Erica · PM Research MemoAll memos

PM RESEARCH MEMO

Title: Linear vs. Exponential — AGI Has Arrived; Oil, Bonds and the Fed Are the Wrong Regime
Author / source: Jordi Visser (Jordi Visser Labs / 22V Research)
Source title: Linear vs. Exponential: AGI Has Arrived But Oil, Bonds and the Fed Oh My
Source URL: https://www.youtube.com/watch?v=VPuzUZlvk0U
Video ID: VPuzUZlvk0U
Channel: Jordi Visser Labs (@JordiVisserLabs)
Published: Sunday 13 Sep 2026 ~08:30 ET (yt timestamp 12:30 UTC; recorded Fri 11 Sep 2026, opens on 9/11 25th anniversary)
Duration: 49:12 (2952s)
Memo date: Sunday, 13 September 2026 (America/Toronto)
Transcript: /workspace/youtube-transcripts/VPuzUZlvk0U.md · Brief: /workspace/youtube-transcripts/VPuzUZlvk0U_brief.md · Optional JSON: /workspace/youtube-transcripts/VPuzUZlvk0U.json
Caption source: YouTube automatic English ASR only (en-orig json3 via yt-dlp). No manual captions. Names, tickers, and numbers are provisional — see ASR locks below. Timestamps are cue timestamps from the retrieved timedtext.
Source type: Solo thematic / macro education talk. Speaker frames content as education / empowerment, not advice. This memo contains no trade recommendations from this desk.
Product: Regime map and research hypotheses for allocation research. Not advice. Not a recommendation to buy or sell any security.
Source discipline: Primary sources are this transcript and its brief only. Companion Jordi / Gavin Baker / Hunt / Nordvig / Green / Alma / SV memos on this desk are different sources — cited only as External / other desk companions where the tape cross-references a named person; do not silently import numbers from those memos into this underwriting. Speaker-stated figures are used as spoken and attributed; they are not independently verified (External check needed).

How to read this document: Restatements of the talk are Source. Interpretive links and underwriting judgments are Inference. Any fact not spoken in the transcript is External check needed. Language such as “lean AI book / ETH / Robinhood / tokenization” is source expression / research hypothesis, not an Erica / desk recommendation.

ASR name locks (from brief):

ASR heard Likely / note
wars / Worsh / Warsh Kevin Warsh
Besson / Bessent / Bent Scott Bessent
Jensen Yuang Jensen Huang
Astro / Astra Astra (OpenAI agentic system in his frame; identity External check needed)
Navier Stokes / NavierStokes Navier–Stokes
Vera Rubins / Reubens Vera Rubin (NVIDIA next-gen)
Mumu / Moomoo Moomoo
Robin Hood Robinhood
Salana Solana
Arbitum / Arbitron Arbitrum
William Mogayier William Mougayar
Timuros (Jackson Hole) unclear clip / prior guest ref — flag
Shawn Kaitton / Caner personal 9/11 memorial (friend) — not market
aida “the AI / the emotion” — garbled
Paul Tudtor Jones Paul Tudor Jones
Striker Stryker (w/ Boston Scientific cyber)
22vresaerch 22vresearch
Codeex Codex
Grock / Grockbot Grok / agent UI comparison
Demark / Duck Miller DeMark / Stan Druckenmiller
pers PERs / peer comps / perps — flag
manquan Memorial Day / July 4th photo banter

Calendar context (Source / brief): Post Thu PPI + Fri CPI week; Fed hike odds cited ~65→85/90%; next FOMC Wed Sep 16; Freedom Tech DC Sep 22–23, 2026.


EXECUTIVE SUMMARY


SOURCE-ACCURATE SUMMARY

Chapter-ordered with cue timestamps. Short quotes only where they carry the claim (≤20 words).

9/11 open / framing (00:00–02:37)

  1. (00:00–00:42) Recorded Friday September 11; personal memorial for friend Shawn (ASR: Kaitton / Caner) — not market content. Agenda: “linear verse exponential thinking… oil bonds and Fed. Oh my.”

  2. (00:42–01:10) Core punchline early: if you’re paying attention to oil, bonds, and the Fed, “you’re in a linear world.” Big news should be “AGI has arrived.” Not whether yields were up 10 bp, oil up $10, or Fed odds 65%→85%.

