• Type: thesis-shift (subtype: sharpens) — Hypothesis A "what gets sold" granularity revised; first sell-the-news confirmation in the Hypothesis C cohort
• Source: S-1/A full read + picks-and-shovels supply-chain map (2026-06-04-spacex-s1-full-read-section-notes, 2026-06-04-spacex-s1-picks-and-shovels-supply-chain-map) + code-computed breadth across 8 SaaS/AI precompute summaries (summaries{semis,memory,cloud-saas,mag7,ai-infra,ai-scan,web-cloud,tech}.json, 6/3 close) + spacex-s1-supply-chain.json` screen (6/4 intraday).

User question that drove this: "we should expect the spacex ipo to suck some liquidity out of the saas / ai sectors which really have rallied a lot and are way above sma at this point." The instinct is sound on the mechanism; the data refines the target.

1. Deal-size firms at the upper end → the liquidity-sink premise is real. S-1/A carries 555,555,555 Class A @ assumed $135 ≈ $75B, all primary (upper end of the prior $40-80B range) — the largest IPO ever, ~3× Aramco's $25.6B, and all-primary means genuinely net-new equity supply, not a rotation of existing shares. Musk voting 82.4% (82.3% w/ greenshoe) at the priced cover — refines the earlier S-1 "85.1%" figure. This reinforces, not replaces, the Fast Entry forced-buyer mechanic from the 2026-05-26 deep-dive: $75B of new supply + >$600B NDX-tracking forced buyers at ~T+15 both have to be sourced from somewhere.

2. "What gets sold" — REVISED with breadth data. It's AI-infra silicon, not broad "SaaS/AI." The crowded, way-above-SMA cohort that is the real funding-source risk is the semis / memory silicon layer, NOT SaaS apps:

• semis (13 names): 13/13 above SMA50 and 13/13 above SMA200, median RSI 73.1, median 3M +68.7%, sitting −3.2% off 52wk highs (6/13 within 3% of highs). Maximum extension, minimum cushion — sold first to "make room."
• memory (7 names): median RSI 79.4, median 3M +127.5%, −2.2% off highs. The single most-stretched cohort.
• cloud-saas / SaaS apps (13 names): the opposite — only 5/13 above SMA200, median 3M +3.5%, −39.2% off 52wk highs, 0/13 near highs. SaaS apps were already the source-of-funds for the AI trade over the past year; there's no fat gain left to harvest. They are not the drain target.
• mag7: middling — median RSI 51.2, −12.1% off highs. Not the stretched part either.
• Net: this adds the silicon layer (semis/memory) to the prior Hypothesis-A "what gets sold" list (CRWV/NBIS/IREN/APLD/CORZ from the 2026-05-26 entry) and explicitly rules out broad SaaS. The funding-source pressure is reflexively aimed at the most-extended longs, which are AI-infra silicon.

3. First sell-the-news confirmation — in the Hypothesis C (halo) cohort, exactly where the 5/29 refresh said the risk concentrated. The small-cap space complex, which ran +55–136% over 3M into the IPO window, took −11% to −24% over the last 7 days (RDW −23%, FLY −24%, LUNR −22%, RKLB −20%, BKSY −24%, VSAT −21%, ASTS −13%, PL −11%) while the supply-chain picks-and-shovels held green over the same 7 days (MU +16%, RMBS +15%, AVGO +14%, COHR +10%, SEI +7%, ALAB +7%). Diversified / non-competing space adjacents also held (IRDM −4%, GSAT −3%, SATS −5%). Read: the halo run-up is largely spent and the sell-the-news unwind has started in the most speculative, most-directly-competing names first — the textbook post-IPO mean-reversion the 2026-05-26 entry flagged (RKLB/ASTS at 60–450× P/S). This is Hypothesis C resolving toward "halo, then air out," with dispersion-by-relationship: pure competitors hit hardest, picks-and-shovels untouched.

Disciplined counterweight (keeps Hypothesis C alive at the index level): the drain has an opponent — halo inflow + record money-market cash. A chunk of the $75B can be funded from sidelined cash without selling a single AI share, and a hot AI+space IPO can pull momentum money toward the complex short-term. The net sign on the AI-infra cohort is a tug-of-war, so the higher-conviction read is dispersion, not direction: whether or not the complex falls outright, the most-extended semis/memory names carry the most "make-room" downside; already-corrected SaaS and middling mag7 have far less to give.

New watch trigger (stood up 2026-06-04 in TASKS-SCANS.md + README Watch For): the drain-beats-halo tell is semis/memory rolling below SMA20 in the post-IPO window (debut 2026-06-12 through T+14 ~2026-06-26). If the extended silicon cohort breaks SMA20 while SPCX trades, Hypothesis A is winning the tug-of-war; if it holds or melts up through the window, halo/inflow won and Hypothesis A stays unconfirmed at the cohort level. Pairs with the existing AI-infra-cohort-weakness watch (CRWV/NBIS/NVDA/CORZ/IREN).

Discipline note: breadth is a 6/3 close snapshot; the space read is a 6/4 intraday snapshot. These are observations on the existing book, not buy/sell calls — consistent with the standing "we are NOT trading this IPO" frame.