Premarket Sep 16: Fed day, 10y ~5%, Hynix–Intel
Futures modestly higher into Fed; 10y eased under 5%; oil still three-digit. Hynix–Intel memory talks; pacing reframed as more compute.
The lead
S&P futures modestly higher — ES +16.5 (+0.22%), NQ +126.75 (+0.44%), Dow e-minis +93 (+0.18%) at ~6:50am ET after Tuesday’s S&P −0.45% close at 7,585.73 (Reuters).
- 10-year eased under 5% — last ~4.973% (down >2bp) after Tuesday’s post-2007 high near 5.041%; 30y ~5.347% (CNBC).
- Oil still three-digit but off highs — Brent ~$107.47 / WTI ~$103.60 early Wednesday after Saudi cargoes-via-Oman relief (Reuters); Tuesday settles were ~$108.75 / ~$105.83.
- Fed day — path over the print. Decision + SEP/dots 2:00pm ET; Chair Warsh presser 2:30; FedWatch ~92.7% for a 25bp hike (CNBC); Reuters frames Warsh language as the market risk.
- Broader tape: Senate stalled the Clarity Act (crypto bill); CNBC flags household cost pressure from oil + rates (oil/rates).
- Overnight archive: SK Hynix–Intel U.S. memory talks; SemiAnalysis moratoriums map; Baker extended with Friar/Katti compute quotes; Satya All-In + SemiAnalysis Ep.031 on pacing-as-more-compute.
Open question: Does Warsh’s presser reprice the back end of the curve, or do three-digit crude and a sticky ~5% 10y keep equities discounting higher-for-longer even if the 25bp hike lands as priced?
Positioning
AI capex durability — operators and labs still framing pacing as more compute, not a guide-down. Baker’s overnight extension cites OpenAI CFO Sarah Friar (“getting more compute to keep that flywheel going”) and Sachin Katti (“even more compute” for safety/alignment) against Monday’s infrastructure selloff; pushback holds that pacing is about when spend lands, not the level. Nadella on All-In treats common-sense control + diffusion as compatible with build/lease/rent kit matching (~60% of cost asr). SemiAnalysis Ep.031 inverts “pace the frontier” into higher safety/eval intensity (OpenAI CoT monitoring ~20% of rollup compute asr) while physical bring-online stays below demand for 2–3 years. SemiAnalysis’s moratoriums map — 1,525MW slip vs 20GW inside restricted boundaries — separates permit friction from cancelled hyperscaler spend. No top-four sequential capex guide-down in window; oil + ~5% 10y remain competing discount-rate pressure into the Fed, not a theme falsifier alone.
HBM supply binds — onshoring talks, not near-term wafers. CNBC/Reuters exploratory Hynix–Intel U.S. memory path (Ohio lease and/or cloud-inclusive JV) sits beside Hynix’s already-announced $4B Indiana HBM plant; company relay says no plans confirmed. Room debate (discussion) correctly weights hyperscaler JV structure over the bilateral headline — still talks ≠ process transfer, tools, or yield this quarter. Theme’s third-supplier / architecture-cut falsifiers unmet.
Inference margin inversion — contested TCO snapshots, no lab GM print. SemiAnalysis says MI355X is closing agentic-inference perf/TCO vs GB300 via SGLang/MoRI/UMBP; replies’ InferenceX $/M-token cites still favor GB300 on other models/SLOs and treat the delta as stack maturity. No consecutive price-cut gross-margin disclosure from a frontier lab overnight.
Top stories
Fed at 2:00pm ET — Warsh path language is the equity variable. Markets ~92–93% priced for a first hike since 2023 into the 3.75%–4.00% range; SEP/dots and the 2:30 presser matter more than the binary print against a sticky long end (CNBC; Reuters).
10y eased to ~4.97% — oil still three-digit off session highs. CNBC 10y ~4.973% / 30y ~5.347% Wednesday morning after Tuesday’s 5.041% high; Brent/WTI still above $100 after early pullback (CNBC; Reuters).
SK Hynix–Intel — exploratory U.S. memory manufacturing talks. Reuters via CNBC: lease of Intel Ohio capacity and/or JV that could include major cloud firms; shares rose premarket; Hynix said no plans confirmed; Indiana $4B HBM plant remains the separate announced path (news).
