←所有簡報

DISCIPLES OF CLAUDE

AI × 市場 · 會議簡報
期號
60秒摘要
  • Cash goes short, then gold: Ethan's plan is 0–2 year T-bills as the cash equivalent now, rotating into gold when the Fed starts cutting; Charlotte called it good points. No date was put on the rotation.
  • Bonds are a sell, not a buy: long duration, junk and investment-grade corporates were all marked dangerous — hyperscaler data-center debt and widening spreads were the reason (Ethan); nobody argued the other side.
  • Energy is not a sector trade: only self-sustaining sources — nuclear/uranium and geothermal — that already hold the infrastructure and contracts; CEG is the room's aligned long on locked-in power agreements. The Saudi pipeline strike was called not yet priced in.
  • Two shorts named, both rate-driven: KRE (regional banks, commercial real estate) and TCPC (private credit); Charlotte agreed. TSM was flagged as dangerous short term even by a long-term bull.
  • The "safety" pause is the tell: Ethan expects the frontier labs to slow scaling under a safety banner rather than admit a power ceiling; an admission would hit the whole Nasdaq. Bullish long term, bearish short term.
本場數據
3Present
—Minutes
18Names discussed
6Commitments
房間

What the room argued about

Three attended; the night moved from a cash plan for the rate shock through energy, chips and two shorts to a decision model that outputs only a verdict and a confidence.

01

Macro and bond market

  • The week's market drop was driven solely by the market re-pricing the rate hike; the underlying energy thesis is unaffected (Ethan).
  • Cash plan: hold reserves in short-duration 0-2 year T-bills (low duration risk) as a cash-equivalent; rotate out of cash/money-market funds into gold once the Fed starts cutting, since gold is expected to shoot up right as cuts begin (Ethan; Charlotte: "good points").
  • Long-duration bonds are dangerous given the energy crisis and physical grid limits; high-yield junk bonds have already collapsed; investment-grade corporate bonds are at risk because hyperscalers rely on this financing for data-center debt, and spreads will widen as the Anthropic/OpenAI scaling-stagnation story plays out (Ethan).
  • Debt rollovers at higher 10-year yields will compress margins across the board; the room is presently only selling, not buying, long-duration bonds, and is avoiding companies with high debt issuance and floating rates in this environment (Ethan).
02

AI frontier labs: IPO delays and the "safety" framing

  • Ethan's read: OpenAI and Anthropic are delaying their IPOs to avoid filing an S-1 that would expose unit economics, hardware depreciation, and revenue growth slowing under the energy crisis; Ryan separately referenced reporting on Anthropic's profit-calculation methodology omitting major expenses, and agreed the companies "have no morals."
  • If either lab admitted publicly to a scaling/power ceiling, Ethan expects a broad Nasdaq selloff since every AI name would be affected; he expects labs to instead frame slower scaling as a "safety" pause rather than admit the constraint. Ryan agreed this tracks with prior discussion.
  • Ethan remains bullish on AI long-term but bearish short-term on the physical and financial case for continued scaling at the current pace.
03

Energy and geopolitical disruption

  • Ethan does not treat energy as a single buyable sector; he favors only self-sustaining sources that don't lean heavily on water, natural gas or grid electricity - specifically nuclear/uranium and geothermal - as the names that win while more grid capacity gets built. The winners will be companies that already hold the physical infrastructure and contracts.
  • Drone strikes launched from Iraq destroyed a petroleum pipeline in Saudi Arabia (the route meant to bypass the Strait of Hormuz), sending crude gapping up; Saudi Aramco is cutting its October deliveries to Europe and India as a result (Ethan). Ethan called this not yet fully priced in, since the market narrative is still bullish on AI.
  • Ryan raised rumors of a possible Russian move against a NATO member as a way to exit the Ukraine war without admitting defeat, citing speculative scenarios (including from asking Gemini) rather than a settled read; Ethan said he doesn't know the Russia/Ukraine geopolitics well enough to weigh in.
  • Ethan: the clearer trigger for a wider escalation would be an attack on a US data center, which is part of why he rates defense - specifically companies building lidar/laser surveillance technology for data centers - above energy as a long-term holding. He also flagged Russia's global uranium position as a possible future lever if US nuclear buildout raises uranium demand.
04

