Evening Analysis — 2026-06-09

Dek

China just told Washington "no" to its face on rare earths — and that one word did more for the thesis tonight than a week of AI headlines.

The Big Shift

The day's hardest signal was a refusal. China reaffirmed its export ban on rare earths to Japan *despite* a direct US request to lift it, confirmed by two independent sources off the same Nikkei report (¹, ²). Why it matters: Beijing held the line under explicit diplomatic pressure — the single cleanest proof that export controls are a sustained policy tool, not a bargaining chip to be traded away. What it signals: Japanese magnet buyers now have a structural reason to fund non-China supply, and the one event that would break the rare-earth thesis (China relenting) just got less likely. That re-rates the lowest-conviction name (MP) the most, because it had the most room to move on exactly this.

Analysis

Materials are the night's real story, and it's a clean confirm. The rare-earth refusal pushes conviction on MP UP and floors the "Own-the-Bottleneck" thesis — the core idea that a single-country chokepoint equals durable pricing power. The risk to that idea is "allied supply matures and breaks China's grip." Tonight does the opposite: China demonstrated it will hold controls even when an ally (the US) asks nicely on behalf of another ally (Japan). That's the bullish read for anyone outside China who can dig, refine, or magnetize.

Copper now has to stand on its own feet. The Mideast de-escalation thread — Trump floating an Iran deal in "two or three days" and Hormuz reopening "immediately" (⁴) — drains the fear premium that had been lifting copper and power equities together. The implication is sharper than it looks: if copper's recent rise was bought by *risk appetite* rather than *demand*, then the SCCO and FCX trades must now be carried by structural AI buildout alone. That's why the SCCO-over-FCX split keeps widening — the market is separating the miner with real supply discipline from the one riding the macro bid. Note the contradiction, though: ³ reports fresh US strikes on Iran after an Apache was downed in Hormuz (³), so "de-escalation" is a hope, not a fact. The S&P swung violently and nearly erased a 2.3% plunge on these same headlines (⁵) — the tape itself can't decide.

Compute and power: the demand side is holding its breath for Oracle. No new co-location or PPA filing landed today, so the generation-owner thesis (CEG, VST — owning the power that AI needs behind the meter) held rather than advanced. The structural backdrop stayed firm: Applied Digital signed a hyperscaler for a 210MW campus (⁶), Jefferies says AI data-center demand still far outstrips supply despite record capex (⁷, and New York's grid operator warned reserves are shrinking (⁸) — every thread points to power as the binding constraint. The equipment-block 8-K cadence kept printing too (VRT and KEEL today), a six-day sector-wide filing pattern that keeps a floor under the picks-and-shovels names (VRT, ETN, PWR, GEV). What it implies: the supply-side case is intact; the demand-side case needs Oracle's call tonight to confirm hyperscaler spend is still accelerating, not plateauing.

The counter-thesis got louder, but not heavier. The "efficiency kills the power trade" argument strengthened for a third straight day — a Korean inference chip claiming 48GB VRAM at 180W (⁹), a $501 5070ti, GPU power-throttling tips, Apple's on-device CoreAI. The idea: if inference gets radically cheaper per token, the power and copper demand underpinning CEG/VST/SCCO softens. But every item is a forum post (T3) with no T1/T2 step-change behind it. These are leads, not a conviction move — worth watching, not yet worth trading.

One political cloud is darkening. Polling shows a national data-center moratorium gaining momentum and most Americans wanting one (¹⁰), while Microsoft is preemptively seeking a Nevada tariff to shield ratepayers from data-center costs (¹¹). The implication for the power thesis: the bottleneck may shift from *can we build it* to *will communities let us* — a slower, messier risk than a grid-capacity number, and one that doesn't show up in capex charts.

What Would Prove Us Wrong

• China actually resumes rare-earth shipments to Japan (a confirmed lift of the ban, not a rumor). This is the direct thesis-breaker for MP and Own-the-Bottleneck — it would erase the scarcity pricing the whole materials trade depends on. Watch Nikkei and MOFCOM for any reversal of today's refusal.
• Oracle guides hyperscaler capex flat or down tonight, or signals an efficiency-driven slowdown in compute buildout. That hits the generation-owner thesis (CEG, VST) and the structural-demand leg under copper (SCCO) — the trades that just lost their geopolitical bid and now have nothing else to stand on. Measurable: the actual capex number and forward guidance versus consensus.
• A T1/T2 inference-efficiency proof point — a benchmarked, sourced collapse in cost-per-token or power-per-query, not a Reddit thread. That would upgrade the counter-thesis from "leads" to a real threat against the entire AI-power complex.

