Weekly Macro & Outlook — 2026-09-11

Dek

The week AI's power problem stopped being about steel and started being about signatures — four statehouses, one court, two burning shipping lanes, and a dead tariff later, the scarcest input on the desk is permission.

The Week in Macro

Monday set the pattern, and it wasn't the one anybody was hedged for. A Brazilian court suspended every environmental permit at Sigma Lithium's Grota do Cirilo mine and ordered work to stop (¹), while the first data center ever approved on US federal land hit a legal wall in Clark County, Nevada (²). Two continents, two industries, one mechanism: projects that were financed, engineered and permitted got unwound by a judge. For a year the thing you couldn't buy was equipment — transformers still take two to three years. This week the thing you couldn't buy was consent, which is faster to lose and much harder to insure against.

By Friday it had gone from anecdote to trend line. Pennsylvania became the fourth state in three months to restrict data-center development, and its utility regulator opened a formal proceeding on rates, returns, and *curtailment* — the right to cut power to a big customer when the grid is tight (³, ⁴). Massachusetts tied permits to clean power (⁵). A Polish council voted unanimously to kill a project; a 120MW build in Lombardy was withdrawn when the developer couldn't produce grid data (⁶). Five more US states are flagged as primed for pauses (⁷). The clearest tell that the industry knows: OpenAI hired a JPMorgan community-engagement lead, because data centers are becoming a midterm-election issue (⁸).

Meanwhile demand stopped being a forecast and became a purchase order. Microsoft is targeting 38 gigawatts of data-center capacity by 2032, against roughly 12GW today (⁹). That's 26GW of new load — about the size of Chile's entire electricity system — from a single buyer, and it has to be firmed with gear you order now. Set that against shrinking siting options and the conclusion is mechanical: value migrates to whoever *already holds* energized, permitted, interconnected capacity, and away from anyone whose story is a pipeline of sites still awaiting approval. Texas's grid operator will quantify how much of the queue is real by December (¹⁰) — a scheduled event risk nobody has priced.

War repriced the fuel bill, and the fuel bill repriced the generation stack. Brent broke $100 and held near $101 as US–Iran fighting closed in on the Strait of Hormuz and Houthi strikes degraded Saudi Arabia's Red Sea escape route — both export doors impaired at once, the decade-long tail risk (¹¹, ¹²). Hormuz traffic has fallen to single digits. Expensive oil pulls LNG exports, tight LNG lifts domestic gas, and the IEA now expects coal demand to *rise* 1.2% as utilities substitute (¹³). Translation for compute: generation with no fuel bill — nuclear, solar-plus-storage — gets more valuable; anything that burns gas gets less. Buyers who were negotiating on price start negotiating on price *certainty*.

And copper taught the week's most useful lesson: the policy bid and the physical bid are different things. Copper hit an all-time high Monday on tariff front-running plus genuinely short mines (¹⁴). By Thursday Reuters reported the White House tariff plan had stalled on affordability grounds, and Freeport and Teck fell 7% in a session (¹⁵). Nothing changed in the ground: Chinese imports were the weakest in six years and global mine supply still heads for its first annual drop since 2017. What died was the bet that Washington would gift US producers a protected price. The supply response is visible — hyperscalers phoning Ivanhoe about Congo ore (¹⁶), BHP and KGHM hunting deposits together, Rio winning consent at Winu — but every one of those barrels of new metal arrives in 2030 or later. Stop paying for policy. Keep paying for scarcity.

