Weekly Macro & Outlook — 2026-10-16

Dek

America ran out of patience for AI data centers before it ran out of power, so this week the buildout started making its own electricity and moving to states that still say yes.

The Week in Macro

The biggest change this week: the main limit on AI data centers is now permission, not electricity. Oregon froze air and water permits for data centers until July 2027 (¹). A Virginia state senator proposed a three-year statewide pause in the country's biggest data-center market (²). Memphis delayed a vote on its own 12-month ban (³). California regulators put off a vote on a big Google site for the fourth month running (⁴). Heatmap counts more than $260 billion in cancelled projects this year (⁵). This matters because a governor can now remove demand, not just delay supply. Growth will flow to Texas, the Southeast and the Plains, and anything already approved and already connected is worth more.

The industry's answer is to skip the line. Google signed a 20-year deal for a 702-megawatt fuel-cell plant in Wyoming. Utility AEP will spend about $4 billion building it, and AEP also ordered up to 1.6 GW of Mainspring linear generators, compact gas-burning units that make power on site (⁶). The second way to skip the line is squeezing more out of existing nuclear plants, called "uprates." Google is paying for uprates at 11 Constellation reactors (⁷). The Energy Department reportedly offered Vistra up to $4.2 billion for 433 MW of uprates (⁸). The point: the fastest new megawatts come from places that already have permits.

Demand itself isn't slowing. PJM (the Mid-Atlantic grid), ERCOT (Texas) and SWPP (the central Plains) all set peak-load records this summer (⁹). PJM agreed to study SoftBank's roughly 10 GW Ohio project (⁶). Britain had to borrow up to 1,200 MW of emergency power from France on an ordinary October evening (¹⁰). The money question got louder, though. Gartner called AI data-center spending "unsustainable" and warned of a possible $3.5 trillion write-down (¹¹). AI stocks sold off alongside a hot US inflation report.

In materials, copper broke our shortage story, but only partly. The International Copper Study Group expects the world's surplus of refined copper to more than triple next year as new smelters (plants that turn ore into pure metal) come online (¹²). Mines still look tight. Antofagasta fell more than 6% after its 2027 guidance missed (¹³). Tech companies are courting Anglo American for a stake in its mine supply, and Beijing still holds the key to Anglo's Teck merger (¹⁴). Washington keeps stockpiling: Mercuria delivered 12,000 tons of Congolese copper to the US Project Vault reserve (¹⁵). Metals bounced Friday (¹⁶). The takeaway: copper is plentiful at the refining step and scarce at the mine. Metal is not what's holding AI back. Permission is.

Where the Theses Moved

• VST — UP. The reported $4.2B Energy Department offer for uprates fits Vistra's October 5 filing of a new major agreement and new debt. That turns uprates from the weakest part of the thesis into a confirmed one. Power that is already permitted and connected is this week's clear winner.
• SCCO — DOWN (trimmed, not cut). The refined-copper surplus forecast contradicts our "structural deficit" pillar. SCCO's position as the lowest-cost producer still protects its margins, which is why we trimmed rather than cut.
• Own-the-Bottleneck — DOWN. Copper breaks our "processors beat miners" rule: the glut is at the smelter and the scarcity is at the mine. The bigger idea holds up, but the bottleneck worth owning this week was permission, not processing.
• BE — HOLD. The 702 MW Google–AEP plant shows on-site power demand is real. But AEP's 1.6 GW Mainspring order is the first utility-scale sign of a cheaper rival, the trigger we flagged as a risk.
• GEV — HOLD, slight negative. A major utility chose fuel cells and linear generators over gas turbines for fast power. It's one data point, but it chips at the "turbines are the near-term answer" pillar.
• VRT / FCX — HOLD. Gartner's warning and the AI selloff weigh on sentiment, but no hyperscaler has cut spending. FCX simply follows the copper price, and its conviction was already cut to 0.49.

