Weekly Macro & Outlook — 2026-05-29

The Week in Macro

The defining story of the week was a quiet but decisive move in *where* the AI build-out is stuck. Demand stopped being the question — Tuesday's Mag-7 earnings locked in hyperscaler capital spending of roughly $725B for 2026 (¹). With demand settled, the binding constraint shifted to delivery: who connects the load to power, and who pays for it. That's a more interesting place to invest, because it's where the money actually changes hands.

The clearest evidence was the rise of "behind-the-meter" power — generating electricity right at the data center and skipping the multi-year wait to plug into the public grid. T5's plan to run a Dulles, Virginia site on 60MW of fuel cells as *primary* power (²), and today UK Reabold's pitch for a 100MW off-grid gas data center in Yorkshire (³), show on-site generation going from emergency workaround to default plan. The catch worth holding onto: this week's new proof points were *gas*, not fuel cells — a competing path, not clean validation of the fuel-cell story.

Regulators pushed the same direction from the other side. Oregon approved a tariff shifting connection and infrastructure costs onto hyperscale customers (⁴), and Pennsylvania set a standard making data-center developers pay entirely for their own power (⁵). And the friction turned political: Ohio paused data-center tax credits pending review (⁶), and the backlash began splintering the American right over land and energy use (⁷). The message: the next leg of risk is social and regulatory, not physical.

There was one genuine crack in the central thesis. NERC, the U.S. grid-reliability body, *cut* its Texas peak-demand forecast by 3.7 gigawatts because data centers can now be switched off during emergencies (⁸). If hyperscalers will accept being curtailed, the "hard grid ceiling" softens — which trims the scarcity premium that grid-bypass plays depend on. Pair that with the recurring "AI capex fatigue" chatter and TSMC's claim that energy-*efficient* compute matters most to customers (⁹), and you have the seeds of a demand-flattening story. Soft so far, but it's the one thread that could break half the book at once.

On materials and geopolitics, the week leaned toward de-escalation. A tentative US-Iran truce extension sent oil lower (WTI −1.1% to $87.60) and put tankers back through the Strait of Hormuz (¹⁰). That cuts both ways: reopening transit relieves the sulphuric-acid cost crunch squeezing copper-miner margins (¹¹), but it also deflates the aluminum squeeze (¹²) and the war premium baked into metals. Mixed for miners, mildly positive on their cost lines.

Where the Theses Moved

• MP — conviction DOWN. The week's cleanest disconfirm. Two new Western magnet-supply deals landed (Ionic Rare Earths' binding oxide sales to Advanced Magnet Lab; Defense Metals' MOU with Hanwha) while REMX fell −1.5% today. Price moved *opposite* to a posterior that had jumped to 0.79 on a thin tech-theft story — the "only integrated Western producer" pillar is eroding at the margin. ¹³
• BE — conviction HOLD, but the thesis got more complicated. Behind-the-meter-as-default hardened all week, which is the heart of the BE case. But every new on-site proof point was gas, not fuel cells — so the trend confirms the *category* while leaving a competing path wide open. Already priced into the 0.68 prior. ³
• CEG / TLN — HOLD, now in live tension. The same behind-the-meter wave that helps BE quietly questions the co-location scarcity premium these names rely on. The wrinkle: CEG's conviction actually *rose* even as the trend cuts against it — so this is now a direct, testable disagreement rather than a vague worry. ¹⁴
• The equipment layer (ETN, VRT, PWR, GEV) — re-rated up, now largely priced. Convictions jumped earlier in the week on the "equipment wins regardless of who generates the power" logic (ETN 0.79 → 0.89). Most of this week's flow is now recurring confirmation, not new edge.
• GEV — HOLD. Entergy's gas builds make up a third of MISO's fast-track interconnection queue, ~70% serving data centers — confirming gas as the near-term firm-power answer. Partly offset by a grid-firming tender where big batteries beat gas. Net neutral. ¹⁵
• FCX / SCCO — HOLD. Hormuz de-escalation should ease the acid-cost crunch on copper output (mild margin positive), but copper was flat (−0.6%) and Panama's First Quantum audit is a live supply-return risk. No move.