  3. (01:10–01:53) Will cover Astra, podcasts to get up to speed, how he built a trading system with agents (esp. for kids / subscribers); attach agents via Moomoo, Robinhood, Interactive Brokers — even paper trading. Entrepreneurial empowerment theme. Also: hard question of how to value Ethereum / crypto ecosystem.

Linear vs exponential regime (02:37–05:42)

  1. (02:37–03:19) Economy used to care about oil/bonds/rates; “that all changed beginning in 2008” — iPhone the year before, transfer payments, government debt. Exponential / Mag7 world has been making money since. Silicon Valley trip 2013 as personal inflection (later Amazon PE story).

  2. (03:19–03:48) Tape setup: higher-than-expected core inflation “yesterday,” higher-than-expected PPI, Fed hike probability up, crude through, war worse — “over the two days… the S&P was up slightly.” “How do these things actually matter if they don’t matter?”

  3. (03:48–04:02) News focused on linear world inputs that mattered when housing/autos drove the economy. “AI capex is driving the economy” and is not rate sensitive (last week’s show).

  4. (04:02–04:31) Fed hike odds “surged to 90%.” Jackson Hole / “Timuros” clip flagged unclear. Market “double dog daring” Warsh; fighting Warsh and Bessent. His preference: make money on the exponential side.

  5. (04:31–04:59) Fed third mandate: moderate long-term interest rates. Trump clearly has Bessent fighting the market; Warsh chosen by Bessent/Trump; “they don’t want rates higher.”

  6. (04:59–05:42) Bessent upsized buybacks — “goes to six billion” — market immediately sells off (“good news hit with bad news”). Plan B: “cut long-term issuance outright.” “I am the House now” — engagement incl. Japan finance minister coordination. Reason “fighting the government is” the wrong trade (ASR: “best trade” in context of not fighting). Speaks against “Bloomberg terminal bros” arrogance.

Internals: SPX structure, revisions, credit, AI sleeve, crypto relative (05:42–09:01)

  1. (05:42–06:26) S&P holding 50-day; looks like flag after big high; “barely up since June 2nd.” Unless economy/earnings fall off, “we can keep doing this for a while.” Rip-then-pain pattern; Nvidia analogy — get used to three-day rips when it feels impossible.

  2. (06:26–07:22) Revisions this week “32 up”; 20-week average of revisions rising; PMI leading indicators “doing great.” HYG relative to IEF rising as bonds sell off (= credit OK). Contrast 2022: HYG/IEF fell with Fed hikes + SPX down. Bond vol / junk spreads / TIPS inflation expectations “not moving” — market says it doesn’t care.

  3. (07:22–08:05) AI thematic portfolio broke above 50-day; low-vol consolidation; concentrated ~10-name sleeve “getting close to the August highs”; “names are all acting well.”

  4. (08:05–08:33) Bitcoin flag/range; Ethereum break — “Ethereum should be outperforming Bitcoin when we’re actually in the bull market”; ETH/BTC “breaking out again to the highest level since January.”

  5. (08:33–09:01) Bessent warns nothing else would matter if China wins the AI race. Don’t fight the AI trade; don’t fight the bond side of what he’s saying — you don’t have to go buy bonds, but fighting AI is a bad idea. Choice: linear or exponential side of the V.

AGI / Astra / Navier–Stokes / GDP constraint (09:01–15:11)

  1. (09:01–09:30) “AGI has arrived with Astra” (Jensen Huang as “godfather of AI”). Trained on order ~100,000 GPUs; next model 4×; then Vera Rubins. “Why fade this?”

  2. (09:30–10:12) Research agents generating “3.1 workdays for every human workday.” Navier–Stokes challenge: 10,000 AI agents parallel 88 hours. Compute → intelligence → thousands of agents. Critical point analogous to “October of last year” high-enough IQ to move into agentic second stage.

  3. (10:12–10:55) Agents on Navier–Stokes-class problems → closer to mass-scale medicine, energy, physics. Agentic side = big change; reason he talks crypto; need more compute / GPUs — “GPU prices just refuse to go down.”