Chips and hardware supply chain

  • Ethan: Nvidia is a "later order of impact" name - any slowdown starts with smaller companies and is weakened by the time it reaches Nvidia, so a sharp Nvidia drop would itself mark the point the AI bubble bursts rather than an early warning.
  • TSM (TSMC) is in a "dangerous short-term position" as frontier labs pause scaling and revise hardware orders down, even though Ethan is bullish on TSM long term; Charlotte was surprised, having expected him to stay bullish.
  • MediaTek was raised as a comparatively safer chip name despite carrying high debt, because it benefits from demand for GPU-free training of open-source models (Ethan).
05

Market opportunities and shorts

  • Volatility-fee beneficiaries: CME and ICE, which collect transaction fees regardless of direction as futures/options volume spikes (Ethan; Charlotte agreed and named ICE herself).
  • CEG (Constellation Energy): both Ethan and Charlotte are bullish - its locked-in power purchase agreements throw off free cash flow that offsets debt regardless of hyperscaler regulatory slowdown; Ethan said most of his current options (including here) expire January 2028, but his plan is to sell as soon as a position gaps up rather than hold to expiry.
  • Shorting candidates: KRE (regional-bank ETF, exposed to commercial real estate) and TCPC (BlackRock's private-credit vehicle) - both seen as unable to keep lending profitably as yields rise (Ethan; Charlotte agreed).
  • Software names Ethan is not bullish on: Salesforce, Snowflake and CrowdStrike (Charlotte pushed back mildly, saying she "kind of likes Salesforce"); also not bullish on AI-rally names he sees as propped up mainly by government support - Nokia, BlackBerry and Dell - questioning how long that support holds.
  • Palantir: Ethan is bullish short-term on an uncapped DoD budget and an AI-regulatory-sovereignty role, but not long-term.
  • Booz Allen Hamilton, alongside a smaller private auditing firm Ethan referred to as "Cassie" (Consolidated Analysis Center, no public ticker): Ethan expects both to win AI-regulation auditing work given Booz Allen's existing NSA/CIA/DoD clearance base, with a risk Booz Allen tries to monopolize that role as the sector's de facto regulator; Ryan and Charlotte agreed with the logic.
06

JV decision model and AI orchestration

  • Ethan described a model he calls "JV" as roughly 20x faster and 40x cheaper than a normal LLM because it strips out the reasoning layer entirely and only outputs a decision plus a computed confidence interval - cutting hallucination risk since it can't invent unsupported context, which Ethan sees as useful for programmatic decision points inside an AI pipeline.
  • Proposed combined workflow: route anything under a chosen confidence threshold (Ethan floated 90-95%) back through an LLM for reasoning, then back to JV to calibrate; the exact threshold was not settled.
  • Ethan framed JV as a calibrated form of reinforcement learning based on binary classification, distinct from RLHF; Charlotte linked the reasoning/decision split to Kahneman's "Thinking, Fast and Slow" (System 1 vs. System 2) and recommended the book to Ryan.
  • Charlotte proposed also using JV as a hallucination check: feed an LLM and JV the same information, and if the LLM's answer diverges from JV's across repeated tests, treat the LLM as the one hallucinating; Ethan agreed this could work as a safety check.
  • Personal-agent tools: Charlotte recommended the Meta Muse personal-assistant app (she uses it for tasks like reviewing statements and making calls) and offered to send Ryan her referral code. Ethan said he doesn't use it because it isn't personalized enough for his own workflow, and that he built his own agent setup instead.
07

Platform build and calendar feature

  • Ethan demoed a new feature on the group's tracking site: a company-events calendar that auto-populates from every ticker the group adds, with category filters and export to Apple/Google Calendar. Charlotte added Chinese-language translations to parts of the site; financial terms need a custom embedding model rather than Google Translate to translate accurately. Ethan asked the group to keep sendin
08

