies on an economic projection adjacent to ongoing geopolitical risk.

FCX/SCCO: These calls are restatements of existing supply tightness and pricing reality, indicating low genuine surprise.

VRT: The contradiction with pillar 2 is offset by specific project announcements (Neom, Milan), making the overall call supported rather than unsupported.

Sources

1. ERCOT to complete Texas data center audit by December datacenterdynamics.com
2. Data center development halted in Masovian Voivodeship, Poland datacenterdynamics.com
3. The Solar Farm-to-Data Center Switcheroo Hits a Snag heatmap.news
4. These 5 States Are Primed for a Data Center Pause heatmap.news
5. Humain breaks ground on Neom’s Oxagon data center datacenterdynamics.com
6. Greenfield and Finsbury Infrastructure to invest €360m in Milan data center datacenterdynamics.com
7. Copper Hovers Below All-Time High as US Holiday Spurs Caution bloomberg.com
8. Grid OEMs are racing to build in the U.S., but transformers still take 2–3 years - MarketScale news.google.com
9. Crude Poised for Significant Leg Higher, Energy Aspects Says bloomberg.com
10. Gold Edges Lower as Traders Focus on Oil Price Swings and Dollar bloomberg.com
11. Attack on high tension grid reddit.com
12. A new claim on the Batanes, from an old Chinese playbook aspistrategist.org.au
13. UN Resolution 2758: when interpretation becomes institutional power aspistrategist.org.au
14. BioSync: Transformer-Based Cross-Modal Fusion for a Multimodal Physiological Digital Biomarker arxiv.org
15. IPGeoAI: Transformer-Based Geolocation with LLM Semantic Fusion arxiv.org
16. Human-Human & Human-Robot Interaction Transformer (H2INT) for Robot Navigation in Dense and Uncertain Crowds arxiv.org