Where the Theses Moved

TLN — down. A one-plant, one-counterparty, one-state bet just acquired a live state-regulatory variable. Susquehanna's AWS deal is already approved, so this isn't a direct hit, but Pennsylvania restricting data centers *and* opening a curtailment docket is exactly the risk a concentrated position can't diversify away.
KEEL — down. Pennsylvania is a core leg of the 2.2GW pipeline. The thesis held that spreading sites across PA, Washington and Quebec de-risks siting; this week tested that and the spread didn't help, because every jurisdiction is growing its own politics simultaneously. Still no anchor lease.
GEV — down. The shale trade group's own complaint is the tell: PA's limits threaten gas demand growth, which is the demand that underwrites turbine orders. Squeezed from both ends — fewer places to site gas-fired load, higher fuel cost when you do. Backlog is untouched; the *next* tranche of orders is the question.
WULF — up. The mirror image of the clampdown. Every moratorium raises the scarcity value of a site that is already energized and interconnected, and Lake Mariner is precisely that. Caveat, stated plainly: this is a read-through from a policy story, not a company event. Contract ramp and financing remain the real risks.
FCX — hold, but the policy leg is now worth zero. Reuters supplied the missing *reason* for the tariff stall: affordability. A tariff blocked by process comes back; one blocked by consumer prices heading into a midterm year does not. Any remaining premium has to be earned by Grasberg output, not Washington.
MP — hold, two opposing signals. Tighter rare-earth magnet rules for defense contractors are a genuine positive — that's the domestic-content mechanism the thesis needs. But Japan is pushing hard on rare-earth-*free* magnets plus non-China ore, which attacks the premise at the root: if the West engineers around neodymium, the chokepoint stops being one.
Own-the-Bottleneck — confirmed, but incomplete. Microsoft's 38GW is the strongest validation the framework has had. The gap: permission is now behaving like a scarce input with its own pricing power, and it isn't one of the four pillars. It needs to be.

Next Week — Prediction Calls

No US refined-copper tariff is announced by Friday close. ~80%. Affordability is now the stated blocker and it doesn't get better in five days. *Falsified by:* any formal Section 232 action or executive order on refined copper — which would gap FCX up more than 8% in a session and prove the policy trade isn't dead.
At least one more US state or major county announces a data-center moratorium, restriction, or clean-power permitting condition. ~75%. Four states in three months, five more flagged as primed. This is the fastest-compounding trend on the desk, and it bears directly on KEEL (negative) and WULF (positive).
LME three-month copper does not set a new closing record on any day next week. ~70%. The fear premium is out of the price; a structural deficit doesn't re-rate in five sessions. Bears on FCX and SCCO — and if I'm wrong, it means physical tightness is biting faster than the mine-supply data suggests.
At least one new hyperscaler firm-power deal of 500MW or more is announced — nuclear, gas, or fuel cell. ~65%. The run rate has been roughly one a week and Microsoft alone needs ~26GW of new procurement. Bears on CEG, VST, GEV. I expect it to be fuel-free generation, not gas.
Brent closes above $90 every day and Hormuz traffic does not return to normal volumes. ~75%. No negotiating track exists. *Falsified by:* a ceasefire or enforceable nuclear deal — which would deflate the scarcity premium in CEG and VST while relieving the fuel squeeze on BE and GEV. This is the biggest two-way risk in the book.
The Pennsylvania PUC issues no final order on data-center ratemaking or curtailment. ~90%. Dockets opened Friday take months. Anything faster would be a genuine shock and a real event for TLN and CEG.
KEEL announces no anchor AI/HPC lease. ~90%. Nothing in the pipeline suggests otherwise, and Pennsylvania just got harder.

What I'm most likely wrong about: the WULF upgrade. "Energized sites get scarcer" is a logical inference from a policy story, not evidence about the company. If WULF doesn't produce a contract or ramp milestone within two weeks, I take that move back.

Thesis Impact

The week in one line

The bottleneck moved from *things* to *permission*. Monday it was courts (Sigma Lithium halted, Nevada's federal-land data center blocked); midweek it was war (Brent >$100, both Saudi export doors under fire, coal demand revised up); Thursday it was Washington (the copper tariff quietly died and took 7% of Freeport with it); Friday it was statehouses — Pennsylvania became the fourth state in three months to restrict data-center development, and its PUC opened a docket on rates, ROE and curtailment.