Next Week — Prediction Calls

• Panama's First Quantum / Cobre Panama audit does NOT trigger a near-term restart. The audit was due today amid renewed anti-mining protests; expect political noise but no production timeline. Copper-deficit thesis holds and COPX stays above ~$85. *Wrong if a concrete restart timeline is announced.* — bears on FCX/SCCO.
• The Hormuz truce extension holds and WTI drifts into the low $80s. Transit stays open, the sulphuric-acid cost pressure on lithium/copper/nickel eases, and the aluminum squeeze unwinds further. *Wrong if strikes resume or WTI breaks back above $92.* — bears on FCX, materials cost lines.
• No scaled Western rare-earth-magnet competitor emerges — the MOUs stay MOUs. Expect more announcements, none funded or at production scale, and REMX keeps lagging the MP narrative. *Wrong if a funded, at-scale, integrated Western producer is named.* — bears on MP.
• At least one more behind-the-meter / on-site data-center power announcement lands, and it's gas-turbine, not fuel-cell. That keeps the BE-vs-gas competition open rather than resolving it in BE's favor. *Wrong if a named BE primary-power data-center contract is signed — that would be a BE re-rate.* — bears on BE.
• A new hyperscaler nuclear PPA or reactor-uprate milestone surfaces for VST or CEG, with Henry Hub gas stable near $3.2–3.3. The co-location names defend their premium with a fresh signed deal even as behind-the-meter pressure builds. *Wrong if a previously signed PPA slips or cancels.* — bears on CEG/VST, tests the CEG-vs-BE tension.
• The "AI capex fatigue" thread gets louder but does NOT move a hyperscaler to cut guidance. Hedge-fund skepticism and the TSMC efficiency line keep circulating, but no Mag-7 name walks back 2026 spending. *Wrong if any hyperscaler trims capex guidance* — that's the demand-flattening trigger that would hit the whole equipment layer at once.

Thesis Impact

The week in one read

The macro through-line all week was a shift in where the AI-power bottleneck lives — from "can we generate it" to "how do we connect it, and who pays." Demand stopped being the question (Mag-7 capex ~$725B/2026 confirmed Tuesday). The action moved to delivery: hyperscalers being forced to become their own power companies, regulators (Oregon, Pennsylvania) shoving interconnection costs onto data centers, and developers routing around the grid entirely with on-site generation. T5's 60MW fuel-cell-as-*primary*-power plan, and today UK Reabold's off-grid gas data center, are the cleanest proof points. Behind-the-meter went from workaround to default plan.

That helps BE (sell the on-site box) but quietly nibbles the co-location scarcity premium under CEG/TLN — the tension is now live and testable, since CEG's conviction rose even as the trend cuts against it. Note the wrinkle: this week's new on-site proof points are *gas*, not fuel cells — a competing path, not pure BE validation.

The equipment layer (ETN, VRT, PWR, GEV) is confirmed and largely priced — convictions already re-rated up earlier in the week, so most of this week's flow is recurring. The one lurking disconfirm worth respecting: the recurring "AI capex fatigue" thread and TSMC's "energy-efficient compute matters most" line, both of which feed the demand-flattening trigger that breaks half the book at once. Soft so far, but watch it.

Geopolitics flipped toward de-escalation late week: a tentative US-Iran truce extension, oil softening (WTI −1.1% to $87.60), and tankers crossing Hormuz. If transit reopens it relieves the sulphuric-acid cost crunch squeezing copper-miner margins — but it also deflates the aluminum squeeze and the war-premium in materials. Mixed for the miners, mildly positive for cost lines.

The cleanest disconfirming signal of the week was in rare earths. Two new Western magnet-supply deals landed while REMX fell — price moving opposite to a posterior that had jumped on a soft story.

Prediction calls — next week (falsifiable)

1. First Quantum / Cobre Panama audit (due today) does NOT trigger a near-term restart. Copper deficit thesis holds; COPX stays above ~$85. *Wrong if a restart timeline is announced.*

2. Hormuz truce extension holds → WTI drifts into the low-$80s and sulphuric-acid/critical-mineral cost pressure eases. *Wrong if strikes resume or WTI breaks back above $92.*

3. No scaled Western rare-earth-magnet competitor emerges — MOUs stay MOUs, and REMX keeps lagging the MP narrative. *Wrong if a funded, at-scale integrated Western producer is announced.*

4. At least one more behind-the-meter / on-site data-center power announcement, and it's more likely gas-turbine than fuel-cell — keeping the BE-vs-gas competition open. *Wrong if a named BE primary-power DC contract lands (that would be a BE re-rate).*

5. A new hyperscaler nuclear PPA or uprate milestone for VST/CEG, with nat gas stable near $3.2. *Wrong if a signed PPA slips/cancels.*

Thesis lines

• MP | Conviction: DOWN | Surprise: MED | Two NEW Western magnet-supply deals (Ionic Rare Earths–AML binding oxide sales; Defense Metals–Hanwha MOU) erode the "only integrated Western producer" pillar at the margin, and REMX −1.5% today diverges from a posterior that jumped to 0.79 on a soft tech-theft story — CONTRADICTS the scarcity-premium pillar. ¹³
• BE | Conviction: HOLD | Surprise: LOW | Behind-the-meter-as-default hardened all week and today's off-grid 100MW gas DC (Reabold) extends it — CONFIRMS grid-bypass, but already in the 0.68 prior and the new proof points are gas, not fuel cells (a competing on-site path). ³
• GEV | Conviction: HOLD | Surprise: MED | NEW: Entergy's gas builds are a third of MISO fast-track interconnects, ~70% serving data centers — CONFIRMS gas as the near-term firm-power answer; partly offset by a firming tender where batteries beat gas. Net neutral on 0.66. ¹⁵
• CEG | Conviction: HOLD | Surprise: LOW | The behind-the-meter wave keeps questioning the co-location scarcity premium, but it's recurring and already weighed; no NEW negative, nat gas steady at $3.28 — thesis intact, tension noted. ¹⁴
• FCX / SCCO | Conviction: HOLD | Surprise: LOW | NEW Hormuz de-escalation (oil −1.1%, truce extension) should ease the sulphuric-acid cost crunch on copper output — mild CONFIRM of margins; but copper flat (−0.6%) and the First Quantum Panama audit is a supply-return risk. No move. ¹⁰
• ISRG | Conviction: HOLD | Surprise: LOW | Tesla's robotaxi fleet shrank to ~20 cars and humanoid names keep stumbling — relative CONFIRM that ISRG is the only proven scaled robotics model, but T3/recurring and the 8-K is just a board change. ¹⁶
• FSLR | Conviction: HOLD | Surprise: LOW | Solar+batteries "doing great" and batteries beating gas in a firming tender — mild CONFIRM of the cheap-fast-power pillar, but T3/recurring and already in the 0.67 prior. ¹⁷