Google and Apple

  • Google: Charlotte thinks it's undervalued and plans to research the thesis further; Ethan pointed to Google's TPU-training-factory buildout with a member's employer as a demand signal, but said he doesn't yet have a strong bullish thesis of his own on Google, and flagged weak consumer-hardware/wearables demand in the current economy.
  • Apple: both Ethan and Ryan are bearish. Ethan's case - Apple is a luxury good whose sales fall as consumer credit tightens; Apple Intelligence is weak despite being the company's AI move; heavy reliance on Foxconn/TSM supply chains leaves it exposed to copper constraints; ecosystem lock-in (iMessage/FaceTime/Apple Pay), not superior hardware, is what retains users. He noted Apple's historically un
09

Smartphone hardware

  • Samsung's foldables were credited with zero-lag folding displays and strong hardware, undercut by heavy bloatware that Ethan says keeps the Snapdragon chip from running at full efficiency.
  • Recent iPhones switched from titanium to aluminum frames for better heat dissipation, at some cost to dent resistance; Ethan noted phone cases trap heat, so the material change is more forgiving for case users, and called it a fair performance/durability trade-off (Ryan agreed: "that's fair").
10

Workplace AI and phishing

  • Ethan said he has failed several recent corporate phishing tests at a member's employer, including a sophisticated AI-generated email framed as a year-end performance-review feedback request; he noted AI is making these tests harder to catch and joked about asking an AI assistant to vet suspicious messages before clicking.
11

Decisions (aligned)

  • Cash plan: hold reserves in short-duration T-bills now, rotate to gold once the Fed starts cutting (Ethan; Charlotte affirmed).
  • Avoid long-duration bonds, junk bonds, investment-grade corporate bonds and private-credit exposure in this environment (Ethan; Charlotte agreed throughout).
  • Energy is not a buy-the-sector trade; only self-sustaining sources (nuclear/uranium, geothermal) are favored (Ethan; Ryan: "that makes sense").
  • KRE and TCPC flagged as shorts on rate-driven commercial-real-estate and private-credit stress (Ethan; Charlotte: "I agree").
  • CME/ICE and CEG are the room's aligned longs on volatility fees and locked-in PPAs (Ethan and Charlotte both bullish).
  • The aluminum-over-titanium iPhone frame trade-off is a fair engineering trade-off, not a defect (Ethan; Ryan: "that's fair").
  • Forward commitments: Charlotte will send Ryan her Meta Muse referral code; Ryan will spend time this week researching the Russia/Ukraine/NATO situation; Charlotte will review the JV model's technical details and research whether Google is undervalued; the group will keep sending Ethan tickers for the calendar feature and will look at the mathematical representations behind JV versus other algorith
持倉

The watchlist

Watch18
CMEEthan, Charlotte - bullish - volatility-driven transaction-fee beneficiary regardless of market direction.未指名持有人未說明時限
ICEbullish - same volatility/transaction-fee logic as CME.Charlotte未說明時限
CEGEthan, Charlotte - bullish - locked-in PPAs generate free cash flow offsetting debt; Ethan's options here expire January 2028, sell plan is on any gap-up.未指名持有人未說明時限
KREbearish/short candidate - rising yields hurt commercial real estate and regional-bank lending.Ethan未說明時限
TCPCbearish/short candidate - rate-driven private-credit stress at BlackRock's vehicle.Ethan未說明時限
PLTRbullish short-term, bearish long-term - uncapped DoD budget and AI-regulatory-sovereignty role.Ethanshort-term
BAHbullish - favored to win AI-regulation auditing work on its existing government-clearance base; risk of monopolizing the role.Ethan未說明時限
NVDAresilient/last-order-impact - would only drop sharply at the point the AI bubble itself bursts.Ethan未說明時限
TSMbearish short-term, bullish long-term - hyperscaler order pause hits forward bookings.Ethanshort-term
CRWVobservation only - named by Charlotte as a data-center company example, no stance given.未指名持有人未說明時限
CRMEthan bearish, Charlotte mildly bullish ("I kind of like Salesforce") - disagreement noted, not resolved.未指名持有人未說明時限
SNOWbearish - grouped with CRM/CRWD as software names he is not bullish on.Ethan未說明時限
CRWDbearish - same grouping as CRM/SNOW.Ethan未說明時限
NOKbearish - AI-rally name Ethan sees as propped up mainly by government support.Ethan未說明時限
BBbearish - same government-support-dependent grouping as NOK/DELL.Ethan未說明時限
DELLbearish - same grouping.Ethan未說明時限
GOOGLCharlotte bullish (undervalued, researching further), Ethan neutral/uncertain - TPU-factory buildout with a member's employer as a demand signal, offset by weak consumer-hardware demand.未指名持有人未說明時限
AAPLEthan, Ryan - bearish on direction - luxury-good demand exposed to consumer-credit tightening, weak Apple Intelligence/Siri, Foxconn/TSM supply-chain and copper exposure; Ethan still likes it as an options-IV trade despite the bearish view.未指名持有人未說明時限
啟示