  4. (10:55–11:52) GDP formula changes: digital labor + accelerating productivity; agents work 3–5× human effective time (no lunch/flirting/daydreaming/emotion). New constraint: “compute, energy, and our ability to deploy intelligence.”

  5. (11:52–12:21) Productivity already accelerating via profit margins. 1930s industrial GDP metric “wrong” for this; GDP doesn’t count IP / lots of things.

  6. (12:21–14:14) Navier–Stokes importance: energy, transportation, compute, manufacturing, medicine, climate. Economy/TAM gets bigger from science breakthroughs while corporate duration gets shorter (terminal value). Software already lived this; hardware (Nvidia cheap PE) living it. Multiple compression when growth cliffs — “where should your multiple be if you’re not growing anymore?”

  7. (14:14–15:11) Practical implications: better turbines, aircraft, cooling, ships, reactors, batteries. Prefer Anthropic economists over street rate-forecast linear thinking. Greg Brockman OpenAI interview ~24 minutes — Astra description; merging chat + agents + desktop + Codex + tools in one place.

Empowerment / Brockman / AI trading desk build (15:11–31:14)

  1. (15:11–17:19) Grokbot comparison: agents usable though less “video game” UI. Critical for kids 15–24 and knowledge workers who dislike their jobs. Brockman quote on empowerment / superpowers. Subscriber multi-part series on AI entrepreneurial mindset: ask, find, build, test, fail, learn, adapt, repeat.

  2. (17:19–18:43) New ~1.5-hour subscriber video: 14 prompts, GitHub save, ~90%+ of work AI-done. Coding was “Chinese” to him; no longer needs translators. Corporate moats shortening → entrepreneurship more important; many firms fire under pressure, not because AI is thriving for them.

  3. (18:57–21:05) Labor Day Monday: built Astra 6-agent trading desk on small account (Probe, Prism named). Put $500 in; pot to $5,100 by morning (as spoken — backtest/toy; not a performance claim to underwrite). System rejects ~99% of ideas; remaining filtered hard — “mini multistrat” of uncorrelated models. Emotionless agents as edge.

  4. (22:16–25:49) Human edge remains on idea generation. Built technical system for thematic portfolio: structural trend / reset / coiled / ignition using ATR and rate of change on his ~100-name AI universe. Midcycle slowdown = exit example (Micron). Expects structural bull but not another 5-bagger path for Micron from here; multiple compression means most days down/flat and a few days make the year — miss those, miss the move.

  5. (25:49–31:14) Subscriber “treasure hunt” GitHub: code, prompts, research log, Socrates mode. Moomoo / Robinhood / IBKR agentic APIs + paper trading to run mini multi-strat sleeves. Accounting GitHub example: found ≥13 relevant projects. Sundar Pichai line: orchestrate agents now or spend 2027 catching up.

Depreciation / Baker / PTJ / cyber / PEG (31:14–37:50)

  1. (31:14–31:42) Microsoft plans to triple data-center computing power. Demand for Astra “unprecedented” — cites 6.2 million views / running out of space (External check needed what “views” measures).

  2. (31:42–33:07) H100 rental prices “almost at the same level as they were three years ago”; should have fallen ~40–50% in a normal aging process; instead “rental price is up 22% on the month.” Demolishes Burry/Chanos-style depreciation-risk bear case in his telling. Vera Rubins far better than H100s yet rents hold.

  3. (33:07–35:14) Gavin Baker ~1 month ago: market assumed GPU rents fall; older GPUs holding/rising → hyperscaler cloud FCF estimates too low; repricing existing compute “could remove hundreds of billions of dollars of expected credit needs.” Tomato analogy ($2 cost → sell compute at much higher reset). Capacity shortage; MSFT reiterated. Baker detail is Visser’s restatement only — External / other desk for primary Baker sources; do not merge those memos’ numbers.

  4. (35:14–35:43) Paul Tudor Jones negative piece on AI markets (not anti-tech); they exchanged emails; engaged / shocked at pace.

  5. (35:43–36:40) Greg Jensen: AI may need to kill people before regulation. Iranian hackers / UK power plant; Boston Scientific / Stryker earnings hit from cyber; Anthropic fourth AI-hacking incident (Opus 4.6); Iran used US model vs Navy. Cyber ↑ → also positive on crypto because trust breaks; hurts smaller businesses; doesn’t stop earnings/productivity.

