Working Hypotheses
-
Backward induction: how to understand the semiconductor supply chain reverberations of new advances in AI/software (e.g. agentic workflow).
- How will my company have to shift production? [suppliers]
- How will my company have to change procurement? [consumers]
- How will my national economy (or that of my enemy) be affected? [government]
- How will my portfolio perform? [investors]
-
Parametric insurance of fabs/data centers : can we create a mechanism that monitors telemetry from some phase of semiconductor manufacturing, producing dashboards for individual companies but then feeding into parametric insurance policies that can trigger based on the live statistics.
- Physical piece of hardware that creates parameter
-
Glencore of chips
- AI enabled distributor that begins to mirror Glencore
- Study the difference between Arrow/Avnet and Glencore
- Understand the ways AI can disrupt the operations of these companies
-
Compliance and supply chain fidelity for defense
- Gov does genuinely demand this info
-
Data collection/labeling platform (similar to what Alia’s friend Luna does; I think she works for Mercore) but for physical world stuff
- Ie go and buy up tons of real estate drawings and docs and whatnot from every local developer/home builder (focus on the little guys at first cus they prob won’t understand the value) and then sell/license that to physical AI labs/companies for training their models
-
Take on SAP head on/build an SAP compliment that SAP then buys
- Take the lessons from Athey’s class about threat of compliments to build a comiment
-
[Some sort of AI for GC/CLO functionality]
-
Insurance for AI agents doing rogue stuff; customer is GCs/CLOs
I think it’s important for us to ground ourselves in the facts…
-
We have not yet found any signal towards our initial idea
-
We also have not yet spoken to anyone who actually works in the semiconductor supply chain
I think the most important shift for us to make is to stop thinking about anything besides the following question: who is our customer?
Then we need to channel all of our efforts relentlessly towards finding and talking to and understanding that potential customer.
We need to do all of this quickly and efficiently.
I think our goal for tomorrow should be sitting down together, thinking about who specifically is our customer, and then sending 20 emails/LinkedIn messages per person to prospective customers asking to set up meetings
GCs and compliance teams at semicondcutor companies feel increased hours because they have to respond to litigation from USG which can be solved by a data model that provides intellegence into upstream compliance risks
Hedge fund investors focused on the semiconductor industry are constrained in deploying capital because they lack an understanding of how emerging technological trends will impact the different layers of the semiconductor supply chain, which can be solved by a backward induction product that takes a topic and inductively reasons through upstream impacts
senior executives at semiconductor companies feel uncertainty around how to strategically assess and invest in their business because they lack understanding of how emerging technological trends will effect their business, which can be solved by a backward induction product that takes a given trend and inductively reasons through impacts on their business
- NVIDIA isnt sure which chip to design because they’re not sure how claude opus 4.7 is going to change demand for compute from claude opus 4.6
- they don’t understand the impact of openclaw on their demand
Lenders, equity investors, and companies who are investing capital into the building of new facilities or purchase of new equipment are facing increased risk and corresponding expense due to uninsurable or expensive to insure risks, which can be solved by creating a decide that emits telemetry and can be used to underwrite a parametric insurance policy
PMs at NVIDIA feel uncertainty in terms of how to design their products because they don’t understand the implications of major trends in consumption, which can be solved by a data model that takes a given trend and inductively reasons through upstream impact