Inflection Radar

[emergent] AI-Driven Resource Security | The convergence of AI agents optimizing critical minerals recovery (94) and the geopolitical focus on securing supply chains (71, 100) suggests a pivot where compute capability is directly applied to solving physical resource scarcity, bypassing traditional mining bottlenecks. | Touches: NEW | ¹⁸

[emergent] Decoupling in Defense Procurement | Localized security concerns (Ontario banning Chinese drones, 100) and geopolitical friction (Taiwan/China, 71) are forcing immediate, visible policy shifts toward domestic or allied supply chains, creating immediate, non-negotiable procurement mandates. | Touches: NEW | ¹⁹

[emergent] Memory Architecture as AI Bottleneck | The market signal is shifting from pure compute scale (Nvidia hype) to fundamental architectural limits. Startups are betting that memory bandwidth and efficiency (76) will dictate the next wave of AI performance gains, making memory/interconnect IP a higher-leverage target than raw FLOPS. | Touches: NEW | ²⁰

[emergent] Strategic Resource Control Focus | The sustained focus on Iran's stockpile (77, 78, 90) alongside major energy regulatory filings (85-89) indicates that the immediate geopolitical risk premium is not just on *use*, but on *control* and *removal* of strategic, finite assets (Uranium, energy infrastructure). | Touches: NEW | ²¹

Prediction Calls for Next Week:

1. AI Compute: Expect increased focus on memory/interconnect layer investment announcements over pure model size increases.

2. Geopolitics/Defense: Watch for policy statements or regulatory filings that explicitly link domestic/allied procurement mandates to AI/ML optimization tools (e.g., using AI to vet supply chain compliance).

3. Energy/Materials: Increased regulatory scrutiny (FERC/DOE) will likely focus on the *digital* aspect of energy transmission planning, not just the physical assets.

QA & Caveats

• Call 1 is supported by the context of the Panama audit and copper supply, but the specific outcome (no restart) is a prediction, not a direct citation.
• Call 2 is supported by the context of the oil softening and the Hormuz situation.
• Call 3 is supported by the existence of rare earth supply deals, but the absence of a *scaled competitor* is a forward-looking claim that cannot be confirmed by current MOUs alone.
• Call 4 is supported by the BE thesis, but the shift to gas is presented as a competing path rather than a confirmed outcome.
• Call 5 lacks a specific, direct source confirming a milestone for VST/CEG.

Sources

1. news.google.com news.google.com
2. datacenterdynamics.com datacenterdynamics.com
3. UK's Reabold Resources seeks partner for 100MW off-grid gas-powered data center in Yorkshire datacenterdynamics.com
4. Oregon PUC approves PGE’s large-load tariff framework for data centers utilitydive.com
5. datacenterdynamics.com datacenterdynamics.com
6. bloomberg.com bloomberg.com
7. Data Centers Are Splintering the American Right heatmap.news
8. utilitydive.com utilitydive.com
9. Energy efficient compute is most important attribute for customers, TSMC claims datacenterdynamics.com
10. Oil Prices Soften as Traders Bet on Renewal of US-Iran Truce bloomberg.com
11. CHARTS: How the sulphuric acid crunch is driving up critical minerals costs northernminer.com
12. Aluminum Squeeze Deepens as Spreads Tighten and Inventories Drop bloomberg.com
13. Ionic Rare Earths, AML partner up to support US permanent magnet supply chain mining.com
14. Hyperscalers didn’t set out to be power companies. The grid left them no choice. utilitydive.com
15. Entergy’s gas projects are one-third of MISO’s fast-track interconnection process utilitydive.com
16. Tesla robotaxi fleet shrinks to just 20 unsupervised cars reddit.com
17. Robotaxi has the wrong kind of momentum, but wind, and solar are doing GREAT electrek.co
18. US deploys AI agents to speed critical minerals recovery northernminer.com
19. New security rules ground Chinese-made drones in Ontario suasnews.com
20. This chip startup just raised $135M on a bet that AI’s biggest bottleneck isn’t compute — it’s memory techcrunch.com
21. What it would take to destroy or remove Iran’s uranium stockpile - Foundation for Defense of Democracies news.google.com