Next Week -- Prediction Calls

• Copper holds $10,000. LME copper (the London Metal Exchange benchmark) closes above $10,000/t every day through Oct 23, because mine-side tightness outweighs the smelter surplus. A close below it is a miss, and SCCO goes DOWN again. Bears on SCCO, FCX.
• Another US state or major county proposes a data-center pause. By Oct 23, at least one more state legislator, governor or large county formally proposes a moratorium or permit freeze, beyond Oregon, Virginia and Memphis. Bears on greenfield developers and supports already-connected power (VST, CEG).
• No hyperscaler cuts spending. Despite Gartner and the selloff, no hyperscaler (Microsoft, Alphabet, Amazon, Meta, Oracle) announces a cut to data-center capital spending next week. Any guidance that does come out holds or rises. Bears on VRT, ETN.
• Another "bring your own power" deal over 100 MW, and not from Bloom. A utility or hyperscaler announces a new on-site power deal over 100 MW next week, and the supplier is gas engines, turbines or linear generators rather than solid-oxide fuel cells. If it happens, BE moves DOWN. A turbine win would steady GEV.
• Another nuclear uprate or life-extension deal. A new uprate, restart or license-extension deal is announced next week, either Energy Department financing or a hyperscaler contract. Bears on VST, CEG.

Thesis Impact

• SCCO | Conviction: DOWN | Surprise: MED | The International Copper Study Group now expects the world's surplus of refined copper to more than triple next year. That CONTRADICTS pillar 1, the structural deficit. The prior of 0.78 still assumes a shortage, unlike FCX at 0.49. SCCO's lowest-cost margin floor still holds, so trim conviction rather than cut it. Call: LME copper (London Metal Exchange benchmark) stays above $10,000/t through Oct 23. A close below it means DOWN again. | ¹²
• FCX | Conviction: HOLD | Surprise: LOW | Metals bounced Friday after a week hit by a hot US inflation print and an AI-stock selloff (Bloomberg, T1). Mines are still tight: Antofagasta's 2027 output guidance missed. The surplus is in refined metal from smelters (plants that turn ore into pure copper), not in ore. Mixed news, already reflected in the cut prior of 0.49. Call: no new Grasberg disruption next week, and FCX tracks the copper price. | ¹⁶
• Own-the-Bottleneck | Conviction: DOWN | Surprise: MED | This is the week's sharpest point against us. For copper, the glut is at the processing step and the scarcity is at the mine. Antofagasta's guidance missed, and tech companies are courting Anglo American for mine supply. That CONTRADICTS pillar 2, "processors beat miners / refining not mining." Copper is the counterexample. Our weekly through-line also confirms that permission now binds harder than metal. Call: next week brings more mine-side guidance cuts, not smelter tightness. | ¹³
• BE | Conviction: HOLD | Surprise: LOW | AEP will spend about $4B on the 702 MW Bloom fuel-cell plant for Google. That CONFIRMS demand for on-site power. But AEP's separate order for up to 1.6 GW of Mainspring linear generators is a competing on-site technology at utility scale. That is the "cheaper alternative" trigger, now in early form. Both were already reflected after Wednesday and Thursday. Call: the next utility or hyperscaler on-site order over 100 MW goes to a non-SOFC technology (not a solid-oxide fuel cell). If it does, BE moves DOWN. | ⁶
• GEV | Conviction: HOLD | Surprise: LOW | The same AEP deal shows a big utility choosing fuel cells and linear generators, not turbines, for fast on-site power. That mildly CONTRADICTS pillar 1, "turbines are the near-term firm-power answer." It's one data point and was already seen on Thursday. Call: no turbine order cancellations reported next week. | ⁶
• VRT | Conviction: HOLD | Surprise: LOW | Gartner called AI data-center spending "unsustainable" and warned of a possible $3.5T write-down. Bloomberg (T1) confirms AI stocks sold off this week. That leans against pillar 2 (orders track hyperscaler capex), but no hyperscaler has actually cut spending. The Nscale fire and cooling leak in Norway is a single-site event. The prior of 0.39 already reflects this. Call: no hyperscaler announces a capex cut next week. | ¹¹

Inflection Radar

Synthesis of the Week

The macro through-line is Resilience via Digitization. Across power, materials, and defense, the primary focus is on securing critical physical assets—whether that's copper supply chains, reliable grid infrastructure, or operational airspace. The underlying technological enabler for this resilience is a maturation in AI/Compute: the industry is pivoting away from raw model size toward extreme efficiency (inference cost, quantization) to make advanced capabilities deployable at the edge.

• Power & Materials: Regulatory bodies (FERC) are attempting to streamline physical bottlenecks (LNG, pipelines), while geopolitical pressure keeps critical mineral procurement (Copper/Congo) a top-tier policy concern.
• Defense & Drones: The sector is rapidly moving past simple drone deployment toward integrated, networked systems that require robust, encrypted connectivity and advanced C-UAS capabilities.
• Compute & AI: The academic focus has crystallized on solving the *cost* of intelligence—optimizing inference (70, 71) and developing novel learning paradigms for physical systems (Robotics/AI).

