Evening Analysis — 2026-05-29

The Big Shift

The clearest thesis-moving signal tonight is small but pointed: MP Materials' chairman and CEO Jared Litinsky sold $26.2 million of stock, right after a strong run and on a day the rare-earth ETF (REMX, a basket of rare-earth miners) fell 1.3% (¹). Why it matters: the whole MP thesis rests on a "scarcity re-rating" — the idea that Western rare-earth supply is so chokepointed that the stock should keep climbing. When the top officer trims into strength, it quietly says *he* doesn't see much more near-term upside. What it signals: not a break (one sale, possibly pre-planned), but a single notch down in conviction — and a reason to watch whether more insiders follow or whether the rare-earth price tape keeps moving *against* the bullish story.

Analysis

Materials: the bull case is intact on paper, but the tape and the insiders disagree. The supply-crunch logic got another endorsement today — financier Frank Giustra arguing copper *must* rise as electrification, AI, and grid build-out outrun new mines (²). But this confirms the thesis without adding new information, and the prices went the other way: copper flat at $642, COPX (copper miners) −0.3%, REMX −1.3%. The implication for the materials thesis: the *story* is fully priced and now needs fresh catalysts (a real supply shock, a price breakout) to re-rate further. Insiders trimming and ETFs drifting down is what a crowded, fully-believed trade looks like before it either consolidates or rolls over.

Power and the Strait: the war premium is deflating, which pulls two ways. Hormuz ship transits are rising with US help (³) and a tentative US–Iran deal is taking shape (⁴), with WTI crude down to $87.76. That eases one of the materials theses' hidden costs: the sulphuric-acid crunch (acid is needed to process lithium, copper, nickel) was driven by the Hormuz closure (⁵). If the strait reopens, that input-cost tailwind for miners' *prices* fades — cutting against the "everything-mineral re-rates" trade even as it helps the broader macro. Net implication: a Hormuz reopening is bearish for the commodity-scarcity narrative, not bullish.

Compute and capex: the demand engine keeps roaring, but skeptics are stirring. Hyperscaler capital spending is now pegged at $725 billion for 2026 (⁶) — the single biggest source of demand under every power, copper, and equipment thesis on the desk. But "AI capex fatigue" chatter is now showing up from hedge funds (⁷). Implication: the demand pillar holding up power and materials is still intact, but for the first time the *question* of when spending plateaus is being asked aloud. That's the variable that, if it turns, takes down the most theses at once.

Power supply: the "who generates the electrons" fight is sharpening into a testable split. Tonight's strongest cross-domain thread is behind-the-meter generation — power made *on-site* at the data center — flipping from backup plan to default plan, which makes Bloom Energy (BE) a direct AI beneficiary and puts pressure on the grid-co-location premium baked into nuclear names like CEG and TLN. Today gave fresh fuel: an off-grid 100MW gas-powered data center in Yorkshire (⁸) and a flat argument that "hyperscalers had no choice but to become power companies" (⁹). Implication: BE and the co-located-nuclear thesis are now in *direct tension* — if on-site generation becomes the norm, the scarcity value of grid-connected nuclear capacity erodes. This is the pair to watch resolve.

The equipment layer remains the cleanest expression of the whole complex. The thread that "the gear layer wins regardless of who generates the power" keeps strengthening — switchgear, transformers, power electronics get bought first no matter whether the electrons come from the grid, gas, or fuel cells. TSMC reinforcing that energy-efficient compute is now customers' top priority (¹⁰) points the same way: efficiency and power-delivery hardware is where the durable demand sits. Implication: amid the noise in commodities and the BE-vs-nuclear fight, the equipment thesis is the one carrying the least binary risk.

