Research notes · September 9, 2026

Seven notes from six publishers: Meta Muse launch, Capital Wars maps a 5-6 year liquidity cycle, and a hedge fund trims its AI-chain shorts

On 2026-09-09 the tracked publishers put out seven pieces from six sources. Capital Wars ran a long interview with Michael Howell on global liquidity cycles. TMT Breakout published two notes: a morning wrap on Meta's new Muse AI agent and an end-of-day roundup of Goldman Sachs and Citi tech-conference recaps spanning dozens of companies. Damnang and SemiAnalysis each ran deep research pieces, on the AI memory-hierarchy standards fight and on-device versus datacenter robot inference. James Bulltard posted a personal trading-milestone update centered on Intel, and Cassandra Unchained posted a hedge-fund position update trimming both its long and short books. Every statement below is attributed to the publisher or sell-side desk that made it, in that source's own numbers; Quant Data did not verify any of it and takes no view.

What was published

"Points extracted" counts the discrete claims (numbers, theses, risks, catalysts, falsification points) pulled from each piece during ingestion. It measures how much a piece asserts, not how good it is.

What each piece said

Directional statements on the record

Each row is a dated fact: a named source expressed a direction on a named ticker. Where a publisher was relaying a broker's rating change, the originating voice is in the "via" column. None of this is a recommendation from Quant Data.

Numbers the publishers put on the record

• Capital Wars: Liquidity cycle length is about 5-6 years, tied to the average duration of world debt; about 80% of primary issuance refinances existing debt rather than funding new capital projects.
• Capital Wars: Term premium explains about 80% of Treasury long-end yield moves since 2000, versus a roughly 40% historical norm; liquidity is said to have peaked late 2025 and to reach the real economy 12-15 months later, with the yield curve typically starting to flatten about 9 months after a liquidity peak.
• TMT Breakout · META: Muse reached #4 on the App Store; shares closed up 4.5%; the author raised a risk-case valuation from about $550 to $600 (roughly 17x CY27 EPS of $36); Morgan Stanley estimates a $30 trillion consumer-agent TAM.
• TMT Breakout · DELL: Evercore models FY27 EPS above $30 (versus $25.88 consensus) and FY28 above $40 (versus about $31 consensus).
• TMT Breakout: Cantor's CY28 stretch EPS estimates: ADI $23 (vs. $19.86 consensus), NXPI $23 (vs. $20.49 consensus), MCHP $6.50 (vs. $5.17 consensus).
• TMT Breakout · HOOD: Mizuho raised its 4Q26 equity trading volume forecast to $1.074 trillion, from $823 billion previously, implying 50%+ year-over-year growth.
• James Bulltard · INTC: Cumulative gain of $12,871,006.69 on about $1.5M starting capital over 1,066 trading days; $3.3M already withdrawn for taxes; 2026 YTD return of 55% versus 13% for the market.
• TMT Breakout · MRVL: FY27 data-center revenue guided to $15-16B, with FY27-28 combined at $30B, up from a $20B framework given a year earlier.
• TMT Breakout · WDC: Per-TB HDD pricing up 18-19% year over year; most of 2027 capacity is already committed under long-term agreements.
• TMT Breakout · NET: Shares up 10.5%, the day's biggest gainer, as management reaffirmed confidence in a 50% Rule-of-50 growth target.
• TMT Breakout · VRT: Shares down 9.5%, the day's biggest decliner, after a SemiAnalysis report of field equipment failures corroborated by two data-center operators each over 1GW.
• SemiAnalysis · NVDA: After accounting for utilization, cloud-offload TCO is about $0.17/hr per PFLOP versus $0.28/hr on-robot, putting cloud cost at roughly 60% of on-robot cost industrially and about 15% in household settings; a single B300 GPU can serve 7 robots within a 1.6-second latency budget.
• Damnang · NVDA: DeepSeek V4 Pro has 1.6 trillion total parameters but activates only about 49 billion (roughly 3%) per token.
• Damnang: TrendForce estimates that in 2026 Q1, 64GB DDR5 RDIMM revenue and margin per wafer already exceeded HBM's.

Reproduced as stated by each publisher. Not verified by Quant Data.

What would prove them wrong

Several publishers name the observation that would falsify their own view. Those are the most useful sentences in any research note, so they are collected here.

• Capital Wars: A MOVE index spike that pushes repo collateral haircuts from about 1% to 2%, halving the collateral multiplier, would signal a liquidity shock.
• Capital Wars: The yield curve should begin flattening roughly 9 months after the late-2025 liquidity peak (i.e., 2026 Q3-Q4) to confirm the liquidity-cycle thesis; a material rise in term premium would instead point to genuine sovereign-credit concern.
• TMT Breakout · META: Muse's App Store ranking needs to hold in the top 2-3 and Meta's +4.5% move needs to hold (absent broad market weakness), or the valuation-upgrade case weakens.
• TMT Breakout · ORCL: Barclays flags that unresolved air-permit and gas-pipeline issues at Oracle's 2.45GW New Mexico campus could delay the planned 1H27 start and force FY28+ cloud estimates lower if not resolved within 1-2 quarters.
• James Bulltard · INTC: Reported cumulative gains, the 2022 and 2026 year figures, and the Intel entry/exit are all self-disclosed with no third-party audit; verification would require the author's paid database or live-feed position records.
• TMT Breakout · MRVL: Marvell's $30B FY27-28 data-center guide, particularly the NPO/CPO ramp raised from a $0-300M range, needs to show up in coming quarterly results.
• TMT Breakout · VRT: Whether the Semianalysis-reported field equipment failures are confirmed or refuted in 10-Q filings or customer commentary over the next 1-2 quarters, and whether 2H guidance gets cut.
• TMT Breakout: Nebius's CEO says the current GPU pricing ladder ($10-15M/MW bare metal up to $40-50M/MW for immediate delivery) should normalize within 18-24 months; continued acceleration instead of convergence would say supply tightness is more persistent than guided.
• TMT Breakout · MDB: MongoDB's hosted MCP server six-week usage data, due at the September 29 investor day, will show whether Atlas traffic is materially pulled forward.
• SemiAnalysis · NVDA: If deployed robots-per-GPU stays below roughly 4 (the industrial threshold), the cloud TCO advantage does not hold; robot-model scaling laws remain, in the report's own words, an open question.
• Damnang · NVDA: Watch HBM capacity per GPU, stack height and stack count in the generation after Rubin Ultra: continued declines would confirm the despec trend, a rebound would say the adjustment was temporary; also watch whether HBF lands in an actual shipping accelerator (not samples) versus remaining a supply-chain rumor.
• Cassandra Unchained: If Q4 AI capex or cloud guidance beats expectations (Nvidia around November, Oracle around December), the structural short thesis on the AI compute chain would be falsified; watch for position closures.

Where Quant Data's own data touches the same tickers

For tickers named above that have a stock page on this site, the settled options positioning from the same period. These are Quant Data's computations from end-of-day open interest, dated to the settlement shown, and are unrelated to what any publisher said.

Common questions

Sources: paid research publishers accessed through the GeoScope aggregator under a licence that permits redistribution. Statements and figures are attributed to their publishers and reproduced as stated; Quant Data has not verified them and expresses no view on them. Educational only: not investment advice, not a recommendation, not a forecast.