Buyer concentration tightening
countervailing a named mechanism, not a conclusion
These names sell into a concentrated buyer set — above 2,500 on a Herfindahl, and several above 5,000, which is two or three customers. Pricing power sits with the buyer, and a qualification win reads as revenue while it is actually dependence.
The path
| node | effect |
|---|---|
| IREN Limited | -1.00 |
| Oracle Corporation | -1.00 |
| Custom AI ASIC (XPU) | -1.00 |
| Interconnect bandwidth (scale-up and scale-out) | -1.00 |
| Cipher Mining Inc. | -1.00 |
| Applied Digital Corporation | -1.00 |
| Core Scientific, Inc. | -1.00 |
| Lambda, Inc. | -1.00 |
| JX Advanced Metals Corporation | -1.00 |
| Kioxia Holdings Corporation | -1.00 |
| Lenovo Group Limited | -0.71 |
| MediaTek Inc. | -0.69 |
coverage
customer_hhi SQUARES customer shares, so error in the undisclosed supply weights is squared with it. This is the least reliable quantity the desk computes and the ordering should be treated as indicative only.
Arguments about this mechanism
None yet. No thesis names this mechanism’s subject, which is a gap in the bank rather than a fact about the mechanism.
Arguments that run through it
These name a node on the path rather than the subject, so they pass through this mechanism without being about it. Shown separately and never graded as claims about it — hubs above the graph’s own 90th-percentile degree are excluded, or every argument touching NVIDIA would attach to every mechanism NVIDIA touches.
- GPU collateral decay gates buildout faster than physical supply
- Long-context inference: memory constraint compresses cloud margin before GPU-poor plays feel it
- MoE inference: converted miners capture memory scarcity, hyperscalers leak it
- The Neocloud Rent Is Consumed By The Asset Before It Reaches Equity
- NVIDIA's Rent Compresses Through Software, Not Silicon
- Optical attach is set by topology, not by shipments — linear optical TAM models are mis-specified