What Would Prove Us Wrong

More MP insider selling or a sustained REMX downtrend. One $26.2M sale is a notch down; a second insider exit or REMX breaking lower over the next week would turn the MP scarcity-re-rating thesis from "fully priced" to "rolling over." Watch the next round of Form 4 filings and whether REMX holds ~$100.
A confirmed Hormuz reopening with crude falling toward the low $80s. That would collapse the sulphuric-acid cost tailwind and undercut the broad commodity-scarcity trade (copper, lithium, uranium miners) — bullish macro, bearish for the materials theses that have been riding war-driven input costs.
A hyperscaler trimming 2026 capex guidance, or one large data-center project cancelled for funding (not siting) reasons. "Capex fatigue" is just talk until a number gets cut. A single guidance-down from a Magnificent-7 name would hit every downstream thesis at once — power (CEG/TLN/VST), equipment (ETN), and copper demand.

Thesis Impact

MP | Conviction: DOWN | Surprise: MED | Chairman/CEO Litinsky sold $26.2M in stock — a NEW, T2 insider sale landing right after the recent run-up and on a day REMX fell -1.3%. Insiders cashing out isn't thesis-breaking (no 10b5-1 context given, sales can be planned), but a sale this size from the top officer is a mild CONTRADICTION of the "scarcity premium re-rating" story — when the chairman trims into strength, it dents the conviction that he sees more upside. One sale, so a single notch down, not a structural break. | ¹

Everything else tonight was recurring data-center/grid/copper/uranium flow (LOW surprise, already in the priors), opinion pieces restating theses (Giustra on copper — CONFIRMS but no new info), or T3-only leads (MTP inference speedup, Hormuz transit chatter, Nvidia chip tease) that can't move conviction without T1/T2 corroboration. The Hormuz reopening (T1, Bloomberg) is real but touches no tracked thesis directly.

Inflection Radar

[dismissive] Robot Field Failure | Allegations of secret field testing damaging property highlight unmanaged operational risk in early-stage robotics deployment, suggesting regulatory/liability gaps before mass adoption. | Touches: NEW | ¹¹

[emergent] Category Theory in LLMs | The CCT architecture proposes grounding language models in category theory, suggesting a structural shift in how models build abstract knowledge beyond pure statistical correlation. | Touches: NEW | ¹²

[emergent] Memory Bottleneck Thesis | XCENA's funding bet on memory, supported by academic work (e.g., CosmicFish), signals a consensus shift away from pure compute scaling toward memory architecture optimization as the primary AI constraint. | Touches: NEW | ¹³

[emergent] AI for Resource Extraction | DOE research deploying AI agents to optimize critical minerals recovery from waste suggests a convergence point: advanced compute solving deep industrial supply chain/sustainability bottlenecks. | Touches: NEW | ¹⁴

QA & Caveats

No issues found.

Sources

Mp materials chairman & CEO Litinsky sells $26.2 million in stock investing.com
Copper prices must rise as supply crunch looms: Giustra northernminer.com
Strait of Hormuz Ship Transits Are Rising Thanks to Help From US bloomberg.com
How far is there to go until the US and Iran end the war? defensenews.com
CHARTS: How the sulphuric acid crunch is driving up critical minerals costs northernminer.com
'Magnificent 7' earnings rush reveals AI spending surge, with hyperscaler capex set to reach $725 billion in 2026 - AOL. news.google.com
AI Capex Fatigue: Why Hedge Funds Are Questioning the Hyperscaler Spending Boom: - HedgeCo.Net news.google.com
UK's Reabold Resources seeks partner for 100MW off-grid gas-powered data center in Yorkshire datacenterdynamics.com
Hyperscalers didn’t set out to be power companies. The grid left them no choice. utilitydive.com
Energy efficient compute is most important attribute for customers, TSMC claims datacenterdynamics.com
Robot Startup Accused of Running Secret Airbnb Field Tests That Allegedly Damaged Rental Properties reddit.com
The Cognitive Categorical Transformer: Category-Theoretic Inductive Biases for Language Modeling arxiv.org
This chip startup just raised $135M on a bet that AI’s biggest bottleneck isn’t compute — it’s memory techcrunch.com
US deploys AI agents to speed critical minerals recovery northernminer.com