  6. (36:54–37:50) AI adopters’ earnings / profit margins higher. PEG “never been cheaper… in the last 30 plus years.” Peak-earnings skeptics: revisions still higher; margins usually roll first — not seeing that. Crypto now ~30% of his focus attention.

Robinhood / tokenization / Clarity / ETH valuation / DC (37:50–49:12)

  1. (37:50–38:47) “Robin Hood convergent” paper: RH as epicenter of AI–crypto–macro nexus — TradFi company + crypto + agentic. Bernstein: 31% upside narrative as fees top Solana; RH chain fees record $6 million daily; chain live ~2 months; Arbitrum surge (~7.87% as spoken) on RH chain news. WSJ didn’t match Arbitrum news — TradFi lag.

  2. (39:15–40:40) Crypto.com deal; “pers” (flag ASR — peer comps / perps). Tokenized stocks advancing. Robinhood at Goldman tech conference takeaways: agentic trading mainstream (“hedge fund in your pocket”); RH chain as global financial infrastructure / tokenization; Clarity Act catalyst for pension credibility; RH wants entire financial relationship (trading, crypto, retirement, banking); prediction markets as gateway; evolving toward AI-native tokenized platform (Apple analogy).

  3. (40:40–41:36) Nasdaq $100m into Kraken, eyeing 2027 tokenized stock trading. US Bank stablecoin tests; Visa stablecoin card “exploding”; European banks euro stablecoin on Ethereum — “all this week.” Coinbase / Brian Armstrong / Patrick Witt: Clarity ready for yes vote; probabilities for passage this year still below 20%, but yes vote still possible.

  4. (41:36–43:16) William Mougayar old TED (~2014–16): Ethereum / blockchain vision — direction 8–10/10, specifics ~5/10. Too aggressive on removing intermediaries (Robinhood as counterexample). Missed stablecoins; biggest miss = AI agents — “they may finally make his thesis work.” Not for humans ultimately.

  5. (43:16–47:06) “Ethereum’s endgame” / how to value ETH: like early internet — eyeballs before earnings before SaaS metrics over ~25 years. Network effects, flows, ecosystem, productive assets, programmable. Amazon 1997→2013 ~27,000%; ETH from lows ~3,000%. Amazon PE never below 100 for most of that stretch; first two quarters PE<100 not until ~2017–18. His 2008 best year was short US financials / long EM/commodities — missed Amazon because it “didn’t make money.” Silicon Valley / Singularity University ~12 years ago = exponential mind-shift. Message to mutual funds / HFs: don’t make the same mistake with crypto if you want bull-market participation. Source expression / hypothesis only. No buy/sell.

  6. (47:06–48:45) Mougayar book Trust Shift; trust as economic asset; Ethereum as trust model; AI makes trust scarce → RWA must go on-chain. Freedom Tech DC Sep 22–23, 2026; Bitcoin Policy Institute / PubKey; tokenization as bridge between digital and linear worlds. WSJ Bitcoin “freedom” piece. Closes on 9/11 memorial gratitude.


SYSTEMS MAP / VALUE CHAIN ANALYSIS

Layer 0 — Macro thermostat (linear world). CPI / PPI → Fed hike odds → Warsh/Bessent theater → long-end buybacks / issuance / Japan MOF coordination → oil & geopolitics. In the speaker’s map this layer still moves headlines but has lost first-order pricing power over SPX and HY credit so long as AI-capex / agentic productivity dominate growth.

Layer 1 — Regime allocator (exponential world). Mag7 / AI-capex complex as post-2008 growth engine; “not rate sensitive” claim (prior show). Policy: don’t fight AI (China race) and don’t fight government on long rates (third mandate / Bessent “house”).

Layer 2 — Intelligence production. GPU / accelerator supply (H100 → Vera Rubin) → hyperscaler data centers (MSFT triple compute) → frontier models / Astra → research agents (3.1 workdays) → multi-agent swarms (10k / 88h Navier–Stokes) → science/TAM expansion.

Layer 3 — Economic transmission. Deployable intelligence → digital labor (3–5×) → rising corporate margins → earnings revisions → PEG compression even as PE debates rage → duration shortening for any firm whose growth cliffs under AI disruption.