The through-line: demand hardened, siting softened. Microsoft put a number on it — 38GW by 2032 against ~12GW today. That's not a forecast, it's a purchase order. Meanwhile the list of places you're allowed to plug it in keeps shrinking. Value migrates to whoever already holds energized, permitted, interconnected capacity — and away from anyone whose story is a pipeline of sites they still have to get approved.

The week's other lesson, from copper: the policy bid and the physical bid are separable. The tariff premium deflated while mine supply still heads for its first annual drop since 2017. Stop paying for policy. Keep paying for scarcity.

---

Thesis moves

TLN | Conviction: DOWN | Surprise: MED | Pennsylvania's governor just restricted data-center development in the one state this one-plant thesis lives in, and the PA PUC simultaneously opened a docket on ratemaking, ROE and curtailment of large loads. Susquehanna's AWS deal is already approved, so this isn't a direct hit — but a concentrated bet on one plant, one counterparty and one state just added a live state-regulatory variable that wasn't in the pillars. CONTRADICTS pillar 2 (clean front-of-meter transition) at the margin. | ³

KEEL | Conviction: DOWN | Surprise: MED | Pennsylvania is a core leg of the 2.2GW pipeline, and it just got harder to build there. Pillar 4 says geographic spread (PA/WA/QC) de-risks siting — the week tested that and the spread didn't help, because every jurisdiction is developing its own politics at once. Still no anchor lease. CONTRADICTS pillars 1 and 4. | ³

GEV | Conviction: DOWN | Surprise: MED | The shale trade group's complaint is the tell: PA's limits threaten gas *demand growth*, which is the same demand that underwrites turbine orders. Stack that on the week's fuel repricing — Hormuz pushing gas prices up enough that the IEA raised coal demand 1.2% — and pillar 1 ("gas turbines are the near-term firm-power answer") is squeezed from both ends: fewer places to site gas-fired load, higher fuel cost when you do. Backlog is untouched; the *next* tranche of orders is what's in question. CONTRADICTS pillar 1. | ⁴

WULF | Conviction: UP | Surprise: MED | The inverse of the state clampdown: every new moratorium raises the scarcity value of a site that is *already energized and interconnected*. Lake Mariner is exactly that asset, and four states in three months have now made it harder to replicate. CONFIRMS pillar 1 — the interconnect, not the buildout, is the moat. Note this is a read-through, not a company event; it does nothing for pillars 2 and 3 (contract ramp, financing), which remain the actual risk. | ³

FCX | Conviction: HOLD | Surprise: MED | Reuters puts a *reason* on the tariff stall — affordability — which was the missing piece. A tariff blocked by geology or process comes back; one blocked by consumer prices into a midterm year does not. The prior already marked the tariff trade down Thursday, so no further cut, but tariff optionality should now be valued near zero and any remaining FCX premium has to be earned by Grasberg output, not Washington. CONTRADICTS the policy leg; pillar 1 (structural deficit) untouched. | ¹⁷

MP | Conviction: HOLD | Surprise: MED | Two NEW items pointing opposite ways. Tightening rare-earth magnet rules for defense contractors is a genuine positive against a 0.06 prior — it's the domestic-content mechanism the DOD facility thesis needs. But Japan is going all-in on rare-earth-free magnets plus non-China overseas ore, which attacks pillars 1 and 4 at the root: if the West engineers around neodymium, the chokepoint stops being a chokepoint. They roughly cancel. The Japan substitution track is the one to watch — it's the quiet version of "China lifts export controls." | ¹⁸

Own-the-Bottleneck | Conviction: HOLD | Surprise: MED | Pillar 1 got its strongest confirmation yet (Microsoft's 38GW is demand as a signed plan, not a forecast) and pillar 2 got a clean natural experiment (the copper tariff premium vanished; the physical deficit didn't). But the framework is now missing a bottleneck: permission. Four states, a Nevada court, a Polish council, a withdrawn 120MW project in Lombardy — siting consent is behaving like a scarce input with its own pricing power, and it isn't in the four pillars. CONFIRMS the thesis, but the pillar list needs a fifth line. | ⁹