SemiAnalysis — 300+ DC moratoriums map to ~1.5GW slip, not buildout death. 20GW inside restricted local boundaries; 1,525MW actually slips; 2.3GW nationwide including New York — count ≠ cancelled megawatts (news; X).
Senate stalls Clarity Act — crypto regulatory path slips. Procedural failure after ethics fights over officials’ crypto earnings; industry likely waits into next year (CNBC Morning Squawk).
Household oil + rates pressure — CNBC ~$1,700 per-household estimate. Rising energy and borrowing costs from the Iran-war oil spike and higher yields (CNBC).
The conversation
Baker pacing — Friar/Katti quotes land on “more compute,” not less. Overnight extension: OpenAI CFO and VP Compute Strategy language used against Monday’s infra de-rating; replies insist the bear case is the shape of the spend curve. Carried read: mechanism claim strengthened by principals; timing/curve still open (discussion).
Hynix–Intel — hyperscaler JV framing vs talks-only. jukan argues cloud-inclusive memory JV is the load-bearing angle; replies flag prior denials, cycle cost, and wafers-not-this-quarter. Carried read: structure question is right; confirmation still sources-say (discussion).
AMD MI355X vs GB300 — TCO closing on some agentic stacks only. SemiAnalysis direction claim vs InferenceX screenshots that still favor GB300 elsewhere; software portability is the check. Carried read: gap moving on named stacks; not a single apples-to-apples close (discussion).
In their words
More on AI pacing: Sarah Friar CFO of OpenAI yesterday on CNBC: “from where I sit today there is so much opportunity to drive growth that I am still highly focused on getting more compute to keep that flywheel going.” … Sachin Katti … “we’ll need even more compute to make sure future models are more safe and aligned.” If you thought “pacing” was negative for AI infrastructure demand, think again.
— Gavin Baker (@GavinSBaker), X, 2026-09-16 — Principal OpenAI compute quotes behind the pacing≠capex-cut read.
Not many people are paying attention, but what matters in the Hynix and Intel news is that Hynix and Intel could form a JV with hyperscalers hungry for memory supply. That's more important than the Hynix and Intel news itself.
— Jukan (@jukan05), X, 2026-09-16 — Structure claim that elevates cloud offtake over the bilateral headline.
BREAKING: AMD MI355X is quickly closing the perf/TCO gap in agentic inference, when compared to GB300 apples-to-apples.
— SemiAnalysis (@SemiAnalysis_), X, 2026-09-16 — Named research claim opening the MI355X vs GB300 TCO debate.
"All this additional safety stuff is actually still incredibly compute intensive."
— Max Kan, SemiAnalysis Ep.031, 2026-09-15 — Mechanism behind pacing→more FLOPs invert [asr].
Podcasts & interviews
All-In — Satya Nadella (AI pacing / kit strategy) — BULLISH / asr. Common-sense control + diffusion over mystical slowdown; kit ~60% of cost with build/lease/rent; ~30m Copilot/enterprise users vs ~250–300m TAM; Quincy campus ~400–500 MW. Recap · Watch
SemiAnalysis Ep.031 — pace the frontier — BULLISH / asr. Safety/ops excellence as compute-hungry; OpenAI CoT monitoring ~20% of rollup; physical supply still binds 2–3 years; HF as shot across the bow. Recap · Watch
What to watch
Fed decision + Warsh presser — 2:00 / 2:30pm ET. Unanimity, SEP path, and whether language sounds dovish vs multi-hike path priced through next year (Reuters; CNBC).
Hynix–Intel confirmation or hard denial. Lease/JV term sheet naming cloud equity/offtake and HBM vs commodity DRAM/NAND scope — or firm “no plans” that sticks (discussion; news).
MI355X vs GB300 same-SLO TCO tables. Published InferenceX or SemiAnalysis refresh across ≥2 agentic workloads (discussion).
East-West / crude path after Oman cargo relief. Whether Brent holds three digits into CPI and the hike-path debate (Reuters).
DC moratorium slip MW vs Wells count. Whether 1.5GW-class slip stays the right scale as local bans proliferate (news).
The Open/Close · Research commentary, not investment advice. Positions may be held in securities mentioned.