Evening Analysis — 2026-06-28

Dek

The day's scariest headline — tankers burning near Hormuz — went quiet by evening, clearing the geopolitical fog off the real story: AI's appetite for power is now the structural trade, and nothing this week made it smaller.

The Big Shift

The US and Iran agreed to stop shooting at each other and sit down for talks, after a weekend flare-up that put a hit supertanker near the Strait of Hormuz — the chokepoint about a fifth of the world's oil passes through (¹). Oil gave back its early spike on the news (²). Why it matters: a sustained Hormuz closure was the one fast-moving event that could have forced energy prices and policy to revolve around crude again instead of electricity. With that tail risk easing tonight, the spotlight swings back to the slow, structural squeeze — AI compute pulling on the power grid — which is where the desk's core thesis lives.

Analysis

The thread that ties the day together is a paradox flagged on our own cross-domain radar: a cheaper model doesn't free the grid — it fills it faster. China's DeepSeek shipped a major upgrade to its V4 model built specifically to speed up "inference" — the work a model does each time it answers — and cut the chip strain per query (³). The intuitive read is "less hardware needed." The thesis read is the opposite: when each answer gets cheaper, people and companies buy far more answers, so total demand climbs. That's why hyperscaler capex projections still point at roughly $600 billion for 2026 (⁴). Efficiency is not relieving the buildout; it's underwriting it.

If demand is the constant, the action is on supply — and the supply answer keeps coming back as nuclear. Walmart signed on with Constellation (⁵), Constellation's Ginna plant filed today for a license to run decades longer (⁶), and the Trump administration laid out its long-promised nuclear buildout plan (⁷). Why it matters: existing reactors getting 20-year extensions is the cheapest, fastest megawatt available, and corporates locking up that output directly is a tell that the grid can't supply them on normal terms. The implication for the thesis: firm, always-on power — not solar fields — is what AI buyers actually pay up for, and the holders of that power (Constellation, Talen) capture the scarcity premium. Talen's stack of material agreements in its latest 8-K fits the same pattern.

The counter-pressure is political, and it's getting louder. A senior House Democrat called for a national moratorium on new data centers, on the heels of a Ratepayer Protection Act meant to stop AI buildout from raising ordinary people's electric bills (⁸). Why it matters: the bottleneck on AI may turn out to be neither chips nor reactors but voters who don't want to subsidize them. That's the live political risk to the "demand grows unchecked" thesis — watch whether it moves from press release to actual law.

On materials, Nigeria says its Kaduna state holds a "world-class" critical-minerals province with a large lithium find (⁹). Why it matters: every reactor, transmission line, and battery in this buildout runs on metals, and new non-China supply is the slow-burn relief valve for the whole chain. It won't move prices this quarter, but it's the kind of discovery that, over years, decides whether the energy thesis is constrained by geology or just by permitting.

And the day's reality check: skepticism is hardening around Elon Musk's orbital data center pitch, with SoftBank's CEO among the doubters (¹⁰). Why it matters for the thesis: the exotic "escape the grid" answers aren't ready, which leaves terrestrial power — nuclear, gas, the existing grid — as the only place to put the next gigawatt. That keeps the bottleneck exactly where our thesis says it is.

What Would Prove Us Wrong

• A data center moratorium or the Ratepayer Protection Act actually becomes law (not just a press release) — this hits the core "AI power demand grows unchecked" thesis directly, because it caps the buildout by political force rather than physics.
• Hyperscaler 2026 capex guidance gets cut materially below the ~$600B mark alongside the DeepSeek-style efficiency wins — that would mean efficiency is shrinking demand, not feeding it, breaking the Jevons logic that underpins the whole power-scarcity trade.
• Hormuz talks collapse and crude spikes back toward $100+ — this disconfirms tonight's "tail risk eased" read and would drag policy and capital back to oil-supply security, crowding out the electricity-and-compute story for weeks.

Thesis Impact

No thesis-moving signal.

Inflection Radar

[dismissive] Hyperscaler AI Over-investment Thesis Flaw | Steve Eisman suggests current investment models funding hyperscalers for AI are misallocating capital, signaling a potential structural correction in market focus. | Touches: NEW | ¹¹

[emergent] FERC Regulatory Headwinds for Data Centers | FERC developing a new plan introduces potential utility and regulatory constraints on data center power draw, creating an infrastructure bottleneck that could cap AI scaling regardless of chip advances. | Touches: NEW | ¹²

QA & Caveats

No issues found.

Sources

1. US and Iran Agree to Halt Attacks Ahead of Talks bloomberg.com
2. Oil Pares Early Gains as US, Iran Halt Attacks After Flare-Up bloomberg.com
3. scmp.com scmp.com
4. AI Infrastructure Investment 2026: $600B Hyperscaler Boom - Intellectia AI news.google.com
5. Walmart Goes Nuclear heatmap.news
6. Constellation Energy Generation, LLC; R.E. Ginna Nuclear Power Plant; Subsequent License Renewal Application federalregister.gov
7. Trump’s Big Nuclear Play Is Here heatmap.news
8. Key House Democrat Calls for a National Data Center Moratorium heatmap.news
9. Nigeria identifies major new critical minerals district, large lithium discovery mining.com
10. SoftBank’s CEO isn’t the only one with questions about Elon Musk’s orbital data center hype techcrunch.com
11. Steve Eisman says investors betting on hyperscalers in the AI race are funding the wrong side of the trade - Startup For news.google.com
12. FERC Has a New Plan for Data Centers heatmap.news