SCCO | Conviction: HOLD | Surprise: LOW | Sold off with the group on the tariff news despite never carrying the tariff premium — that's a mispricing, not a thesis change. Chinese copper rod operating rates rose this week, a small demand-side confirm. CONFIRMS pillars 1 and 2. | ¹⁹

CEG | Conviction: HOLD | Surprise: LOW | Same PA regulatory exposure as TLN (Crane restart, PJM), but 55GW across many states absorbs it, and restrictions on *new* data centers make an existing 95%-capacity-factor fleet more valuable, not less. No move. CONFIRMS pillar 1. | ⁴

LEU | Conviction: HOLD | Surprise: LOW | Fluor as EPC on the Piketon expansion is execution progress on pillar 3, but it's been on the tape two days and is already in the 0.90. CONFIRMS, no move. | ²⁰

BE | Conviction: HOLD | Surprise: LOW | Gas-price floor keeps rising on Hormuz (IEA now lifting coal demand), which is exactly the named thesis-breaker. Prior is already 0.05 — there's nowhere left to cut. CONTRADICTS pillar 3. | ¹³

No move: CLF, ETN, VRT, PWR, VST, FSLR, IREN, ISRG — touched only by recurring items already in the prior.

---

Prediction calls — week of Mon Sep 14 to Fri Sep 18, 2026

Each resolves by Friday close. Wrong = the stated condition fails.

1. No US refined-copper tariff is announced. ~80%. Affordability is now the stated blocker and it doesn't improve in five days. *Falsified by:* any formal Section 232 or executive action on refined copper. If that happens, FCX gaps up >8% in a session and I was wrong about the policy trade being dead.

2. LME copper 3-month does not close above the Sep 7 record high on any day. ~70%. The fear premium is out; the physical deficit doesn't re-rate in a week.

3. At least one more US state or major county announces new data-center restrictions, a moratorium, or a clean-power permitting condition. ~75%. Four states in three months, five more flagged as primed. This is the week's fastest-compounding trend.

4. The PA PUC issues no final order on data-center ratemaking or curtailment. ~90%. Dockets opened Friday take months. Anything faster would be a genuine surprise and a real TLN/CEG event.

5. At least one new hyperscaler firm-power deal ≥500MW is announced (nuclear, gas, or fuel cell). ~65%. The run rate has been roughly one a week and Microsoft's 38GW target needs ~26GW of new procurement.

6. KEEL announces no anchor AI/HPC lease. ~90%. Nothing in the pipeline suggests otherwise, and PA just got harder.

7. Brent stays above $90 every day and Hormuz traffic does not return to normal volumes. ~75%. No negotiating track exists. *Falsified by:* a ceasefire or an enforceable nuclear deal — which would hit CEG/VST's scarcity premium and relieve BE/GEV's fuel problem, the single biggest two-way risk in the book.

8. No second domestic HALEU producer is announced. ~95%. Free roll, but it's the LEU thesis-breaker and worth checking weekly.

What I'm most likely wrong about: the WULF upgrade. "Energized sites get scarcer" is a logical read-through from a policy story, not a company event, and read-throughs are where conviction gets padded without evidence. If WULF doesn't produce a contract or ramp milestone in the next two weeks, take that move back.

Inflection Radar

[STAGE] Rare Earth Supply Chain Buildout | US rare earth magnet facility breaking ground in SC marks tangible progress toward de-risking critical materials supply chains. | Touches: N/A (NEW) | ²¹

[STAGE] Regulatory Friction in Power | Court rejecting DOE 'emergency' orders signals continued legal challenge to federal utility overreach and market stability. | Touches: N/A (NEW) | ²²

[STAGE] AI Evidence Rigor | Critique of LLM training data reliance (detectable duplication) forces the field to prioritize verifiable evidence over generalized claims. | Touches: N/A (NEW) | ²³

