Daily Alpha · X
· Premarket Alpha · by Buzzberg Research
AI hardware share shifts, memory pricing, long-end rates and distillate policy generated the most decision-useful disagreement.
Themes on this desk
Long-end yields and Fed path
The 30-year yield is at a 24-year high, Bloomberg reports a buyers' strike, the BoE warns AI valuations could spill into sovereign bonds, Williams pushes back on October hike pricing, and France's spread over Germany hits a 2012-crisis wide
Memory and HBM pricing
TrendForce sees HBM ASPs up 121% in 2027, Micron faces a calendar-quirk guide with multi-year price caps, and Samsung is modeled at a record Q3 on memory pricing.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →White House weighing feasibility of a diesel export ban
Unusual Whales relays Treasury Secretary Scott Bessent saying the White House is examining whether an export ban is feasible in terms of overall refining capacity and whether a full or partial ban would work.
The White House is examining whether an export ban is ‘feasible in terms of the overall refining capacity and whether
A US diesel export ban would reprice domestic refining margins, global distillate trade flows, and diesel-exposed transport and industrial costs; even the study of it introduces policy risk into refiners and exporters.
Watch Formal policy proposals, refining-capacity assessments, and any change in US distillate export volumes or crack spreads.
Source →Amazon's equity/warrant stakes span the AI hardware supply chain
Serenity (@aleabitoreddit) argues Amazon's portfolio of ownership/warrant positions across AI hardware suppliers — including optical (AAOI, Celestial/MRVL), connectivity (ALAB, CRDO), assembly (JBL, FLEX, FN), ASIC partners (QCOM, Alchip), and PCB (Gold Circuit Electronics) — may end up more compelling than Nvidia's own holdings.
Bro, $AMZN semi ETF portfolio is probably going to end up more compelling than what $NVDA owns...
If Amazon is locking in supply and pricing across multiple layers of the AI hardware stack via equity stakes, it could compress margins for merchant suppliers while entrenching Amazon's own ASIC/cloud cost advantage.
Watch Follow-on 13F/13D or private-placement disclosures confirming additional Amazon stakes, and whether named suppliers disclose Amazon as a customer or equity holder.
Source →MediaTek projected to take 100% of Google v9 TPU generation
P Equity Research relays Jeff Pu's view that MediaTek holds 100% share of the v9 generation (Humufish training, Triggerfish inference), with no other v9/2nm projects expected in the window, and that MediaTek should still win the major Icefish SKU while AMD/MRVL or AVGO compete for Merope.
MediaTek: > MediaTek is projected to hold 100% market share for the v9 generation, split between Humufish (training) and Triggerfish
A full-generation ASIC socket win would concentrate Google TPU economics in MediaTek and pressure the case for Broadcom/Marvell share in that program.
Watch Confirmation via MediaTek/Google tape-out disclosures or supply-chain checks; invalidation if a second v9/2nm vendor emerges or Icefish is awarded elsewhere.
Source →OpenAI's GPT-6.1 Sol Ultrafast runs on Nvidia GPUs, not Cerebras
SemiAnalysis reports OpenAI's latest GPT6.1 Sol Ultrafast is not running on Cerebras but instead runs at low batch size on NVIDIA GPUs; bubbleboi reacts with alarm ('Oh shit this is baaaaaad lol'), implying a negative read for Cerebras and a positive for Nvidia.
Cerebras but is instead running at a low batch size on NVIDIA GPUs.
If frontier inference workloads stay on Nvidia GPUs even at low batch sizes, Cerebras's differentiation thesis for latency-sensitive inference is weakened while Nvidia's moat in inference is reinforced.
Watch Whether Cerebras is subsequently announced as a serving partner for GPT6.1 Sol Ultrafast, or whether OpenAI discloses continued Nvidia-only deployment.
Source →Micron preview: calendar quirk and price caps could make guidance look like a miss
tengyanAI argues Micron's forward guide could look like a miss even as the business grows: the just-reported quarter had 14 weeks versus 13 next quarter (~7% headline drag), about half of revenue is moving into multi-year deals whose first tranches cap prices at April-June levels, and HBM is already priced for all of 2026. Their read is a guide at roughly this quarter's revenue +4%, ~$57B with ~87.5% gross margin and ~$35-36 EPS, and they flag $59B+ as the level that would show the caps binding far less than assumed.
https://t.co/deToFx35tL if micron guides $59B+ on a ~$56B quarter.
Sets a concrete bar for interpreting MU's print: a soft headline guide may be mechanical rather than demand-driven, which matters for how memory names trade on the reaction.
Watch Whether MU guides at or above $59B on a ~$56B quarter, and whether management quantifies revenue under capped contracts and adjusts growth for the extra week.
Source →