Some thesis-centric weekend reading, with our take on each, tying them to the house view.
Contents
- CoreWeave goes to junk
- Climate inflation is not rate-sensitive
- Oracle has a power problem
- Data-center opposition moves upstream
- Amazon builds the robot factory
- Burry on the capital cycle
- Goldman found the clean trade
- Thirty-year Treasury at 2004 levels
- AI enters the utility control room
- Utilities call bullshit on the queue
- The biggest economic bet in U.S. history
- Goldman gets to $1.2 trillion
- GMO: what actually breaks the AI bubble
- AI discovers upcoding
- AI is saving scientists seven hours a week
- Walmart draws the line at personalized pricing
- The shorts spread through the AI supply chain
- Australia was not AI alone
1. CoreWeave goes to junk — Bloomberg
A CoreWeave-leased Virginia data center raised $1.1 billion at 9.25%. Fifteen-year lease, one tenant, non-investment-grade credit. This is exactly the financing migration discussed here for months: AI capex spilling out of hyperscaler balance sheets into SPVs, private credit, leases, guarantees and junk bonds.
2. Climate inflation is not rate-sensitive — Allianz
Allianz estimates Europe’s 2026 heatwaves cost $128 billion, with a strong El Niño capable of adding materially to global inflation next year. This supports the point made here repeatedly: a meaningful share of measured inflation now comes from supply shocks that monetary policy cannot fix. Oil, power and weather do not respond to another 25 basis points.
3. Oracle has a power problem — Bloomberg
Oracle invoked force majeure around Project Jupiter in New Mexico over possible power delays. Another one for the file. The binding constraint on AI infrastructure was always going to move from chips to power, transmission and permitting. Oracle is now writing that problem into its contracts.
4. Amazon builds the robot factory — WSJ
Amazon is spending another $100 million on an Indiana robotics plant. More interesting than another model release. The economic payoff comes when AI and automation remove expensive physical labor from the cost structure.
5. Data centers are power systems with buildings attached — Visual Capitalist
Some planned campuses have load profiles comparable with major cities. Calling them “data centers” understates the economics. These are industrial power complexes, with all the transmission, generation, water, land and political constraints that it implies.
7. Burry on the capital cycle — Michael Burry
Burry puts hyperscaler commitments, leases and guarantees in the trillions and points to investment intensity near historic extremes. This is very close to the argument made here for some time: AI can be economically transformative while the capital cycle around it destroys enormous amounts of capital. Railroads did it. Telecom did it. There is no reason AI should be different.
8. Goldman found the clean trade — FT
Goldman reportedly made more than $200 million from AI hedge fund Situational Awareness, which had become its largest prime-brokerage client. In a capital boom, financing the participants can be considerably better business than owning the assets. Picks and shovels, except the shovel is, you know, us.
9. Thirty-year Treasury at 2004 levels — FT
Long borrowing costs are at multi-decade highs while AI capex, energy shocks and sovereign issuance all compete for capital. This is exactly why current U.S. inflation and rate data are so hard to read conventionally. AI capex and energy are pushing prices and long yields higher for reasons that another Fed hike does little to address.
10. AI enters the utility control room — Utility Dive
New York is auditing utility use of AI across electric, gas and water systems. Sensible. The interesting AI risk sits inside tightly coupled infrastructure, where small model errors can propagate into large operational ones.