[STAGE] Wheeled-Humanoid Robotics | Identifying a technical gap in retargeting human motion for wheeled-humanoids suggests the next frontier of autonomous physical deployment. | Touches: N/A (NEW) | ²⁴

[STAGE] Local Compute Procurement | Localized spending spikes (Piqua bonuses) following data center procurement signal immediate, ground-level demand for compute infrastructure. | Touches: N/A (NEW) | ²⁵

***

Synthesis & Prediction Calls:

The macro through-line is a shift from theoretical capability to localized, verifiable implementation. Power and materials are de-risking via physical buildout (SC rare earth, Indiana microreactor), while AI research is undergoing a necessary methodological correction, moving away from black-box claims toward auditable controls (GEOSTEER) and evidence requirements ²³.

Prediction Calls for Next Week:

1. Policy/Power: Expect continued regulatory challenges to centralized power structures; the friction seen in ²² will likely repeat or escalate, favoring decentralized generation models.

2. AI Compute: The market narrative will pivot sharply from "what LLMs can do" to "how we prove they work." Any company making generalized claims without citing specific architectural controls (like activation steering) will face immediate skepticism.

3. Physical Infrastructure: Cross-domain pressure points will emerge where compute meets physical assets: Expect increased focus on AI optimization for logistics, energy grid management, or autonomous industrial deployment (e.g., linking ²⁵ procurement to ²⁴ robotics).

QA & Caveats

The WULF upgrade is a read-through; if it doesn't produce a contract or ramp milestone soon, the conviction should be reduced.
The "Own-the-Bottleneck" call is an analytical framework observation, not a direct prediction supported by any single source.
Prediction calls 1, 2, and 7 rely on geopolitical stability and policy inaction, which are external variables, not direct thesis confirmations.

Sources

bloomberg.com bloomberg.com
The Solar Farm-to-Data Center Switcheroo Hits a Snag heatmap.news
Pennsylvania Data Center Limits Are a Threat to Gas Firms, Trade Group Says bloomberg.com
Pennsylvania PUC to consider ratemaking, ROE and curtailment as data centers grow utilitydive.com
Massachusetts hits data centers with new clean power rules techcrunch.com
Plans for 120MW data center withdrawn in Lombardy, Italy datacenterdynamics.com
These 5 States Are Primed for a Data Center Pause heatmap.news
Amid data center public backlash, Yvette Eden Ruiz joins OpenAI’s Community Engagement team from JPMorgan datacenterdynamics.com
Microsoft targets 38GW of data center capacity in 2032 - report datacenterdynamics.com
datacenterdynamics.com datacenterdynamics.com
bloomberg.com bloomberg.com
twz.com twz.com
Coal demand to rise 1.2% as Hormuz closure drives gas prices: IEA mining.com
bloomberg.com bloomberg.com
Mining stocks rally comes to abrupt halt as copper, silver prices plummet and gold slides mining.com
bloomberg.com bloomberg.com
White House copper tariff plan stalls amid affordability concerns, sources say - Reuters news.google.com
“Rare-Earth-Free Magnets, Raw Materials From Overseas Mines”: Japan Goes All Out to Build an Independent Rare Earth Supp news.google.com
[SMM Copper Cathode Rod Flash] The operating rate rose this week mainl - SMM Metal news.google.com
Fluor to serve as EPC contractor for Centrus’s Piketon plant expansion - American Nuclear Society news.google.com
USA Rare Earth breaks ground at $1.2B rare earth magnet manufacturing facility in South Carolina mining.com
Court rejects DOE ‘emergency’ order delaying coal plant retirement as overstep utilitydive.com
Detectable Only Where It Is Confounded: What Verified Duplication Counts Say About Membership Evidence in Language Model arxiv.org
Morphology-Aware Human Motion Retargeting for Wheeled-Humanoid Loco-Manipulation arxiv.org
One-time bonuses paid to 11 Piqua city employees after data center procurement - Miami Valley Today news.google.com