Weekly Macro & Outlook — 2026-06-05

Dek

The week AI's bottleneck slid down the stack — from chips nobody can make fast enough, to power nobody can deliver fast enough, to money nobody's sure how to raise — and then the whole hard-asset complex sold off in one synchronized heap, gold included, which is the tell that this was a margin call, not a verdict on demand.

The Week in Macro

The story this week was a single moving target: what, exactly, is stopping the AI build-out. It started as silicon. By Wednesday it was power. By Friday it was money. Demand for compute never wavered — AirTrunk committed $30B to build 5 gigawatts of data centers in India, proof the race is now global, not just American (¹). What changed is everyone's understanding of the binding constraint, and that shift reprices who wins.

Power became the loud, repeated theme. The grid is now the chokepoint, not the chip. An expert warned gigawatt-scale build-outs could trigger rolling blackouts; Google paired a Texas data center with its own 1GW power project; Meta started pitching literal tents over servers to cut costs; a 10,000-acre campus got scrapped in New Mexico. The pattern — "build your own power, site where the electrons are" — strengthened every single day. That's a structural demand floor under firm baseload (nuclear, gas) and the electrical-equipment block (VRT, ETN, PWR, GEV) that wires it all up. Why it matters: when power is the scarce thing, whoever owns committed megawatts and committed switchgear wins, because those clear before the money gets tight.

Then the money got tight — or at least the worry about money did. Early in the week, Wall Street started pricing the build-out two ways at once: traders floated GPU "futures" so hyperscalers could hedge compute costs, while credit-default-swap jitters (the cost of insuring hyperscaler bonds against default) ticked up for the first time. Goldman put a number on it — $5.3 trillion in hyperscaler capex — and said private infrastructure capital will have to carry a growing share (²). The question moved from "will they build" to "how do they fund it."

That financing anxiety detonated Friday. The entire hard-asset complex liquidated together: copper −4%, uranium −10%, rare earth −9%, steel −4% — and gold −3.5%. Gold is the tell. When the safe-haven falls alongside everything else, you're watching forced selling and deleveraging, not investors deciding the demand story is dead. The physical story is intact — Goldman actually *cut* its copper supply forecast 350kt this week, widening the deficit (³). The flows are just ugly.

Geopolitics stayed a simmering tail risk. Iran's Strait of Hormuz toll regime hardened into permanent policy, with traffic near zero and peace talks stalled (⁴) — yet WTI drifted *down* to $90 on the week, because the market has stopped pricing reopening and started pricing around the toll. A permanent tax on shipped fuel is a quiet, standing subsidy for anything that makes power on-site. That feeds straight back into the power thread.

Where the Theses Moved

• MP (rare earth) — conviction DOWN. The moat broke. A *second* federally-funded Western integrated player firmed up: ElementUSA/Colorado School of Mines won a $67M DOE processing award, stacking on USA Rare Earth's $1.6B Commerce package earlier in the week. The "only integrated Western supplier" pillar is now a Washington-subsidized multi-horse race. REMX −8.6% Friday sided with the bears; the MP posterior collapsed from 0.64 to 0.47.
• FCX (copper) — conviction DOWN. Two hits to the same pillar. Indonesia advanced export controls Monday (a named thesis-breaker), and Friday Goldman cut copper supply *citing Grasberg's slow recovery by name*. Bullish for the copper price, bearish for FCX's own output. The deficit widens; FCX may not be the one to harvest it.
• SCCO (copper) — conviction HOLD. Same Goldman cut, opposite read: it confirms the structural deficit, and SCCO carries none of FCX's Grasberg/Indonesia baggage. Friday's COPX −9.9% is the everything-down flow, not a Peru/Mexico problem.
• CEG / VST (nuclear/power) — conviction HOLD, thread strengthening. The week's dominant story — power as the global AI constraint — is a structural floor under firm baseload. The uranium crash (URA −10%) is flow inside a sell-everything tape, not a demand crack; no PPA news cut against the thesis.
• LEU (enrichment) — conviction HOLD. URA −10.3% / URNM −9.5% looks alarming but landed with zero new HALEU/SWU fundamentals and gold falling too. The Russian-SWU-ban + sole-US-enricher case is untouched. Watch, don't move.
• ETN / VRT / PWR / GEV (equipment) — conviction HOLD, watch the disconfirmer. Demand kept confirming (continental data-center starts all week, another VRT 8-K). But a national data-center moratorium is gaining poll momentum, with local backlash, police controversy, and scrapped campuses piling up. Near-term floor intact; permitting is now a real medium-term tail risk.

