Research notes · September 8, 2026

Eight notes from five publishers: OpenAI's Astra reshapes the AI capex debate, GS/Citi TMT conference recaps, and a wave of sell-side rating changes

On 2026-09-08 the tracked publishers put out eight pieces from five sources. AlphaPick logged a mechanical, valuation-driven trim of its Alpha Picks position in POWL. FundaAI published two deep dives — an initiation on Taiwan's Delta Electronics and a framework piece on OpenAI's new Astra (GPT-6) model and its implications for compute demand. TMT Breakout filed three pieces: an EOD wrap of the Goldman Sachs and Citi TMT conferences, a morning wrap bundling AI-capex forecasts and a stack of sell-side rating changes, and a writeup of OpenAI CFO Sarah Friar's conference remarks. Capital Wars issued its weekly global-liquidity update, and James Bulltard posted a personal options-trading recap. Every statement below is attributed to the publisher that made it; Quant Data did not verify any of these numbers and takes no view on them.

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

• AlphaPick · POWL: Combined return across two portfolio inclusions >1000%; EBIT margin 19.67%, net margin 16.49%, levered FCF margin 16.45%; Alpha Picks portfolio +348.00% since July 2022 vs S&P 500 +104.02%.
• FundaAI · 2308.TW: DC/DC (VPD) revenue share ~8.3% of sales in 2026 rising to ~12.2% in 2027 at ~50% gross margin; FY27/FY28 revenue growth of 53%/28% with EPS of TWD77.3/102.4; shares at ~17.8x FY28E EPS.
• FundaAI · NVDA: Astra trained on 100,000+ GPUs at roughly 10x the compute of the prior model; OSWorld computer-use score rose from 65.6% to 72.6%; coding's addressable market put at ~$1 trillion versus knowledge work at >$10 trillion.
• TMT Breakout: SOX +1.3% versus TMT momentum +7% and QQQ -8bps on the GS/Citi TMT conference day.
• TMT Breakout · AVGO: CEO Hock Tan's framework: 1GW of compute ≈ $30B ARR against ~$10B run cost; global inference cost ~$200B/year versus model revenue ~$150B/year.
• TMT Breakout · QCOM: AWS deal: ~$4B in warrants (25M shares at $161.26 strike) tied to up to $60B of Qualcomm server-chip purchase commitments.
• TMT Breakout · HPE: Networking orders +36%, cloud/AI orders +75%, purchase commitments above $30B, FY27 revenue growth guided to 13-17%.
• TMT Breakout: Evercore ISI: 2030 AI-related data-center demand of 149-189GW (total capacity 208-250GW); Morgan Stanley: 2027 hyperscaler capex +60% to ~$550B incremental spend, slowing to ~12% growth in 2028.
• TMT Breakout · DELL: This quarter's AI server bookings ~$61B, backlog ~$95B, enterprise AI-factory deployments up to ~6,500 (+60% in six months), per Bernstein channel checks.
• TMT Breakout · MSFT: OpenAI enterprise run-rate revenue up 32% month-over-month in July (overall ~20%); Codex users grew from ~100,000 to ~25 million; advertising hit a $1B run rate in 7 months.
• James Bulltard: WTI crude up ~20% in 9 sessions, nearing $95/barrel, with the author expecting a possible return to $100 within two weeks; his own YTD return is +52%.
• Capital Wars: Global liquidity at $195.4 trillion (3-month annualized +2.3%, annualized +5.8%) versus a shadow monetary base of $109.4 trillion contracting at -0.3%/-0.1% annualized; MOVE index at 73.

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.

• AlphaPick · POWL: If POWL's rating falls from Hold to Sell or Strong Sell, the Alpha Picks rules trigger a full exit of the remaining position (same if the company is acquired or merges).
• FundaAI · 2308.TW: August monthly revenue (due early September) needs to extend July's record MoM acceleration, or the VPD ramp thesis is undercut; Q3 gross margin needs to hold flat or rise; HVDC needs to hit Q4 volume production.
• FundaAI · NVDA: Whether computer-use multi-step reliability (e.g., OSWorld 2.0) breaks out meaningfully in the next model generation, confirming a 'Claude 3.7-to-4.5' moment is near.
• TMT Breakout · AMD: If 2027 full-year gross margin comes in meaningfully below 'slightly under' the Q3 56% guide, that would undercut management's MI450 ramp execution claim.
• TMT Breakout · QCOM: If FY27 data-center revenue misses ~$5B or the FY29 path falls short of $15B, the 'high confidence' framing of the AWS deal is undercut.
• TMT Breakout: If SHOP/DASH/CART/BKNG earnings show no evidence of agent-driven erosion of customer relationships or pricing power, the 9/8 Instinct-driven sell-off would look like an overreaction.
• TMT Breakout: If actual 2027 capex guidance doesn't show the deceleration Morgan Stanley expects for 2028 (~12% growth versus ~60% in 2027), the enabler/software rotation thesis weakens.
• TMT Breakout · NVDA: If Google's native TorchTPU support for vLLM/SGLang doesn't open-source around mid-October, or third-party benchmarks don't confirm the TPU v7 price/performance edge, Nvidia's software-moat concern is overstated.
• TMT Breakout · MSFT: If OpenAI's promised New York financial-services announcement doesn't land, or lands weak, the enterprise-vertical-expansion narrative should be discounted; same for the promised year-end ChatGPT/ChatGPT Work merger.
• James Bulltard: If WTI doesn't return to around $100/barrel within about two weeks, or there's no geopolitical de-escalation news, the 'falling oil lifts stocks' narrative is disproven; the bull-flag reading also fails if SPY doesn't hold the 21-day average and break the top of its four-month range.
• Capital Wars: A MOVE index break above 80 sustaining, or shadow-monetary-base 3-month/annualized growth failing to turn positive, would confirm the liquidity surface reading is a volatility-suppression illusion rather than real creation.

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.