Ten investment bets I actually hold on how AI reshapes markets. Every weekday morning a machine reads the news, prices and fundamentals, then grades each bet's evidence from strongly supports to strongly against and publishes it unedited, agree with me or not. The kill criteria were written down before the money moved. You decide what you believe.
the chip is the machine's verdict on today's evidence for each bet. the % is how that bet's current stock pick has done against the market since it was named.
my honest ordering: today's evidence strength first, and each pick's edge over its benchmark since it was named. it will be wrong sometimes — that's what the receipts are for.
each row is one thesis: a specific bet about where value lands as AI plays out. the list is ordered by how firmly today's evidence backs each bet.
the chip grades today's evidence only. a thesis can be losing money while its evidence reads green, and the other way around.
the right column is the scoreboard. each thesis names one stock as its current pick, and the % is that stock's return minus its benchmark index since the pick was named. +2.0% means it beat the market by 2 points, −2.0% means it trailed by 2. just named means the pick is brand new, so its score starts at the next market close.
the small squares are the last two weeks of daily evidence reads, oldest to newest. the arrow is movement in this ranking since the previous run.
a short is a bet that a stock FALLS. bad news for that company reads green here, and its pick % counts a falling price as a gain.
One email each week: which theses gained or lost evidence, what tripped a flag, what the receipts caught. 10 theses tracked, scored against the market, published unedited.
The theses aren't independent bets — they share drivers. A green reinforce edge means two theses ride the same underlying force from different angles; a red tension edge means the same bet taken from opposite sides. $TSLA, for example, sits on both T2 (Physical AI, long) and T6 (the Musk premium, short): if Optimus ships the premium is earned, if it doesn't the premium compresses. Node color is today's evidence, size is how strong the read is. The ring around each node is that bet's last ten runs of evidence, so a long green arc means the evidence kept going its way for days. Bright fast synapses fired today, both ends of that wire moved at once. Hover anything for the numbers.
every card is one bet. tap it for the companies behind it, their numbers, the kill criteria and the daily evidence trail. for a short bet the colors invert, falling prices count as green there.
Intelligence becomes free; value flows to applications with proprietary data and workflows, not model labs
VLA models turn robots into software platforms; public markets underprice it because leaders are private
Enterprise AI is hard; value flows to integrators and AI-literate services while models commoditize
AI bottleneck shifted from compute to data movement; CPO/photonics wins whether the buildout continues or rationalizes
Deficits get inflated away; scarce assets reprice. Position built during the stress test, not the party
The Musk empire trades on narrative, not cash. Tesla's auto cash cow is cracking — Q2'26 operating profit fell 57%, free cash flow went negative, and regulatory credits (once ~2pts of margin) collapsed as the EV credit expired — yet it's priced at ~145x earnings; the private crown jewels are only reachable at steep premiums to NAV. The bet: the premium compresses toward the cash the cows actually produce.
Companies generating outsized revenue and profit relative to their market cap — high margins, real free cash flow, low multiples — are mispriced. The market treats durable profit machines like they're stagnant; as the cash keeps compounding, they re-rate upward.
The AI buildout's binding constraint is shifting from chips to electricity. Datacenters are signing decade-long power deals faster than anyone can build generation; firms that own or equip gigawatts capture pricing power for years
AI turned memory from a cyclical commodity into the binding constraint — HBM is sold out years ahead, conventional DRAM and NAND supply is being cannibalized to feed it, and pricing power has shifted to a three-player oligopoly for the first time in decades
Per-seat software loses its unit of account when agents do the work. Mid-market SaaS with seat pricing and thin data moats gets repriced first: seat counts stall, net revenue retention slips, and the multiple compresses. The bet: these names underperform the software index
the next scheduled earnings reports that will stress-test these bets.
one line per bet per weekday, written by the machine and published exactly as written.
Prices and relative performance daily (vs each thesis's benchmark — QQQ, SMH or SPY), company fundamentals (growth, margins, valuation, cash flow), and fresh headlines collected per mechanism signal — the things each thesis actually depends on.
Every weekday Grok reads the day's evidence and scores each signal −2…+2 against the thesis mechanics, with a one-line explanation. I publish both unedited — even when it disagrees with me. That's the tape.
I wrote the kill and double-down criteria before taking the positions — they're shown verbatim on every page. Sustained negative evidence puts a thesis into kill review. No quiet goalpost-moving.
A one-person experiment in accountable investing. Nothing here is investment advice or a recommendation to buy or sell anything. Do your own work — that's the point of showing the data.