Evening Analysis — 2026-05-30

The Big Shift

SoftBank said it will spend up to €75bn to build 5 gigawatts of new data-center capacity in France, partnering with Schneider Electric at the Port of Dunkirk — one of the largest single AI-power demand prints we've logged (¹). Why it matters: it's hard, capital-committed evidence that power, not chips, is the binding constraint on AI — the core of our Own-the-Bottleneck thesis. What it signals: the demand curve keeps steepening, but this is another point on an already-steep line, not a re-rate. With that pillar near its ceiling (we put the odds at 0.85), confirming news now moves the needle far less than it did six months ago.

Analysis

Power. The SoftBank print lands on top of a wall of smaller buildouts the same day — Ascenty's $1.2bn/150MW in Brazil, a $283m pre-leased site in Malaysia, new 100MW projects in Finland and off-grid in Yorkshire (²). The thread building across days is that "physical energy access is now the primary bottleneck": Entergy's gas plants make up a third of MISO's fast-track grid queue, almost all of it to serve data centers (³), and developers are increasingly going behind-the-meter (their own on-site generation) to skip the grid wait entirely. Implication for the thesis: the bottleneck is real and getting worse, which keeps the equipment layer (VRT, ETN, PWR) structurally bid — but see the caveat below.

The efficiency counter-thread. Two days running, the compute signal says the real gains come from smarter software, not more silicon — Nvidia is shipping a 4-bit quantized Qwen model that runs the same workload on far less hardware (⁴), and TSMC's own deputy COO now says energy-efficient compute is what customers ask for first (⁵). This is the cleanest bear case for the whole chain: if each chip does more work per watt faster than demand grows, the power crunch eases and the buildout overshoots. Implication: it doesn't break the thesis yet, but it's the variable that could turn "perpetual shortage" into "glut" — worth weighting more each day it recurs.

Materials and geopolitics, now decoupled. The Strait of Hormuz is reopening — ship transits are rising with US help, Qatar is haggling over a temporary toll, and a tentative US-Iran deal is in reach (⁶). WTI fell 1.3% to $87. That pulls the war-driven cost-push premium out of copper, lithium and nickel. Implication for FCX/SCCO: the copper theses now stand on structural electrification demand alone — no geopolitical kicker left. The tape agrees the structural floor holds (copper flat at $642, copper miners off just 0.3%), so we hold both, but the easy tailwind is gone. Giustra's "prices must rise as supply can't keep up" call (⁷) is the long-run support; it isn't a near-term catalyst.

The equipment block looks priced. VRT, ETN, PWR and GEV have re-rated together as a bet on data-center electrical content, and posteriors keep climbing (ETN 0.86, PWR 0.81). But conviction has now outrun fresh evidence: the SoftBank deal names Schneider as the equipment partner, not Eaton, and no liquid-cooling order or design win is attached for Vertiv. Implication: we're holding, not adding — the news confirms the demand source but gives no company-specific reason to push these higher.

Rare earth fading as a trade. The magnet-chokepoint story (MP) is structurally intact, but the tape disagrees — rare-earth miners fell 1.8% today and chairman Litinsky just sold $26.2m of stock. Implication: the chokepoint as a *pattern* is real (it shows up across China's robotics and humanoid push, ⁸), but MP as a *position* is losing momentum and insider behavior isn't helping.

What Would Prove Us Wrong

Efficiency outpaces demand: a second or third major model release this week showing large per-watt gains (like the NVFP4 quantization), plus any hyperscaler trimming its 2026 capex guide below the ~$725bn consensus (⁹). That hits Own-the-Bottleneck and the whole equipment block.
Demand actually cancels, not just slows: more deals like Sabey backing out of its Montana data center (¹⁰), or the local-permitting backlash hardening into blocked projects (¹¹). A pattern of cancellations — not single sites — would crack Pillar 1.
Copper breaks its structural floor: COMEX copper closing below ~$600 (down from $642) with copper miners (COPX) confirming, now that the Hormuz premium is gone. That would signal the electrification-demand story alone can't hold the price, hitting FCX/SCCO.