Name the mechanism, not the sector

The room agreed energy wins and still refused to buy it as a sector: only names that already own the infrastructure and the contracts qualify. The shorts passed the same test — KRE and TCPC were named because a specific mechanism, rate-driven rollover stress, breaks them, not because their sectors are out of favor.

未決問題

尚未定論的問題

Whether Booz Allen Hamilton (with "Cassie") actually wins, and then monopolizes, AI-regulation auditing work - Ethan's guess, not adjudicated.
Whether the Russia/NATO invasion rumor is a real threat or a way to exit the Ukraine war - Ryan's speculation, Ethan declined to weigh in citing limited geopolitical knowledge; Ryan to research further.
The JV/LLM confidence threshold for routing decisions back and forth (90% vs. 95%) - floated, not fixed.
Whether Google has a strong bullish thesis beyond the TPU buildout - Charlotte leans undervalued, Ethan says he can't articulate one yet but suspects one may exist.
Whether Russia would actually use its uranium-supply leverage or Strait of Hormuz position - raised as a hypothetical by Ethan, not resolved.
日程

值得關注的日期

跟進事項

誰負責什麼

Charlotte Zhou · 未定截止日期Share Refer Code: Send the Muse application referral code to Ryan.
The group · 未定截止日期Analyze Algorithms: Investigate the mathematical representations of each algorithm to understand their differences.
Charlotte Zhou · 未定截止日期Review Model Details: Examine the technical details of the JV model to assess its functionality and limitations.
Ryan Lee · 未定截止日期Research Geopolitics: Spend time this week researching the geopolitical situation regarding the Russia and Ukraine conflict and its potential impact on NATO members.
Charlotte Zhou · 未定截止日期Analyze Google: Conduct research into the investment thesis for Google to evaluate if the company is currently undervalued.
The group · 未定截止日期Provide Tickers: Submit additional stock tickers to Ethan for inclusion and tracking on the company events calendar.
未指名負責人 · 未定截止日期CME · State a horizonThe room named this position without a timeframe.
Charlotte · 未定截止日期ICE · State a horizonThe room named this position without a timeframe.
未指名負責人 · 未定截止日期CEG · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期KRE · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期TCPC · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期BAH · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期NVDA · State a horizonThe room named this position without a timeframe.
未指名負責人 · 未定截止日期CRWV · State a horizonThe room named this position without a timeframe.
未指名負責人 · 未定截止日期CRM · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期SNOW · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期CRWD · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期NOK · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期BB · State a horizonThe room named this position without a timeframe.
Ethan · 未定截止日期DELL · State a horizonThe room named this position without a timeframe.
未指名負責人 · 未定截止日期GOOGL · State a horizonThe room named this position without a timeframe.
未指名負責人 · 未定截止日期AAPL · State a horizonThe room named this position without a timeframe.
本週市場史
SEPTEMBER 20, 1929

The fraud surfaced only after the lenders said no

On September 20, 1929, Clarence Hatry confessed to fraud before Britain's Director of Public Prosecutions, and the London Stock Exchange suspended every share in his group, paper worth about £24 million. Montagu Norman's Bank of England had refused him a bridge loan for his $40 million United Steel merger, and he had pledged forged City of Wakefield stock to banks; the group owed £19 million against £4 million of assets. The Dow, which peaked at 381.17 on September 3, was 10% lower by month-end, and the real break came in late October.

Fraud hides inside a boom's easy credit and surfaces when lenders say no — the scandal usually marks the turn rather than causing it.
— C. LEBRUN