Next Week — Prediction Calls

• The selloff bounces. If Friday was flow not demand — and the synchronized drop with gold says it was — COPX, URNM, and XME each close ≥4% above Friday's level by next Friday. *Wrong if* any makes a fresh lower low and holds it, which would mean real demand is selling.
• A demand-floor deal prints. At least one new hyperscaler PPA or behind-the-meter power deal (CEG/VST/BE/GEV) gets announced. The power-is-the-constraint thread is too hot to go a full week quiet. *Wrong if* the wire is silent on new generation deals.
• The backlash gets teeth. The data-center moratorium thread produces at least one concrete action — a state or local moratorium vote, or a permitting denial — not just op-eds and polls. *Wrong if* it stays talk. This is the first real medium-term threat to the equipment block.
• MP keeps fading. REMX underperforms XME next week with no MP-positive catalyst (offtake win, China re-tightening export controls). *Wrong if* China reinstates or tightens rare-earth export controls — the one event that re-rates MP overnight.
• FCX bleeds versus SCCO. If Indonesia formalizes its June export rules, FCX underperforms SCCO by ≥2 points. *Wrong if* Indonesia carves out Grasberg or delays the rules.
• The financing crack widens, the demand story doesn't. Hyperscaler CDS and AI-capex funding worries get *louder* next week — more two-sided coverage — even as physical deals keep landing. *Wrong if* credit chatter fully reverses and metals plus equities rally together on demand alone.

Thesis Impact

The week in one line

The binding constraint on AI migrated down the stack all week — silicon → power → money. Demand never cracked (AirTrunk's $30B/5GW India commit, Goldman's $5.3T capex math, a wall of new data-center starts). What cracked is the *financing layer*: GPU futures and hyperscaler CDS jitters appeared Monday, and by Friday the whole hard-asset complex was in a synchronized liquidation — copper −4%, uranium −10%, rare earth −9%, steel −4%, and gold −3.5%. Gold falling with everything else is the tell: this is deleveraging/margin-driven selling, not a demand verdict. The physical story (deficits widening, power as the new oil) is intact; the flows are ugly. Two structural threads also resolved this week — behind-the-meter power as the default AI build (strengthening) and the rare-earth moat breaking bearishly (faded).

Theses the week's signal materially touched

• MP | Conviction: DOWN | Surprise: MED | A *second* federally-funded Western integrated rare-earth play firmed up — ElementUSA/Colorado School of Mines took a $67M DOE processing award on top of USA Rare Earth's $1.6B Commerce package earlier this week. CONTRADICTS the "only integrated Western supplier" pillar; this is now a Washington-subsidized multi-horse race. REMX −8.6% confirms the price siding against the bull. | ⁵
• FCX | Conviction: DOWN | Surprise: MED | Goldman cut its global copper mine-supply forecast 350kt *citing Grasberg's slower-than-expected recovery specifically* — bullish for the price (deficit), bearish for FCX's own output pillar, and it stacks on Monday's Indonesia export-control move (a named thesis-breaking trigger). CONTRADICTS the "Grasberg output sustained" pillar. | ³
• SCCO | Conviction: HOLD | Surprise: LOW | Same Goldman cut widens the structural deficit — CONFIRMS pillar 1 — and SCCO carries none of FCX's Grasberg/Indonesia beta. Today's COPX −9.9% / copper −4% is the everything-down flow, not a demand crack. No Peru/Mexico disruption. | ³
• LEU | Conviction: HOLD | Surprise: MED | URA −10.3% / URNM −9.5% is a large single-day move, but it sits inside a sell-everything tape (gold down too) with zero new HALEU/SWU fundamentals — flow, not demand. The Russian-SWU-ban + sole-US-enricher thesis is untouched. Watch, don't move. | ⁶
• CEG / VST | Conviction: HOLD | Surprise: LOW | The week's dominant thread — power, not chips, as the global AI constraint (AirTrunk 5GW India, $710B capex, blackout warnings) — is a structural demand floor under firm baseload. CONFIRMS the nuclear-for-AI pillar. Henry Hub $3.22 and the uranium drop are flow; no PPA news to move on. | ¹
• ETN / VRT / PWR / GEV | Conviction: HOLD | Surprise: MED | The buildout demand keeps confirming (continental DC starts all week), but watch the *disconfirmer building underneath*: a national data-center moratorium is gaining poll momentum, with local backlash, police controversy and scrapped campuses piling up. Near-term demand floor intact; permitting is now a real medium-term tail risk to flag. | ⁷

Every other tracked thesis: no thesis-moving signal — recurring buildout/nuclear/regulatory items already in the prior.

Prediction calls for next week (falsifiable)

1. Mean-reversion bounce. If today's selloff is flow not demand (it is), COPX, URNM and XME each close ≥4% above Friday's level by next Friday. *Wrong if* any of them makes a fresh lower low and stays there — that would mean demand, not leverage, is selling.

2. Uranium stabilizes. URA does not print another >5% down day next week absent a new fundamental; LEU/CEG/VST hold their priors. *Wrong if* uranium legs down again on no news (would signal forced selling isn't done).

3. MP keeps fading. REMX underperforms XME next week and no MP-positive catalyst (offtake, China re-tightening) lands. *Wrong if* China reinstates/tightens export controls — the one thing that re-rates MP.

4. Demand floor prints. At least one new hyperscaler PPA or behind-the-meter power deal (CEG/VST/BE/GEV) is announced — the power-is-the-constraint thread is too strong to go a week quiet. *Wrong if* the wire is silent on new generation deals.