Thesis Impact

Own-the-Bottleneck | Conviction: HOLD | Surprise: MED | SoftBank commits up to €75bn to build 5GW of new French DC capacity (Schneider partner, Port of Dunkirk) — a large, genuinely NEW demand print that CONFIRMS Pillar 1 (power is the binding AI constraint). But it's another point on an already-steep curve, not a re-rate, and the pillar is near ceiling (P=0.85) | ¹

ETN | Conviction: HOLD | Surprise: LOW | Same SoftBank buildout = more electrical content into DCs, CONFIRMS the diversified-exposure thesis — but the named equipment partner is Schneider, not Eaton, and at P=0.86 with conviction already past the evidence, no ETN-specific news here to move it | ¹²

VRT | Conviction: HOLD | Surprise: LOW | More DC capacity CONFIRMS the power/cooling demand source, but no liquid-cooling design win or order figure attached — already reflected in the prior | ¹²

FCX / SCCO | Conviction: HOLD | Surprise: LOW | Hormuz reopening (WTI -1.3%) keeps draining the war cost-push tailwind from copper, leaving both theses on structural electrification demand alone — mildly CONTRADICTS the near-term setup, but it's recurring (0d) and the tape disagrees (copper flat at $642, COPX -0.3%), so structural thesis and prior both hold | ⁶

Inflection Radar

[emergent] AI Memory Bottleneck | XCENA's funding bet that AI's primary constraint is memory, not compute, signals a structural shift in hardware investment focus. | Touches: NEW | ¹³

[emergent] Geopolitical Uranium Flow | Kazakhstan's overtures regarding Iran's uranium stockpile, reinforced by multiple reports, signal potential non-Western supply chain realignment for nuclear fuel. | Touches: NEW | ¹⁴

[emergent] Asymmetric Defense Tech Push | Concerns over US arms sales pauses forcing Taiwan toward indigenous, non-traditional defense technologies. | Touches: NEW | ¹⁵

[dismissive] Consumer Robotics Failure Mode | Reports of robot startups allegedly damaging private property during field tests suggest regulatory gaps and consumer risk in early deployment. | Touches: NEW | ¹⁶

[emergent] AI Inference Funding Pivot | Groq's reported $650M raise indicates a market pivot toward optimizing AI inference efficiency rather than just raw compute power. | Touches: NEW | ¹⁷

[emergent] Infrastructure Execution Risk | DOE's SPARK funding success hinges on the readiness of utilities to execute transmission development, highlighting policy-to-physical gap risk. | Touches: NEW | ¹⁸

QA & Caveats

ETN call is not supported; the source cites Schneider, not Eaton, undermining the specific equipment link.
VRT call over-reads the source; the source only confirms capacity buildout, not specific cooling technology wins.
All other calls are supported by the cited context or represent known market contradictions.

Sources

SoftBank says it will invest up to €75 billion to build French data centers techcrunch.com
Ascenty announces $1.2bn investment to deploy 150MW data center capacity across Brazil datacenterdynamics.com
Entergy’s gas projects are one-third of MISO’s fast-track interconnection process utilitydive.com
nvidia/Qwen3.6-35B-A3B-NVFP4 · Hugging Face reddit.com
Energy efficient compute is most important attribute for customers, TSMC claims datacenterdynamics.com
Strait of Hormuz Ship Transits Are Rising Thanks to US Help bloomberg.com
Copper prices must rise as supply crunch looms: Giustra northernminer.com
scmp.com scmp.com
news.google.com news.google.com
Sabey backs out of proposal to build data center in Butte, Montana datacenterdynamics.com
Data Centers Are Splintering the American Right heatmap.news
SoftBank plans up to 5GW data center buildout in France, investment of up to €75bn datacenterdynamics.com
This chip startup just raised $135M on a bet that AI’s biggest bottleneck isn’t compute — it’s memory techcrunch.com
Kazakhstan offers to take Iran’s uranium stockpile, IAEA chief says - Al Arabiya English news.google.com
US arms sales pause would push Taiwan toward asymmetric-defense tech: Analysts defensenews.com
Robot Startup Accused of Running Secret Airbnb Field Tests That Allegedly Damaged Rental Properties reddit.com
After Nvidia’s $20B not-acqui-hire, AI chip startup Groq reportedly raising $650M techcrunch.com
Once you secure SPARK funds for transmission development, what comes next? utilitydive.com