Postmarket Sep 15: rates/oil dominate; chips bounce into Fed
S&P −0.45% as 10y tagged ~5.04%; SOX bounced while indexes fell. Ramp Astra 13% vs Fable 8%; AWS Bahrain unrestorable; Kyber slip to 2028.
The lead
S&P −0.45% — closed 7,585.73 (−34.25) vs Monday 7,619.98; Nasdaq −0.78% to 25,981.57; Dow −0.63% to 52,093.11 (CNBC ~4:20–4:38pm ET).
- Rates and oil owned the session. The 10-year tagged an intraday high of 5.041% — highest since 2007 on the CNBC/Tradeweb print — then settled ~5.006% (prior close ~4.961%). Brent settle ~$108.75 / WTI settle ~$105.83 (prior closes ~$105.68 / ~$101.39) on CNBC Quote Cache; midday Motley Fool had Brent ~$108.41 / WTI ~$104.76 (Fool).
- Monday’s AI de-rating did not finish the job on chips. SOX closed +0.40% to 11,175.55 after Monday’s ~−5.9% wipeout; Motley Fool flagged Nvidia/Micron/AMD/Intel rebounds while Mag7 and banks dragged indexes lower (Fool).
- FedWatch still ~92–93% for a Wednesday 25bp hike; path language and Chair Warsh’s presser remain the event (CNBC).
- Afternoon archive prints: Ramp Astra 13% vs Fable 8%; AWS Bahrain unrestorable / one UAE AZ unrecoverable; GS-via-X Kyber NVL144 / Rubin Ultra to 2028.
Morning’s open question — finish Monday’s AI de-rating, or do oil + a ~5% 10y dominate while semis digest guidance vs essay language? — resolved toward rates/oil dominating the index tape while chips mean-reverted, not a second-day SOXX rout.
Open question: Does Wednesday’s Warsh presser reprice the back end of the curve, or do three-digit crude and a sticky ~5% 10y keep equities discounting higher-for-longer into the hike?
Positioning
AI capex durability — delivery timing and geo-risk, not a guide-down. A Goldman expert call relayed on X slips Kyber NVL144 racks for Rubin Ultra to 2028 (>12 months past a 2027 target) and cuts 2030 US 800VDC share of new datacenters to 21% from 25% on inference mix — capex can stay elevated while specific rack/power SKUs slip (secondhand; primary GS note not opened). Separately, AWS told customers Bahrain-hosted resources/data are unrestorable and one UAE AZ is unrecoverable six months after Iran-war drone damage (news) — a concrete Gulf geo-risk print on the hyperscaler build, not a sequential capex cut. Neither print is a hyperscaler sequential guide-down — the theme’s falsifier stays unmet.
Inference margin inversion — frontier spend share shifted on Ramp and OpenRouter. Ramp’s Ara Kharazian puts Astra at 13% of enterprise AI spend vs Fable at 8% this week and frames Anthropic’s pace-the-frontier call as already losing adoption; OpenRouter shows OpenAI spend ahead of Anthropic for the first time in >2.5 years (discussion). Replies contest sample (Global-2000 vs Ramp card spend; developer vs enterprise). Carried read: spend share moved; lab gross-margin falsifier still untested — no consecutive price-cut GM disclosure in window.
Top stories
10y tagged ~5.04% — oil still three-digit into Fed. CNBC/Tradeweb 10y high 5.041% / last ~5.006%; Brent settle ~$108.75, WTI ~$105.83; FedWatch ~92–93% for a 25bp hike Wednesday with path language the real variable (CNBC Fed; WSJ bond headline in RSS).
Chips bounced; indexes still red. SOX +0.40% while S&P −0.45% / Nasdaq −0.78% — Monday’s concentrated AI de-rating partially unwound in semiconductors even as rates pressure hit the broader tape (CNBC; Fool).
AWS — Bahrain region unrestorable; one UAE AZ data unrecoverable. First public update since April: damage exceeded design tolerance; two other UAE zones still being worked; UAE Stargate (5 GW / ~10 sq mi with OpenAI/Nvidia/Oracle/Cisco) cited as Gulf AI campus context (CNBC; news).
Kyber NVL144 / Rubin Ultra racks — 2028 vs 2027 target. Goldman Nvidia 800VDC expert call via @ParadisLabs: >12-month slip; 800VDC 2030 share 21% from 25%; inference racks up to 150 kW keep traditional architecture feasible (secondhand) (news).
Ramp — Astra 13% vs Fable 8% on enterprise AI spend. Kharazian: OpenAI winning frontier enterprise spend; pacing call already costly for Anthropic adoption; OpenRouter parallel first OpenAI>Anthropic spend week in >2.5 years (discussion).
The conversation
Ramp Astra vs Fable — pacing already costing frontier spend share? Principal card-spend claim (13% / 8%) plus OpenRouter OpenAI spend lead; replies flag Global-2000 / regulated-pipeline sample and developer-vs-enterprise split. Carried read: week’s spend share favors the pacing-as-cost frame on Ramp/OpenRouter panels; do not generalize to all enterprise without ETR-class confirmation (discussion).
In their words
OpenAI is winning enterprise spend at the frontier. As of this week, Astra takes 13% of enterprise AI spend vs. Fable (8%) per Ramp data.
— Ara Kharazian (@arakharazian), X, 2026-09-15 ~15:30 ET — Falsifiable spend-share print behind the pacing-cost read.
Anthropic took a big risk in its recent call to pace the frontier. It's frontier model has already fallen behind on adoption.
— Ara Kharazian (@arakharazian), X, 2026-09-15 — Causal claim tying the pace call to lost adoption.
OpenRouter users spent more on OpenAI models than on Anthropic models last week. This hasn't happened for more than 2.5 years.
— tae kim (@firstadopter), X, 2026-09-15 — Parallel OpenRouter venue confirming OpenAI spend lead (Peter Walker: Astra top by spend / Luna top by tokens).
What to watch
Fed decision Wednesday 2:00pm ET — path over the print. Statement, SEP/dots, Chair Warsh presser 2:30; markets ~92–93% priced for 25bp (CNBC); whether 10y holds ~5% after the presser is the equity variable.
Ramp / OpenRouter spend-share persistence. Multi-week Ramp AI Index series and ≥4 consecutive OpenRouter weeks of OpenAI>Anthropic spend (or revert) through mid-October (discussion).
Kyber / Rubin Ultra confirmation. Primary Goldman note or NVIDIA delivery commentary validating the 2028 rack slip and 800VDC share cut (news).
Gulf DC hardening / Stargate dispersion. UAE site-dispersion and hardening follow-through after AWS’s unrestorable print (news).
East-West pipeline / crude path into CPI. Whether Saudi East-West outage duration keeps Brent three-digit and feeds the hike-path debate.
The Open/Close · Research commentary, not investment advice. Positions may be held in securities mentioned.