Layer 4 — Financing / depreciation arb. Spot GPU rental (H100 +22% m/m, flat vs 3y) → cloud compute repricing at contract resets → hyperscaler FCF upside vs street → less credit issuance need than bears assume (Baker restatement: hundreds of billions). Counterparty: street depreciation schedules still assume scrap/fast fade.

Layer 5 — Trust / rails / crypto. Cyber degradation of soft targets → demand for verifiable trust → stablecoins / tokenization / ETH as settlement & trust layer → Robinhood (and peers) as TradFi × agentic × on-chain distribution. Clarity Act = institutional credibility / pension seal more than existence condition. Agents as native users of crypto rails (connects to prior nexus talks — companion External; no number merge).

Layer 6 — Portfolio construction implication (source method, not desk advice). Select ~100-name AI infrastructure/capex universe → ATR / rate-of-change states (structural / reset / coiled / ignition) → concentrated ~10-name sleeve for expression → crypto sleeve (~30% of attention, not stated as % of AUM) with ETH/BTC relative as bull-market thermometer → paper-trade agentic multi-strat for skill-building.

Bottlenecks (Source): compute capacity, energy, ability to deploy intelligence, cyber defense for smaller firms, Clarity / regulatory credibility for pension capital.
Bottlenecks (Inference): verification of Astra/AGI claims; whether GPU rents stay elevated when Vera Rubin ramps; whether HYG/IEF “don’t care” survives a true growth air-pocket; whether tokenization volumes become P&L-material before PE narratives catch up.


SECOND AND THIRD-ORDER EFFECTS

Chain A — Hot prints that don’t bite → regime confirmation (or trap)

  1. First order (Source): Hot core CPI/PPI + hike odds ~90% + oil up + war worse coexist with SPX slightly up and calm credit/vol/BE.
  2. Second order (Inference): Cross-asset desks that sold equities on the linear checklist underperformed this week’s tape; “Fed theater” becomes a fade until joint bond-vol + junk + BE break.
  3. Third order (Inference): If the ignore pattern persists into/through Sep 16 FOMC, narrative migrates from “inflation scare” to “AI-capex immunizes” — raising complacency risk if growth/earnings later roll. Watch: HYG/IEF, MOVE/bond vol, TIPS BE, revision breadth — not just the hike probability itself.

Chain B — Agent productivity → TAM↑ + duration↓ → barbell book

  1. First order (Source): 3.1 workdays / Navier–Stokes swarm / digital labor 3–5× → science breakthroughs expand economy.
  2. Second order (Source): Terminal values shorten; multiples compress when growth cliffs; Nvidia “cheap PE” as living example; hard to make money the old always-up way.
  3. Third order (Inference): Winners concentrate in infrastructure + agent platforms + trust rails; losers are mid-duration software/services with cliff risk and no agent leverage. Portfolio construction shifts from buy-and-hold growth to rate-of-change / structural-trend inside a curated universe — exactly his ATR state machine. Research, not a rec.

Chain C — H100 rents up → FCF up → credit need down (Baker channel)

  1. First order (Source): H100 rents flat vs 3y, +22% m/m; demolishes fast-depreciation scrap narrative.
  2. Second order (Source, via Baker restatement): Hyperscaler cloud FCF estimates too low; compute repricing at contract end can remove hundreds of billions of expected credit need.
  3. Third order (Inference): If true, AI-capex credit-supply scare (hyperscaler issuance flooding credit) is partially self-correcting via pricing power — bullish for equity duration of hyperscalers / NVDA ecosystem, less bullish for “spread blowout from AI debt” bears. Falsifier: rental crash on Rubin ramp; utilization air-pocket; contract repricing fails. External check needed on actual rental indices and hyperscaler FCF bridges. Do not import Baker memo figures.

Chain D — Cyber degradation → trust premium → crypto/tokenization

  1. First order (Source): Escalating AI-enabled hacking (UK power, med-device earnings cuts, Anthropic disclosures, Navy targeting).
  2. Second order (Source): Smaller firms hurt; earnings/productivity of AI complex continue; trust becomes scarce economic asset.
  3. Third order (Source/Inference): Ethereum / on-chain settlement / stablecoins / tokenized RWAs gain “trust” bid; Robinhood-like agents×TradFi×chain distribution captures flows. Clarity <20% this year still leaves yes-vote and 2027 Kraken/Nasdaq tokenization path as catalysts. Falsifier: major L1 failure; app-layer hacks destroy retail trust faster than chain trust builds; Clarity dies and offshore tokenization stalls.