Prediction Calls for Next Week

1. Infrastructure Convergence: Expect increased discussion linking advanced compute efficiency (e.g., low-bit KV caching) directly to grid modernization or distributed energy resource management, treating the power grid as a computational problem rather than purely an electrical one.

2. AI in Hard Assets: The focus on autonomous robotics and drone integration will accelerate. Look for signals connecting AI learning from passive data (74) with physical defense applications (C-UAS/Drones).

3. Geopolitical Commodity Focus: Continued emphasis on the *process* of securing materials (e.g., tax compliance, specific origin tracking) rather than just the commodity price itself.

***

[emergent] AI-Enabled Edge Resilience | The convergence of advanced compute efficiency (low-bit caching, optimized inference) with critical infrastructure/defense applications suggests a shift toward making complex AI models run reliably on low-power, distributed hardware at the operational edge. | Touches: 03 Drones & Defense, 02 Energy & Power, 01 AI & Compute | ¹⁷

[emergent] Autonomous Systems Integration | The combination of advanced human-machine interfaces (voice command, 67) and robust secure networking (ZeroTier, 54) indicates that future military/industrial autonomy will be defined by seamless, reliable remote control and integration across platforms. | Touches: 03 Drones & Defense, 04 Robotics & Autonomous | ¹⁸

[emergent] Regulatory Streamlining for Electrification | FERC's proactive proposals to streamline gas and LNG reviews (79) signal a broader regulatory willingness to accelerate energy infrastructure buildout, which will inevitably intersect with the necessary grid upgrades required by electric vehicle/industrial adoption. | Touches: 02 Energy & Power | ¹⁹

[dismissive] Legacy Drone Acquisition Timelines | The shift in Germany's timeline (59) and the general trend of evolving C-UAS requirements (58) suggest that major, multi-year acquisitions for specific platforms are facing political or technical delays, favoring modular, rapidly deployable solutions instead. | Touches: 03 Drones & Defense | ²⁰

QA & Caveats

• SCCO: The call regarding the LME price trajectory is not directly supported by the source, which focuses on refined copper surplus structure rather than specific price forecasts.
• FCX: The claim about no new Grasberg disruption next week is an unsupported specific event prediction.
• Own-the-Bottleneck: This call is reasonably supported by the context provided regarding Antofagasta's missed guidance, suggesting a shift in constraint from smelters to miners.
• BE: The trigger condition relies on future hyperscaler actions, which is not directly confirmed by the source detailing past technology choices.
• GEV: The claim about no turbine order cancellations next week is an unsupported specific event prediction.
• VRT: The call regarding no hyperscaler capex cuts next week is an unsupported future prediction.

Sources

1. rtoinsider.com rtoinsider.com
2. datacenterdynamics.com datacenterdynamics.com
3. datacenterdynamics.com datacenterdynamics.com
4. rtoinsider.com rtoinsider.com
5. heatmap.news heatmap.news
6. AEP in deal to buy up to 1.6 GW of Mainspring linear generators utilitydive.com
7. heatmap.news heatmap.news
8. news.google.com news.google.com
9. PJM, ERCOT, and SWPP all set peak load records this summer eia.gov
10. bloomberg.com bloomberg.com
11. Gartner warns AI data center capex is ‘unsustainable’ as $3.5T write-down risk looms - Fierce Network news.google.com
12. Copper surplus to widen as refined output rebounds northernminer.com
13. Antofagasta shares sink as 2027 copper outlook falls flat mining.com
14. Tech giants come knocking for a piece of Anglo American as copper scarcity bites, Wanblad tells FT mining.com
15. Mercuria delivers 12,000 tons of DRC copper to US Project Vault - Mining.com news.google.com
16. Metals Rebound as Sentiment Improves After Volatile Week bloomberg.com
17. RoLaQ: Joint Rotation and Range Calibration for 2-Bit KV Caches in Multi-Head Latent Attention arxiv.org
18. New human-machine interface will help Boxer punch above its weight defensenews.com
19. FERC proposes easing regulatory pathway for gas pipeline, LNG projects utilitydive.com
20. Germany changes course on plans to buy MQ-28 drone wingman: Boeing official breakingdefense.com