Worldview
Investment Worldview
✓ Canon · META-WORLDVIEW.md
Updated —
1
Chokepoint Priority Log
*Track changes to the constraint hierarchy over time.*
scroll to explore↓
2
The Single Core Belief
"**Demand is no longer the variable. Constraints are the variable.**"
3
What Has Permanently Changed
AI infrastructure CapEx is committed, multi-year, and driven by structural necessity — not consumer sentiment or quarterly refresh cycles. Hyperscaler CapEx is now modeled like utility infrastructure:
4
Sovereign AI & The Techno-State
In prior eras, national power was measured in industrial output and military capacity. In the current era, a new measure has emerged: who controls the compute, models, data jurisdiction, and the execu
5
The 6 Structural Chokepoints
"⚠️ **This list is dynamic.** Chokepoints shift in priority as constraints get resolved, new bottlenecks emerge, or the relative severity of each layer changes. A chokepoint that is binding today may be partially solved in 12 months (e.g. transformer capacity catches up), while a new one emerges (e.g. liquid cooling becomes the new ceiling). The ranking below reflects the current ordering — **update when new source material or market evidence changes the constraint hierarchy.**"
- ▸Jensen Huang (NVIDIA): *"Power generation and grid capacity"* — #1 bottleneck
- ▸Elon Musk (Tesla/xAI): *"The electrical grid growing too slowly"*
- ▸Andy Jassy (Amazon): *"Decade-long wait times for grid connections"*
- ▸Larry Ellison (Oracle): *"The need for gigawatt-scale power plants"*
- ▸Satya Nadella (Microsoft): *"Fully permitted buildings with power connected"*
6
The Software Control Plane
"**"Logic is cheap. Data is scarce. Control and physics win."**"
- ▸Workload orchestration and scheduling (who runs what, where, with which data)
- ▸Identity, access, and policy enforcement (what models can query what datasets)
- ▸Audit trails and compliance (provenance, approval gates, regulatory evidence)
- ▸Model lifecycle control (deploy, evaluate, rollback, version lineage)
- ▸Operational resilience under real-world threat models
7
AI Factory Waves — Constraint Sequencing
"**Note:** Specific timing is hard to call. Treat the ordering as logic, not calendar. Multiple waves can be co-active. A wave that appears priced-in in one sub-category may still be early in another."
8
Screening Philosophy
"*"If AI infrastructure deployment slows next quarter due to a physical constraint, does this company get richer or poorer?"*"
- ▸Pure unit-volume demand plays without a constraint moat
- ▸Consumer-facing hardware priced on cyclical assumptions
- ▸Node-scaling stories without system-level deployment viability
- ▸Single-region supply chains in geopolitically contested materials
- ▸Companies that are "beneficiaries" of the trend but not controllers of any constraint
9
Open Questions / Things to Watch
- How fast does the CPO transition actually happen? The timeline matters for photonics chokepoint strength.
- ▸How fast does the CPO transition actually happen? The timeline matters for photonics chokepoint strength.
- ▸Does liquid cooling become a hyperscaler standard or remain optional? Changes the thermal infrastructure investability window.
- ▸How quickly can Western rare earth and magnet supply chains scale? Political will is there; execution timeline is not yet clear.
- ▸Will HBF (High-Bandwidth Flash) materially displace HBM in inference workloads? Changes the memory controller thesis.
- ▸Does China's domestic semiconductor acceleration (RISC-V, mature nodes) create new investable chokepoints outside the US/EU stack?
10
Source Notes
- Treat as T2 source material (analyst synthesis + executive statements, not primary SEC/IR filings)
- ▸Treat as T2 source material (analyst synthesis + executive statements, not primary SEC/IR filings)
- ▸C.C. Wei (TSMC) quote on packaging is effectively T1 given source credibility
- ▸Tier: T3 — retail synthesis; LightCounting market data T2-adjacent
- ▸Tickers: AXTI, LITE, AEHR, COHR, AAOI, TSEM, ALMU, AIXA, VECO (new: AIXA, VECO)
- ▸Contribution: InP 70% supply gap quantified at Layer 0; AIXA + VECO stubs created (MOCVD equipment layer); ALMU stub updated; LandMark noted as private watch