Weekly Macro & Outlook — 2026-07-10

Dek

The week the AI buildout stopped worrying about whether tenants would show up and started worrying about wire, transformers, town councils — and, on Friday, learned to pay for all of it with borrowed money.

The Week in Macro

The week opened by defusing a bomb. For weeks the one clean way the whole AI-power trade could break was an energy-price spike — expensive fuel wrecking the economics of gas- and grid-fed data centers. That risk drained away: tankers moved again through the Strait of Hormuz on the US-protected Oman corridor, OPEC+ agreed to add barrels, and Saudi Arabia cut prices to Asian buyers (¹). Oil wobbled again late in the week as US-Iran talks stumbled (²), but the tail risk that shadowed every power thesis is smaller than it was Monday. Cheap electrons keep the buildout math intact.

Midweek settled the demand question for good. Anthropic signed a $19 billion, 20-year lease with former bitcoin miner TeraWulf (³) — a named frontier lab committing for two decades. Then Walmart signed its first-ever nuclear power deal with Constellation (⁴), the first big *non-tech* buyer paying up for firm, carbon-free power. That's a bigger deal than it looks: the fear was that the whole story rested on a handful of hyperscalers. A giant retailer bidding for the same scarce electrons says the scramble has spilled out of the tech club into the wider economy. Permitting eased the same week, with the NRC proposing to streamline environmental review for reactors (⁵).

Then the grid said no. America's biggest grid operator, PJM (which runs power for 13 mid-Atlantic states), set an all-time peak-demand record during an early-July heat dome (⁶) — a record set by *air conditioners*, before most AI load is even plugged in. If a heat wave alone maxes the system, the headroom left for gigawatt-scale computing is effectively negative. The political side hardened too: projects died or shrank at town councils in Bonner, Montana; Tyler, Texas; and Edged's Pennsylvania campus, cut from six buildings to three (⁷). The constraint is now physical *and* political — power is the whole story, and it's tight before AI shows up.

Materials told a story the price tag hid. Copper pushed toward $14,000, but global stockpiles sit at records and keep building — Macquarie flatly calls it "a rally running ahead of reality" (⁸). Translation: the metal isn't scarce in the ground, it's just not where it's needed as finished wire, magnets, and transformers. Rare earths echoed it — billions in US-funded output is flowing to Japan and Korea because American magnet factories haven't been built yet (⁹). The chokepoint is conversion and delivery, not extraction.

Friday delivered the week's real punchline: the buildout learned to borrow. TeraWulf moved to raise $3.5 billion in debt, led by Morgan Stanley, to build for Anthropic (¹⁰) — signed tenant, long lease, no new shares. Wall Street is now underwriting these projects like commercial real estate: credit-worthy anchor, borrow against the lease. That pulls the miner-to-landlord trade off speculation and onto contracts and cash flow — the single thread that ties the whole week together.

Where the Theses Moved

• WULF — conviction UP. The $3.5B debt raise against a signed Anthropic lease confirmed the weakest pillar of the miner-to-landlord thesis: these buildouts can be financed by borrowing, not by diluting shareholders. This is the week's genuinely new information, not a restatement of a high prior.
• CEG — confirmed, buyer base widening. Walmart's nuclear PPA plus the NRC/NEPA permitting easing both cut the same way: the customer pool for firm power is broadening beyond hyperscalers, directly attacking the concentration risk that was the soft spot in the Constellation case.
• FCX — conviction DOWN. Copper near $14k while stockpiles hit records undercuts the structural-extraction-deficit pillar FCX leans on hardest. The bottleneck is conversion, not mining — which strengthens "own-the-bottleneck" logic and weakens the raw-deficit bet.
• SCCO — HOLD. Same copper cross-current, netted out: Giustra's "six new mines a year to 2050" call (¹¹) supports the long-run deficit, but near-term stockpiles undercut it. Neutral for the low-cost producer whose margin floor holds regardless.
• VST — confirmed. The PJM record and NEPA streamlining both help the restart/merchant-power clock, but the signal is recurring against an already-high prior — hardened, not re-rated.
• MP — HOLD. US-funded rare earths still leaking to Asia confirms the gap MP aims to fill but shows the Western chain isn't closed yet; no new execution signal to move a low prior.