Chain E — Government as “house” → don’t fade AI, don’t fight long-end policy

  1. First order (Source): Bessent $6bn buybacks, Plan B cut long issuance, Japan MOF coordination, China-AI-race framing, Warsh in orbit of not wanting higher long rates.
  2. Second order (Source): Fighting the government on bonds is the wrong trade; fighting the AI trade is the wrong trade; need not be long bonds.
  3. Third order (Inference): Policy put under long-end + industrial policy put under AI capex = asymmetric support for exponential book vs linear mean-reversion shorts. Risk: policy credibility break (buybacks fail and issuance cut fails and hike path forces recession) — then linear world returns with a vengeance. Horizon: months for buyback efficacy; years for AI-race industrial policy.

SCENARIO FRAMEWORK

Probabilities below are analyst inference for research prioritization, not forecasts the speaker assigned, and not desk allocations.

Bull (Inference ~25–30%) — Exponential dominance confirmed

Path: Sep 16 FOMC hike theater resolves without equity drawdown; HYG/IEF stays constructive; H100/compute rents remain firm into Rubin; Astra/agent KPIs keep printing; ETH/BTC relative trend extends; Clarity yes-vote or strong institutional tokenization path; Bessent Plan A/B caps long-end.

Market expression (research hypotheses, not recs): AI infrastructure sleeve and agent platforms outperform; hyperscaler FCF revisions up / credit need narratives fade; RH / tokenization names / ETH beta capture TradFi bridge flows; linear oil-bonds-Fed overlays remain noise.

What would need to be true: GPU rental data confirms +22% path; revision breadth stays positive; cyber incidents don’t cascade into systemic outage that kills risk appetite.

Base (Inference ~45–50%) — Grind + digestion + multiple compression

Path: SPX continues flag/grind above 50-day unless earnings crack; AI sleeve works but with high day-count of flat/down and a few gap days that make the year; hike odds volatile around Warsh/Bessent; crypto attention stays elevated with ETH/BTC choppy leadership; Clarity stays <20% this year but 2027 tokenization prep continues; GPU rents stabilize off highs but don’t collapse.

Market expression: Factor returns favor selection + timing inside AI universe (his ATR states) over passive Mag7; PEG stays “cheap” because earnings grow into multiples; duration shortening shows up as software multiple air-pockets without ending the AI-capex cycle.

What would need to be true: No deep growth scare; credit thermometers stay calm; capacity shortage persists enough to support compute pricing.

Bear (Inference ~20–25%) — Linear world reasserts / AGI narrative overruns tape

Path: Joint break in bond vol + junk + BE; earnings revisions roll; GPU rents crack on oversupply/Rubin transition; Astra/AGI claims walk back or regulation shock from cyber (“kill people” path); China AI race framing flips to capital controls / multiple compression without TAM payoff; ETH/BTC relative fails and crypto risk-off hits RH nexus.

Market expression: Linear checklist (oil, bonds, Fed) suddenly does matter; AI book correlates to 1 like 2022; hyperscaler credit need returns as FCF disappoints; tokenization narrative freezes.

What would need to be true: Growth air-pocket plus policy put failure; or model/compute pricing collapse that hits Layer 3–4 simultaneously.

Invalidation markers for his map (Inference from Source falsifiers): (i) SPX loses 50-day with HYG/IEF rolling over and bond vol up; (ii) H100 rents trend down hard YoY; (iii) revisions go negative while margins peak; (iv) ETH/BTC loses the “bull market tell” breakout and stays broken; (v) Bessent tools visibly fail and long-end disorder becomes equity-relevant.


COMPANY/ASSET WATCHLIST

No buy/sell. No target weights. Monitoring list only — names appear because the source discussed them.

