Nasdaq futures open interest jumps $10.6bn as earnings risk builds
@GoldmanSachs identifies a $9.5bn purchase in Nasdaq non-dealer net positioning, a new high.
With non-dealer positioning elevated in the CFTC data for Oct 20, @GoldmanSachs sees significant futures selling risk if Amazon/Microsoft/Alphabet results disappoint.
Funding adds another concern. @GoldmanSachs points to volatile Nasdaq funding costs above those of the S&P, with sector-specific liquidity risk for tech-heavy portfolios.
LLM token costs fell to ~50% of their May peak, according to @Nordea.
Nordea cites a Stanford study that found smaller models match or outrun larger ones on mainstream tasks. Smaller language models use 3% of the energy of large models, while 49% of business leaders are scaling back AI.
The longer-term risk is a mismatch between demand and capacity.
Nordea warns that a shift toward local AI and smaller models could leave data centres with overcapacity against 5-year project timelines.
Once again, Wall Street is long the left tail of compute, not long the right tail like technologists are.
AI is rebounding. Breadth isn't keeping up. @GoldmanSachs' US equities desk put NDX at +1.40%, reclaiming its 100DMA, in its 27 Oct 2026 midday update. SPX ex AI was down 50bps, with 286 names down.
@GoldmanSachs said "AI financing concern" eased after $META and $AVGO funding success. The rebound from oversold conditions was strongest in the AI complex: neoclouds +585bps, optical +5% and inference chips +466bps.
Momentum was +272bps and +8% in 3 sessions after a ~20% drawdown. @GoldmanSachs described flows as stable, with hedge funds favoring tech/financials over consumer/macro. That's a tactical recovery. The desk flagged weaker breadth as a potential source of risk-off pressure.
Meta $25bn common-stock offering about 5x subscribed
$META's expanded $25bn common-stock offering drew roughly five times demand for available shares. The company is targeting up to $98bn to fund AI spending.
More than half of the allotment went to long-term investors and existing shareholders; the top 15 subscribers received about 70% of the offering.
The deal priced at $625 per share, a 2.7% discount to the close the Friday before the announcement.
The equity sale is part of a broader financing package that includes $12.5bn of mandatory convertible preferred and expected at‑market share placements; if the full $98bn is raised it would be among the largest equity financings on record.
The 10-year rolling return on US Treasuries is now -1%, below the lows of December 1959 and September 1981, the latter of which marked the start of the greatest bond bull market in history:
If you are just now evaluating a software factory you are way the f*ck behind.
This should be your top priority.
At a minimum, every night run the Claude security plugin and @vercel’s deepsec on a dedicated machine with a dedicated user (not root) using the latest frontier model and also create a cascading “ladder” of models with the most advanced at the top of the ladder. This should be a launchd or cron task.
In addition, you need a realtime bot listening for / tracking all CVEs and it immediately provides patches and/or notifies you of a pull request for the patch and any operator actions (think DevOps). @grok @bot is the best at this because the CVEs propagate on X.
Then you can add, more nightly runners for performance, testing, documentation, feature drift, etc.