Thesis Impact

MP | Conviction: UP | Surprise: MED | China reaffirmed its rare-earth export ban on Japan *despite* a direct US request to lift it — two independent T2 sources (Mining.com + Northern Miner, both off Nikkei). Beijing holding the line under explicit diplomatic pressure CONFIRMS pillar 3 (export controls sustain Western scarcity/pricing) and pushes Japanese magnet buyers toward non-China supply like MP. Avoids the thesis-breaker (China lifting controls). Most material because MP is the lowest-conviction name with the most room to re-rate on exactly this. | ¹

Own-the-Bottleneck | Conviction: HOLD | Surprise: LOW | Same China rare-earth refusal directly touches pillar 3 (single-country chokepoint = pricing power) and counters the thesis-breaker "alliance supply matures, breaking single-country control." Confirms — but recurring export-control action is literally the expected news here, so it floors the prior rather than lifting an already-high 0.81. | ²

*Noted, not moved:* the de-escalation thread (Trump "deal in 2-3 days," T3 recurring) drains the geopolitical bid flattering SCCO/FCX/CEG/VST, but it's price-interpretation, not new demand info, and ³'s fresh strikes contradict it — HOLD pending Oracle's call. The inference-efficiency counter-thesis strengthened again (Furiosa chip, cheap 5070ti, GPU throttling) but every item is T3 with no T1/T2 step-change — leads only, no conviction move.

Inflection Radar

[dismissive] Solar Permitting Risk | State-level regulatory action can rapidly invalidate large-scale renewable energy infrastructure plans, creating significant investment uncertainty for utility-scale solar developers. | Touches: NEW | [¹²]

[emergent] AI for Nuclear Process Control | Applying advanced AI techniques (like agent protocols) to solve the multi-year, high-cost regulatory bottlenecks inherent in advanced nuclear reactor design. | Touches: NEW | [¹³]

[emergent] Offline RL for Fusion Control | Using historical operational data (offline RL) to train complex plasma control algorithms for fusion reactors, bypassing the extreme cost and risk of real-time physical testing. | Touches: NEW | [¹⁴]

[emergent] Edge LLMs on Spatial NPUs | The convergence of specialized, energy-efficient hardware (Spatial NPUs) with LLMs, signaling a shift toward deploying complex AI models directly onto resource-constrained edge devices. | Touches: NEW | [¹⁵]

[emergent] Robotics for E-Waste Stream | Development of specialized, vision-guided humanoid robotics capable of safely and selectively disassembling complex, hazardous waste streams like lithium-ion batteries. | Touches: NEW | [¹⁶]

[emergent] Data Center Interconnection Bottleneck | FERC proceedings are crystallizing the conflict over who pays for transmission upgrades required by massive, intermittent loads like AI data centers, signaling a major utility/tech friction point. | Touches: NEW | [¹⁷]

QA & Caveats

• The MP call is supported by the cited source (Source 17/8) regarding the China/Japan rare-earth restriction, but the leap to confirming "pillar 3" requires external context not provided in the immediate citation.
• The Own-the-Bottleneck call is largely descriptive of the cited news (Source 8), suggesting it is already priced in rather than a surprise catalyst.
• No issues found.

Sources

1. China reaffirms Japan rare earth ban despite US request mining.com
2. China stands by Japan rare earth export restrictions northernminer.com
3. US launches new strikes on Iran after helicopter downed defensenews.com
4. Trump says Iran deal could be reached in 'two or three days' and Strait of Hormuz will reopen 'immediately' reddit.com
5. S&P 500 Traders Grapple With Swings Reminiscent of Tariff Tumult bloomberg.com
6. Applied Digital secures customer for new 210MW data center campus datacenterdynamics.com
7. AI-driven data centre demand far outstrips supply despite record hyperscaler spending: Jefferies - ET Telecom news.google.com
8. New York state's electricity reserves are shrinking, grid operator says | Reuters reddit.com
9. Furiosa AI selling inference chip to consumer market will be a game changer to local llm reddit.com
10. Most Americans Want a National Data Center Moratorium heatmap.news
11. Microsoft seeks Nevada tariff to shield ratepayers from data center costs utilitydive.com
12. Ohio Supreme Court overturns permit for massive 6,000-acre solar farm reddit.com
13. Overcoming the Regulatory Bottleneck via Agent-to-Agent Protocols: A Nuclear Case Study arxiv.org
14. Offline Reinforcement Learning for Plasma Control in Nuclear Fusion: Codebase and Benchmark arxiv.org
15. From Human Guidance to Autonomy: Agent Skill System for End-to-End LLM Deployment on Spatial NPUs arxiv.org
16. Vision-Guided Dual-Arm Humanoid Robotic Disassembly of End-of-Life 18650 Lithium-ion Battery Packs arxiv.org
17. FirstEnergy asks FERC to require data centers to pay for transmission interconnection costs utilitydive.com