5. Backlash hardens. The data-center moratorium thread produces at least one concrete action — a state/local moratorium vote or a permitting denial. *Wrong if* it stays op-eds and polls.

6. FCX/SCCO spread widens. If Indonesia formalizes June export rules, FCX underperforms SCCO by ≥2 points. *Wrong if* Indonesia carves out Grasberg or delays.

7. Financing crack widens, not the demand story. Hyperscaler CDS / AI-capex funding worries get *louder* next week (more two-sided coverage), even as physical deals keep landing. *Wrong if* credit chatter fully reverses and metals + equities rally together on demand alone.

Inflection Radar

Macro Through-Line Synthesis

The week reveals a convergence on Infrastructure Constraint as the dominant theme. The exponential demand from AI compute (Compute) is immediately colliding with physical limitations in power generation and transmission (Power/Energy). This is not a theoretical problem; it is manifesting in utility-level decisions ⁸ and private capital deploying to build localized power sources ⁹¹⁰.

Geopolitics is reinforcing this constraint via resource control ¹¹ - Uranium). Simultaneously, the frontier of AI is bifurcating: one path is deep integration into established platforms ¹² - Apple Agents), and the other is physical embodiment ¹³¹⁴¹⁵, which demands immediate, reliable, and localized energy access.

The key tension is that technological capability (AI/Robotics) is outpacing the regulatory and physical build-out of necessary supporting infrastructure (Grid/Power).

Prediction Calls for Next Week

1. Power/Grid: Expect increased focus on *localized* or *private* power generation solutions (e.g., microgrids, on-site generation) as the primary hedge against regulatory grid delays.

2. AI/Robotics: The narrative will shift from *capability* (what robots *can* do in simulation) to *utility* (what robots can do reliably in a constrained, real-world environment, like a home or industrial site).

3. Resource/Geopolitics: Expect cross-domain linkage between AI compute build-out plans and critical material sourcing (e.g., linking data center location selection directly to mineral supply chains).

***

Emerging Inflection Signals

[emergent] In-Person AI Agents | The market is signaling a pivot from purely digital AI interaction to physical, in-person, or localized group experiences, suggesting a maturation point for consumer AI adoption. | Touches: NEW | ¹⁶

[emergent] Private Compute Infrastructure | Major financial players are bypassing traditional cloud/utility models to build dedicated, proprietary data centers, signaling a belief that compute density will become a highly localized, asset-backed commodity. | Touches: NEW | ⁹

[emergent] Resource Nationalism in Energy | State-level agreements regarding strategic materials (Uranium) are becoming visible, indicating that energy supply chains are already being treated as geopolitical assets separate from market pricing. | Touches: NEW | ¹¹

[emergent] Waste-Stream Critical Materials | The energy transition is creating a new, immediate materials supply chain opportunity by targeting lithium and other resources within existing waste streams (e.g., wastewater, mine drainage). | Touches: NEW | ¹⁷

[emergent] Platform AI Agent Integration | Apple's approval of Poke signals that the next wave of AI compute monetization will occur through deeply integrated, platform-gated agent services, rather than standalone model access. | Touches: T2 | ¹²

[emergent] Regulatory Constraint on Power | The debate around grid speed and competition (FERC) is crystallizing regulatory risk, making energy infrastructure development a primary, high-stakes bottleneck for all compute expansion. | Touches: T3 | ¹⁸

[dismissive] Reliance on Legacy Power Sources | The DOE ordering a coal unit to remain online specifically to support potential data center load signals that immediate energy demand spikes are overriding long-term decarbonization mandates. | Touches: T2 | ⁸

QA & Caveats

No issues found.

Sources

1. AirTrunk commits $30B to build 5GW of AI data centers in India techcrunch.com
2. news.google.com news.google.com
3. Goldman slashes copper supply outlook as deficits widen northernminer.com
4. Hormuz Traffic Remains Near Zero as Peace Talks Stall bloomberg.com
5. ElementUSA, Colorado School of Mines win $67M award for rare earth processing plant mining.com
6. Louisiana Energy Services, LLC, dba Urenco USA; National Enrichment Facility; Exemption federalregister.gov
7. Most Americans Want a National Data Center Moratorium heatmap.news
8. DOE orders OUC’s 465-MW coal unit in Florida to continue running utilitydive.com
9. Quant trading firm Jane Street plans data center - report datacenterdynamics.com
10. Jane Street Group plans to build its own data center - Bloomberg - Investing.com news.google.com
11. Iran Agrees to Transfer Part of Its Enriched Uranium Stockpile, Likely to China - Мілітарний news.google.com
12. Apple approves Poke as the first AI agent on its Messages for Business platform techcrunch.com
13. GENISOM AI debuts deployable robotics platforms at ICRA 2026 therobotreport.com
14. Is Silicon Valley ready to put robots in people’s homes? Hello Robot is. techcrunch.com
15. LadderMan: Learning Humanoid Perceptive Ladder Climbing arxiv.org
16. The most interesting startups right now want to get you off your phone techcrunch.com
17. Stardust Power joins DOE-backed lithium extraction program mining.com
18. Speed to power requires more transmission, not less competition utilitydive.com