Stock-market dispatch — 2026-07-30

Headless run (Thu). Research output, not advice — brain never trades. Marks are live 07-30 intraday twelvedata quotes; 48/48 symbols priced, 0 429s. Step 4d (weekly ledger) not due (Fri cadence; last mark 07-24). Supabase sync + push are performed by the wrapper, not this run.

Top of mind

The "punish the spender" tell sharpened into a free-cash-flow discriminator, and the tape obeyed it. On the same night (07-29 post-close), MSFT beat and rose ~8% — it held capex steady and stays FCF-positive (Amy Hood) — while META beat revenue +28% but fell ~9.6%, having raised 2026 capex to $125–145B as FCF collapsed to $784M (2026-07-30-autoresearch-megacap-q2-capex-reaction-discriminates). Two beats, opposite reactions: the market now rewards disciplined spenders and fades the FCF-negative raiser — a refinement, not a reversal, of ai-roi-reckoning. The credit market echoes it exactly: S&P downgraded Oracle to BBB- (July) on capex 2× cash flow, while MSFT's rating is untouched — the first realized instance of mega-issuance-peak-to-ai-capex-derate's falsifier, now re-keyed to issuer-level FCF-negative names. Today's tape was the FCF-discrimination in action: VRT −17.3%, MU −9.9%, CBRS −12.1%, ARM/STRL/MP/RKLB −8%, TSM/NVDA/DELL/INTC down, while FCF-positive NOW +4.7% rose a 4th session against the red. The book's top names (MU/TSM asset-heavy compute, NOW FCF-positive SaaS, LIN/CCJ non-AI-capex) sit on the right side of that tell.

Strongest-conviction buys

Ranked shortlist (4c). Prices 07-30 intraday, twelvedata.

#	Ticker	Causal chain (mechanism)	Conviction	Valuation vs base	Fundamentals (one line)	Next catalyst
1	MU	hbm-cowos-as-binding-bottleneck	High (0.80)	Yes — $739.00 (−9.9%, −41% off hi)	~5–6× fwd, HBM booked thru CY2027; the deepening semis rout = better entry on the asset-heavy leg the FCF-discrimination rewards	HBM4 ramp / SK Hynix reads
2	CCJ	ai-capex-to-power-and-materials-cascade + kazatomprom-supply-cut-to-western-uranium-premium	Med-high (0.73/0.60)	Yes — $84.57 (−2.8%, −37% off hi)	Uranium premium intact — DNN/NXE legs beat sector (+6.6%/+8.4%); CCJ the liquid anchor	Cameco Q2 (07-31, tmrw)
3	NOW	agentic-ai-seat-erosion-to-saas-rerate	Med-high (0.62)	Yes — $115.76 (+4.7%, −42% off hi)	Q2 beat+raise held; base $140–150 (~+25–30%); 4th up-session against a red tape = the gap-close the thesis flagged; FCF-positive	Q3 print; ACV conversion
4	TSM	hbm-cowos-as-binding-bottleneck	Med-high (0.70)	At base — $374.67 (−4.5%, −22% off hi)	18–19× fwd; CoWoS booked thru 2027; +custom-ASIC broadening	N2 ramp
5	LIN	ras-laffan-halt-to-lin-helium-pricing-power	High (0.78)	Marginal — $511.17 (flat, −7% off hi)	30%+ margin; caught the defensive bid; near target	Helium contract reads

⚠ LLY (0.60) — $1210.02 (−0.9%), near 52w hi → priced-for-perfection, watchlist. EQT (0.55) — $52.57 (+1.8%) → gas-floor is 2027-forward, watchlist. SLB (0.52) — $48.96 (−2.0%) → watchlist.

Watchlist

Ticker	Thesis	Conviction	Price	Undervalued vs base?	Next catalyst
MU	hbm-cowos-as-binding-bottleneck	High	$739.00 (−9.9%)	Yes	HBM4 ramp
CCJ	kazatomprom-supply-cut-to-western-uranium-premium	Med-high	$84.57	Yes	Q2 07-31
NOW	agentic-ai-seat-erosion-to-saas-rerate	Med-high	$115.76 (+4.7%)	Yes	Q3 print
TSM	hbm-cowos-as-binding-bottleneck	Med-high	$374.67 (−4.5%)	At base	N2 ramp
EQT	ai-gas-demand-to-appalachian-producer-price-floor	Medium	$52.57 (+1.8%)	Marginal	2027 floor
INTC	tsmc-saturation-to-intel-anchor-stack	Medium (0.43)	$81.88 (−5.1%)	⚠ external-anchor unproven	14A decisions H2'26
SLB	slb-oilfield-pivot-to-dc-infrastructure	Medium	$48.96 (−2.0%)	No	Q3 DC run-rate

Undervalued candidates (today)

hbm-cowos-as-binding-bottleneck — MU $739.00, −9.9% in a deepening AI-infra rout, −41% off high; the FCF-discrimination rewards asset-heavy consumption compute, and the supply-constraint leg (HBM booked thru CY2027) keeps getting corroboration — the rout is the entry, not the thesis breaking.
agentic-ai-seat-erosion-to-saas-rerate — NOW $115.76, +4.7% and rising a 4th session against a red tape, −42% off high, base $140–150 — the market continuing to close the seat-erosion gap; FCF-positive, the trait today's tape rewards.