Cluster Names / tickers as spoken Why on the list (Source) Diligence hook
Macro thermometers SPX, HYG, IEF, TIPS/BE, crude, Fed hike odds Linear checklist vs “don’t care” tape Confirm post-CPI/PPI path into Sep 16
Policy / official Bessent (buybacks $6bn; Plan B issuance), Warsh, Japan MOF “House” / don’t fight government Buyback size efficacy; long-end reaction
AI platforms Astra / OpenAI, Anthropic (econ team + hacking disclosures), Grok/xAI (UI compare) AGI/agent claims; cyber Verify Astra product identity; agent KPIs
Semis / compute NVIDIA (H100, Vera Rubin, “cheap PE”), Micron (multiple compression / not another 5-bagger path) Capex / duration / rental GPU rental indices; Rubin ramp
Hyperscalers Microsoft (triple DC compute; capacity shortage) Demand / FCF / credit need Capex guides vs FCF; compute ASP
Credit / depreciation debate “Hyperscalers” collectively; Burry/Chanos as bear foils; Gavin Baker as bull foil H100 +22% m/m; hundreds of bn credit Primary rental data; Baker primary source
Crypto L1 / relative BTC (flag/range), ETH (vs BTC breakout; trust model), Solana (fee comparison) Bull-market tell; valuation frame ETH/BTC; fee market share
Distribution / nexus Robinhood (chain fees ~$6m/d; agentic; Goldman takeaways), Arbitrum, Crypto.com (deal), Moomoo, IBKR AI×crypto×TradFi Fee sustainability; agentic AUM; chain TVL
Tokenization / rails Nasdaq–Kraken ($100m; 2027), US Bank SC, Visa SC cards, EU banks euro SC on ETH Bridge narrative Pilot → volume; Clarity path
Policy / events Clarity Act (<20% this year; Armstrong / Patrick Witt), Freedom Tech DC Sep 22–23, Bitcoin Policy Institute / PubKey Catalyst calendar Vote math; DC read-through
Soft targets / cyber Boston Scientific, Stryker, UK power plant, US Navy (as targets in anecdotes) Trust / crypto co-thesis Incident verification; earnings revisions
Commentary foils Paul Tudor Jones (AI market skepticism), Greg Jensen (regulation trigger), Stan Druckenmiller / DeMark (method refs), William Mougayar (ETH valuation), Sundar Pichai (orchestrate agents / 2027), Jensen Huang Narrative map Primary docs vs ASR

Explicitly not a recommendation list. Appearance ≠ endorsement.


DILIGENCE QUESTIONS & RESEARCH AGENDA

Tape / macro (this week → FOMC)

  1. Exact core CPI / PPI prints and components vs consensus (External check needed — speaker says “higher than expected” without full table).
  2. Live Fed funds futures: path of the 65→85/90% hike-odds claim into Wed Sep 16.
  3. Reproduce HYG/IEF, bond vol, junk spreads, TIPS BE charts for the two post-print days.
  4. Crude “through levels” — which levels? War “worse” — which theater? (External).
  5. Bessent buyback: confirm $6bn upsize and auction/settlement detail; market reaction intraday.

AGI / Astra / agents

  1. What is Astra in product terms (OpenAI agentic system as framed)? Primary Jensen Huang / Greg Brockman quotes vs ASR paraphrase.
  2. Source for ~100,000 GPU training claim; 4× next model; Vera Rubin schedule.
  3. Source for 3.1 workdays per human workday and 10,000 agents / 88 hours Navier–Stokes result — paper, blog, or vendor claim?
  4. Anthropic economics team papers he prefers over GS/MS rate notes — retrieve and summarize without importing other-desk numbers.

Compute / depreciation / Baker

  1. Independent H100 rental time series: level vs 3y ago; +22% m/m confirmation (which venue: Vast, San Flamenco, cloud spot, etc.?).
  2. Primary Gavin Baker remarks (~1 month prior) on GPU rents / hyperscaler FCF / credit need — read primary; do not paste numbers from prior Baker memos.
  3. MSFT “triple compute” — 10-K/guide language vs slide ASR.
  4. Stress: what rent path would restore Burry/Chanos depreciation bear case?

