How concentrated this is
Holdings5
Weighted like4.2 equal ones
LargestNVDA at 27.6%
Half the book2 names
Exposed to the premise90%
Half of this book is 2 names. If they are wrong, the rest will not save it.
Compute is the binding constraint on artificial intelligence, not model design.
Universe v2 (assumed — this link predates the field)· No stated date on this link
Every holding, and why it is that size
Restore the book27.6%Nvidia
Sits on the critical path of nearly every frontier training run.
NVDA · Equity · $223.67 · — · Compute buildout
−+×27%Micron
High-bandwidth memory is the second bottleneck after the die.
MU · Equity · $1027.77 · — · Compute buildout
−+×26.2%Broadcom
Custom accelerators for buyers who want off a single vendor's price list.
AVGO · Equity · $364.38 · — · Compute buildout
−+×9.2%Super Micro Computer
Integration and assembly — real volume, thinnest moat in the chain.
SMCI · Equity · $38.93 · — · Compute buildout
−+×10%0–3 Month Treasury Bill ETF
Cash that still earns while the thesis is early.
SGOV · Treasury · $100.49 · — · Ballast
The case for
Frontier training runs are gated by physical supply, not ideas. Leading-edge capacity is booked years out, high-bandwidth memory is sold before it is made, and grid interconnection queues run for years. Anything on that critical path collects the toll.
The case against
Every capex supercycle in semiconductors has ended in a glut. Efficiency gains, smaller models and custom silicon all cut the accelerators needed per unit of intelligence, and buyers are a handful of firms who can pause in one quarter.
What the weights are worth
Percentages, converted. Nothing is sent anywhere.
Enter an amount to see what each weight is worth. It stays in this browser.
This book has an address
Anyone who opens it rebuilds the same book, weight for weight.
igitur.xyz/b/compute-is-the-binding-constraint-on-1whzpopOr fix it to a date
Put the claim on the public record — with these weights — and it is measured from the day the server saw it, win or lose, permanently.