New chains to investigate (hypothesis-stage)

datacenter-labor-scarcity-to-modular-vendor-content-shift — VRT / FIX / STRL / PWR. From SemiAnalysis (partial): the skilled-trade (electrician/pipefitter) labor shortage projected for 2027 forces datacenter construction into factory modularization (63% of MEP off-site, −85% field electricians, −36% timeline) → content/margin shift to modular power/white-space vendors (VRT ~$3.5M→$7M/MW) and factory-MEP/EPC integrators (FIX/Comfort Systems, STRL ~$6B target). Thin industrials/construction vertical (breadth-2a steer). ⚠ Step 3 single-sourced — but note VRT −17.3% today (crushed into any modular re-rate). To graduate: a VRT/FIX/STRL disclosure tying revenue-mix to modular. → /explore-chain. priority medium.

New theses (now active)

None. Corroboration/refresh + one contradiction day; net-new steered to a thin-vertical hypothesis (per breadth 2a: ai-infra 47% ⚠ over).

Updated theses (existing active)

ai-roi-reckoning — the reaction refines to an FCF-discipline discriminator. MSFT (held capex, FCF+) +8% vs META (raised capex, FCF→$784M) −9.6% on the same night → reward disciplined spenders, fade FCF-negative raisers. as_of→07-30.
mega-issuance-peak-to-ai-capex-derate — first realized credit instance: S&P cut Oracle to BBB- (July) on capex/cash-flow; falsifier re-keyed to issuer-level FCF-negative spread/rating moves, not sector aggregate. as_of→07-30.
open-weight-sputnik-to-frontier-lab-derate — alex-karp names the engine: Nvidia "commoditizes its complement" (the model layer) via open weights → value routes to silicon; innovator's-dilemma pressure on Anthropic/OpenAI. as_of→07-30.
humanoid-actuator-magnet-demand-to-us-ree-stack — NEO ships 2026 (bernt-bornich/1X) = dated Western commercial-volume print; OEMs (amanda-mcmaster/Boston Dynamics) demand US/allied actuator+magnet reshoring → tailwind for MP/USAR. Tonnage gap unchanged.

Contradicted / weakened

oil-roundtrip-to-staples-margin-tailwind — CONTRADICTED by its dated catalyst. P&G Q4 (07-29) guided a ~$1B FY2027 cost headwind (Brent-$90 + freight/trucking/supplier), pricing "neutral," op margin −130bps — not the reversal tailwind. hypothesis → weakened; 1 calibration logged (narrow-framing/reversal-symmetry-assumption — assumed symmetric cost reversal, under-weighted non-oil input costs). PG entity + andre-schulten updated.
4e gate: 23 scored rows; 4 flags <−10% sector (PWR −22.5%, CAT −12.4%, UUUU −11.4%, TJX −10.3%) but all medium/partial tier → no gate fires. 0 new live-feedback calibrations. Kazatomprom mechanism fine (DNN/NXE legs beat).

Open questions worth a human's eye

Does AMZN/AAPL (07-30 pm) extend the FCF-discrimination rule? Amazon's capex is expected to exceed operating cash flow in 2026 (rule → punish any raise); Apple isn't an infra-capex story (rule → doesn't apply). Tonight's prints are the out-of-sample test.
Is the semis rout (MU −9.9%, TSM/NVDA down) the ROI-reckoning finally hitting suppliers (the "last car off the cliff"), or an FCF-negative-name flush that spares the asset-heavy consumption leg? The distinction is the whole MU thesis.
Did VRT's −17.3% come with modular-content commentary? The one datapoint that would move datacenter-labor-scarcity-to-modular-vendor-content-shift from ⚠ unverified to partial.

What I looked at

Run status: headless; steps 0.5–7 in-turn; price fetch paced 8/min — 48/48 priced, 0 429s. Step 4d not due (Fri cadence). Per-concept bull/base/bear narratives carried forward (headless budget); load-bearing marks in the feed + tables.
Breadth (2a): ai-infrastructure 47% ⚠ over (43 distinct chains) → net-new steered to the thin industrials/construction vertical (datacenter-labor→modular). Macro sourcing came via the ingestion veins (robotics/industrials, construction-labor, consumer staples), not extra bucket autoresearch.
Ingestion veins (0.5–0.7): podcast 3 (All-In robotics CEOs; Moonshots Dario-vs-Jensen open-weights; Capital Allocators real-estate — diarized AssemblyAI, ~$0.67); earnings 1 (PG Q4 FY2026; CCJ/LIN 07-31, PWR 07-30, LLY 08-05 not yet out); feeds 1 (SemiAnalysis "LEGO Datacenters", partial/paywalled; FK still paywalled, CP roundups, Apricitas dormant).
Steps 1–2b: 1 autoresearch (megacap Q2 reaction → FCF-discrimination). prospect-chains: no new auto-draft from accumulated wiki (well-connected; today's novel chains arrived via clippings → materialized in ingest 4a).
Sources: 6 promoted + 6 ingested → 3 created (sterling-infrastructure, bernt-bornich, amanda-mcmaster) + 1 hypothesis; ~6 updated (ai-roi-reckoning, mega-issuance-peak, open-weight-sputnik, humanoid-actuator, oil-roundtrip[weakened], procter-gamble; +quotes schulten/bornich/mcmaster/karp). 0 new mechanisms (1 chain hypothesized, 2 strengthened). 1 zero-novelty (Capital Allocators real-estate — over-included).
Live feedback (4e): 23 scored rows; 4 flags <−10% sector all medium/partial → no gate. 0 new auto-calibrations (1 calibration from the PG contradiction, step 5). Best: CF/CBAM +9.0%, NXE +8.4%, DNN +6.6%, NOW +5.3%, STVN +4.9%.
Signal feed (7): full active/armed set (see below). Prices re-marked 07-30; as_of re-dated only for rows whose research moved today.

stock-market