Investing & Markets
I study the contracts and cash flows behind infrastructure, energy, and digital assets.
Infrastructure rights, software ownership, and AI power economics.
At UT Austin and Texas Venture Labs, I work with market research, pricing, financial models, and unit economics. My independent studies retain the documents, assumptions, and calculations behind each conclusion.
Mapping the claims on a Texas toll road
My toll-road investigation separates public title, concession rights, equity, creditors, operating responsibilities, and billing. North Tarrant Express illustrates the distinction: sponsor percentages describe the concession business, while the highway land remains publicly owned. The concession’s revenue rights extend through 2061.
SH 288 provides a different event. When Texas terminated its private concession, the proceeds first repaid outstanding debt, including the federal TIFIA loan; the remainder was available to shareholders. The ownership table records nine facilities or systems with their operator, revenue claimant, term, and supporting sources.
Choosing a comparable revenue basis
The follow-up model separates five questions: recovery of original cost, road debt, system debt, legal tolling authority, and private concession rights. Paying down one balance does not answer all five.
The system table retains seven systems, but the per-$100 comparison includes only CTTS and Grand Parkway on the same TxDOT FY2025 basis. Other systems have different revenue categories or reconciliation limits. Keeping those exclusions visible avoids a precise-looking comparison built from incompatible denominators.
| Question | Evidence required |
|---|---|
| Original cost recovered? | Comparable capital cost and the relevant revenue period. |
| Road or system debt retired? | Debt balances and the facilities pledged to each obligation. |
| Tolls legally required or permitted? | Statutory authority, bond terms, and concession rights. |
Research cutoff: September 2, 2026. Road and system debt are separate tests, as are public tolling authority and private rights.
Classifying liquidity and control separately
I assembled 25 software transactions announced in 2020–2022, with $171.2 billion in mixed-definition headline deal values. Each record follows who still controls the company, rather than treating every IPO or financing event as a completed sponsor exit.
At the August 17, 2026 cutoff, 23 remained in continuing sponsor ownership, one had transferred to creditors, and one had partial liquidity. None met the stated clean control-exit definition. That classification is not an estimate of losses or returns: the public record does not support a cohort IRR, and most deals were only three to five years into ownership.
Turning a gigawatt claim into a model
The AI-power model distinguishes a grid request, facility capacity, IT capacity, rack nameplate, installed fleet, and utilization. Each boundary changes what a “1 GW” announcement can support.
The workbook converts facility power through PUE and external IT overhead to 142 kW, 72-GPU rack equivalents. At PUE 1.145 and 10% overhead, one nominal facility gigawatt gives about 402,574 B300 GPU equivalents. Changing utilization changes annual equivalent compute hours; it does not establish a paid workload or a verified installed fleet.
AI-power workbook · XLSX
Six worksheets with formulas, boundary calculations, and scenario assumptions.
Sensitivity table · CSV
PUE and overhead scenarios with explicit capacity and utilization outputs.
Power-model methodology
Definitions, formulas, source transformations, and evidence cutoff of August 16, 2026.