Crypto / RH / tokenization

  1. Robinhood chain daily fees: confirm ~$6m and Bernstein ~31% upside note.
  2. Arbitrum move on RH chain news — magnitude and persistence.
  3. Nasdaq $100m Kraken stake / 2027 tokenized stocks — primary filing.
  4. Clarity Act: current legislative status vs <20% this-year odds; Armstrong / Witt remarks.
  5. ETH from lows ~3,000% and Amazon ~27,000% (1997–2013) — verify windows; treat as analogy, not valuation math.
  6. Mougayar TED + “Ethereum endgame” + Trust Shift — listen/read; extract valuation framework.

Portfolio process (research of his method)

  1. Define the ~100-name AI universe and ~10-name concentrated sleeve constituents if disclosed to subscribers (not assumed here).
  2. Replicate ATR state labels: structural / reset / coiled / ignition — backtest hygiene (his own point: backtests should fail often).
  3. Agentic broker APIs: Robinhood / Moomoo / IBKR paper-trade constraints and data quality claims vs Yahoo.

Companion / External (no silent merge)

  1. Prior Jordi shows on “AI capex not rate sensitive,” bond trap, AI–crypto nexus — list as companions only.
  2. Hunt / Nordvig / Green / Alma Bessent memos — regime contrasts without importing their figures into this memo’s underwriting.

RISK ANALYSIS

Source / process risks

Market / thesis risks

Desk product risks

Risk map summary (Inference)

Risk Likelihood (near term) Severity if hit Mitigant in his map
Credit/vol finally cares about CPI/Fed Medium High Watch HYG/IEF + BE as overlays
GPU rent fade Medium High for FCF/credit story Track primary rental indices
AGI/Astra claim deflation Medium Medium–High Demand primary docs
Bessent tools fail Low–Medium High Plan B already named; still political
Cyber cascade Medium Medium (asymetric to small caps) Trust/crypto co-thesis — may not hedge equity beta
Clarity delay High (odds <20%) Medium for tokenization timing 2027 Kraken/Nasdaq path

APPENDIX — Unverifiable / provisional numbers (flag for External check)

All speaker-stated; not verified by this desk from primary data in this memo:

Claim As spoken Status
Fed hike odds ~90%; also 65%→85% External check (futures)
Bessent buybacks $6bn / op External check
S&P over 2 days “up slightly” External check
Revisions +32 this week External check
Barely up since Jun 2 External check
Astra training scale ~100,000 GPUs; next 4× External check
Agent productivity 3.1 workdays / human workday External check
Navier–Stokes swarm 10,000 agents / 88 hours External check
Digital labor multiple 3–5× Conceptual / External
PEG Cheapest in ~30+ years External check
H100 rents Unchanged vs ~3y; +22% m/m External check
Baker credit relief “Hundreds of billions” Restatement only; External / primary Baker
MSFT compute Triple DC power External check
Astra “views” 6.2 million Unclear metric; External
Demo account $500 → $5,100 Anecdote; not underwritable
Idea rejection ~99% then further 90% Process claim
Micron path talk “at a,000”; 60% / not 5-bagger Illustrative; External
AI universe ~100 names; ~10-name sleeve Constituents not listed here
Crypto attention ~30% of his focus Attention ≠ AUM
RH chain fees ~$6m daily External check
Bernstein RH ~31% upside / fees top SOL External check
Arbitrum ~7.87% surge External check
Nasdaq–Kraken $100m; 2027 tokenized stocks External check
Clarity Act <20% this year Handicap; External
Amazon return 1997→2013 ~27,000% External check window
ETH off lows ~3,000% External check window
Amazon PE<100 Two quarters ~2017–18 External check
Brockman interview ~24 minutes External check
GitHub accounting hits ≥13 projects Soft claim
Freedom Tech DC Sep 22–23, 2026 Calendar confirm

CLOSING (for packet routing)

Stance one-liner (Source expression, not desk advice): Linear obsessors (oil, bonds, Fed hike theater) are fading the wrong regime — AGI/Astra + AI capex + credit saying “don’t care” is the tape; don’t fight AI or the government; lean exponential (AI book, ETH / Robinhood / tokenization) while multiples compress and duration shortens.

This memo: transcript + brief only · ASR caveats locked · companions External without number merge · no buy/sell · markdown file only · not published to here.now.


End of memo.

Desk copy · not a trade recommendation · Erica · 13 Sep 2026