Next Week — Prediction Calls

• WULF prices or firms the $3.5B raise within ~2 weeks, and at least one other miner-to-landlord name (IREN or KEEL) floats debt against a tenant. The debt-financing template is the week's new mechanism; expect it to be copied fast. *Wrong if* WULF pulls or cuts the raise by more than 30% — that reopens dilution risk and reverts the WULF call.
• Copper stockpiles keep building even as the price holds above $13,500 — the "ahead of reality" divergence persists, keeping pressure on FCX. *Wrong if* two straight weeks of stockpile draws appear, which would revive the deficit story and re-rate FCX up.
• A new firm-power PPA lands from another non-tech or hyperscaler buyer, following Walmart — the CEG/VST buyer-broadening continues. *Wrong if* no new firm-power contract is signed by next Friday, which would mark Walmart as a one-off.
• At least one more grid-tightness print or local data-center cancellation. The heat dome and consent fights aren't over; expect another PJM stress headline or a killed/shrunk project. *Wrong if* a clean interconnection clears with no new local "no" — that would mean the consent-risk narrative is overstated.
• MP stays a laggard absent a fresh China export-control move or a DOD milestone. The metal-to-magnet gap is structural and slow. *Wrong if* China announces a new rare-earth restriction — MP spikes on the scarcity premium.
• Oil stays rangebound and does not re-emerge as a power-thesis risk, even with US-Iran talks fragile — Hormuz flows and OPEC+ supply hold. *Wrong if* a Hormuz disruption sends crude sharply higher, which would re-price the gas-fed data-center cost stack and put the macro tail risk back on the table.

Thesis Impact

The week in one line

Demand stopped being the question; delivery — and now financing — became the whole game.

Macro through-line

Monday–Tuesday: the macro scare deflated. The Hormuz oil-shock that shadowed every power thesis for weeks drained away — tankers moving again on the Oman corridor, OPEC+ adding barrels, Saudi cutting prices to Asia. A sustained energy-price spike was the one clean way the whole gas-and-grid data-center cost stack could have broken. It didn't. That quietly firmed every power thesis without moving any single one, and pointed back at the real chokepoint: getting electrons to the meter.

Wednesday: demand got settled, and the buyer base widened. The Anthropic–TeraWulf $19B, 20-year lease killed the lingering "will anyone fill these buildings" doubt. Then Walmart signed its first-ever nuclear PPA with Constellation — the first big *non-tech* buyer paying up for firm, carbon-free power. That directly attacks the concentration risk in the nuclear theses (CEG, VST): the scarcity bid is broadening across the economy, not just deepening inside the hyperscaler club. Permitting eased the same week (NRC/NEPA streamlining), pushing the binding constraint off paper and onto physical build.

Thursday–Friday: the grid hit its ceiling on air conditioning alone. PJM set an all-time peak-demand record during the heat dome — *before* the AI load is even connected. If a heatwave maxes the system, the headroom for gigawatt-scale compute is effectively negative. That hardens "power is the binding constraint" as hard as it gets. The counter-pressure showed up too: projects died at town councils (Bonner MT, Tyler TX, Edged PA cut in half). The constraint is now physical *and* political.

Materials told a different story than the price. Copper pushed toward $14,000, but global stockpiles are at records and still building — Macquarie flatly calls it "a rally running ahead of reality." That's the week's most important materials tell: supply isn't scarce, it's just not where it's needed. The chokepoint is conversion and delivery, not extraction. That strengthens Own-the-Bottleneck and undercuts the raw-deficit pillar the highest-torque copper name (FCX) leans on. Rare earths echoed it — billions in US-funded output flowing to Japan/Korea because US magnet capacity hasn't caught up: the metal exists, the conversion step doesn't.

Friday's punchline — the buildout learned to borrow. TeraWulf moved to raise $3.5B in *debt*, led by Morgan Stanley, to build for Anthropic. Signed tenant, long lease, no new shares. The market is now underwriting these projects like real estate — the miner-to-landlord trade comes off speculation and onto contracts and cash flow. That's the cross-domain thread of the whole week.

Net thesis shift: Own-the-Bottleneck and the nuclear/grid book got confirmed but not re-rated (already-high priors, recurring signals). The genuinely *new* information is (1) debt financing validating the miner-to-landlord names without dilution, and (2) the copper stockpile divergence quietly eroding the extraction-deficit story.

Prediction calls for next week (falsifiable)

1. WULF prices or firms terms on the $3.5B raise (or it's upsized) within ~2 weeks, and at least one other miner-to-landlord name (IREN/KEEL) floats debt-against-tenant. *Falsifier:* WULF pulls or cuts the raise >30% → dilution risk re-enters, conviction reverts.

2. Copper stockpiles keep building even as price holds >$13,500 — the "ahead of reality" divergence persists. *Falsifier:* two straight weeks of stockpile draws → the deficit pillar reasserts and FCX re-rates *up*.

3. A new non-tech or hyperscaler nuclear PPA lands, following Walmart — CEG/VST buyer-broadening continues. *Falsifier:* no new firm-power PPA by next Friday → the broadening was a one-off.

4. Another PJM/grid tightness print or ≥1 more local project cancellation. *Falsifier:* a clean interconnection clears with no new "no" → the consent-risk narrative is overstated.

5. MP stays a laggard absent a China export-control action or DOD milestone. *Falsifier:* fresh China REE restriction → MP spikes on the scarcity premium.

Thesis-moving signal

• WULF | Conviction: UP | Surprise: MED | $3.5B debt raise (Morgan Stanley-led) to build for Anthropic confirms the weakest pillar — buildout financed by borrowing against a signed tenant, *not* dilutive equity; CONFIRMS the miner-to-landlord thesis and de-risks pillar 3 | ¹⁰
• FCX | Conviction: DOWN | Surprise: LOW | Copper near $14k while stockpiles hit records and keep building — Macquarie calls the rally "ahead of reality"; CONTRADICTS the structural-extraction-deficit pillar FCX leans on hardest, though the 0.45 prior already reflects most of it | ⁸
• SCCO | Conviction: HOLD | Surprise: LOW | Same copper cross-current — Giustra's "six new mines a year to 2050" call supports the long-run deficit, but near-term stockpiles undercut it; nets out neutral for the lowest-cost producer whose margin floor holds regardless of price | ¹¹
• MP | Conviction: HOLD | Surprise: LOW | US-funded rare earths still flowing to Asia because domestic magnet capacity isn't built yet — recurring, and it cuts both ways (confirms the gap MP aims to fill, but shows the Western chain still leaks); no new execution signal to move the 0.26 prior | ¹²
• ISRG | Conviction: HOLD | Surprise: LOW | Teleoperated humanoids performed preclinical surgeries at UC San Diego — a "beyond da Vinci" headline touches the competitor trigger, but preclinical/teleoperated is years from scaled, profitable placement; does not move the "only proven scaled robotics" pillar | ¹³

Inflection Radar

[dismissive] Regulatory Power Vacuum | Former FERC officials warn that stripping agencies of independence exposes consumers to market risks without regulatory guardrails, signaling potential policy instability in grid investment. | Touches: NEW | ¹⁴

[emergent] Transmission Constraint Economics | DOE quantifies that transmission congestion added $12 billion in wholesale power costs in 2024, making physical grid capacity the primary economic bottleneck for clean energy adoption. | Touches: NEW | ¹⁵

[emergent] Agentic Tooling Architecture | New research proposes replacing the costly, repetitive code regeneration loop in LLM agents with dedicated tool-making frameworks, improving latency and reliability for autonomous systems. | Touches: NEW | ¹⁶

[emergent] AI Trajectory Inference (Bio) | The combination of single-cell transcriptomics and LLM agents is emerging as a powerful new framework for reconstructing complex developmental cell paths, linking compute power directly to fundamental biological research. | Touches: NEW | ¹⁷

[emergent] FERC Procedural Pressure | FERC is soliciting public comment on its information collection activity (FERC-555), indicating that regulatory bodies are actively defining the scope and constraints of future infrastructure data requirements. | Touches: NEW | ¹⁸

QA & Caveats

No issues found.

Sources

1. bloomberg.com bloomberg.com
2. Oil Slumps as US, Iran Continue Talks After Renewed Fighting bloomberg.com
3. datacenterdynamics.com datacenterdynamics.com
4. utilitydive.com utilitydive.com
5. federalregister.gov federalregister.gov
6. bloomberg.com bloomberg.com
7. utilitydive.com utilitydive.com
8. Macquarie says copper price rally still running ahead of reality mining.com
9. Billions in US-funded rare earths are Asia-bound: report northernminer.com
10. TeraWulf seeks $3.5bn in funding for construction of data center for Anthropic datacenterdynamics.com
11. Giustra: Copper market needs six new mines a year to 2050 mining.com
12. Billions in US-funded rare earths flowing to Asian markets: report mining.com
13. Beyond da Vinci: Why versatile humanoid robots are the next frontier in surgery therobotreport.com
14. Former FERC officials weigh in on Supreme Court ruling expanding president’s power to fire regulators utilitydive.com
15. What can best ease transmission bottlenecks? More transfer capacity, DOE says. utilitydive.com
16. Tool-Making and Self-Evolving LLM Agents in Low-Latency Systems arxiv.org
17. SpaCellAgent: A Self-Evolving LLM-Based Multi-Agent Framework for Trajectory Analysis arxiv.org
18. Commission Information Collection Activity (Ferc-555); Comment Request; Extension federalregister.gov