Daily Alpha · X
· Post-Market Alpha · by Buzzberg Research
The loudest recurring threads were memory pricing power ahead of Micron, Oracle's repricing credit, Muse adoption and sandbox math, and long yields at 2007 highs.
Themes on this desk
Memory pricing
CK Capital added to PENG into its October 6 earnings citing Micron's DRAM read-through; TheBigBerbowski calls MU the lowest-valued AI hardware name; BofA sees 2027 HBM ASP growth far below 100%.
AI credit
Oracle CDS at a record with roughly 19% implied default odds, $9bn of additional Ellison stock pledged for the Warner Bros deal, and long bonds above 8%.
Muse adoption
M Science says two-thirds of early Muse users were new to LLMs; a SEV-2 VM-escape flaw and Amazon's store block are the counterweight; Korean users are routing around regional limits.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →Meta Muse early data shows market expansion, not share stealing
Jukan cites M Science data that two-thirds of early Meta One users had never used an LLM before, calling it incredibly bullish for AI as a whole because Meta is bringing Instagram users who know nothing about AI into the market; M Science notes Meta might not be stealing share yet but Muse is growing the market.
“Two-thirds of early Meta One users had never used an LLM before.” Meta is bringing Instagram users who know nothing
If accurate, new-to-AI users expand total inference demand rather than shifting share, supporting the compute/memory demand thesis and Meta's distribution advantage.
Watch Independent confirmation of the two-thirds new-LLM-user figure and Meta One/Muse DAU and engagement trends.
Source →VR200 bill-of-materials cost inflation quantified
pequityresearch reports considerable price inflation in several VR200 BoM components due to a supply crunch: memory +435%, PCB +233%, MLCC +182%, ABF substrate +82%.
and PCB. 1. Memory +435% 2. PCB +233% 3. MLCC +182% 4.
Triple-digit component cost inflation on a next-gen rack implies margin pressure for system assemblers and windfall pricing for memory, PCB, MLCC and substrate suppliers.
Watch Subsequent BoM teardowns or supplier earnings showing these increases flowing into revenue and gross margin.
Source →Meta Muse 100M DAU could require 1-4GW power and 7-10GW compute, with inference the real bottleneck
A quoted analysis estimates Meta's Muse consumer agent would need ~1GW of power in the base case (only ~0.1GW from the CPU/VM sandbox layer) and 3-4GW if reasoning-equivalent calls per user are higher; sandbox CPU content is sub-$1B and DRAM ~$2B, far smaller than many expected, while inference remains the binding constraint and can scale faster than users as agents run longer trajectories. @bubbleboi endorses it as 'GOOD work.'
power > > So inference alone could require: ~1-2GW of average power > > A 3-4GW Muse is entirely plausible.
If correct, the Muse buildout is primarily an inference/power story rather than a CPU/DRAM story, shifting the beneficiary mix toward power, accelerators and memory bandwidth and away from the naive per-user VM math that inflated CPU demand expectations.
Watch Meta's disclosed capex/power additions and any confirmation of the rumored 7-10GW compute expansion; falsified if Muse DAU scaling shows sandbox CPU/DRAM demand far above the sub-$1B/$2B estimates.
Source →Oracle CDS at record implies ~19% default probability by end-2031
Holger Zschaepitz reports Oracle's 5-year CDS surged to a record 240.7bps after a force majeure notice for Project Jupiter rattled AI data-center debt markets, implying a 19.1% probability of default by end-2031 assuming 40% recovery.
5y CDS quoted at 240.7bps imply a 19.1% probability of default by end-2031, assuming 40% recovery. Almost 1 in 5!
Credit markets are pricing meaningful Oracle default risk tied to AI data-center obligations, a signal that AI infrastructure debt is being repriced and could pressure related financing and counterparties.
Watch Whether Oracle CDS stays elevated or widens further, and whether the Project Jupiter force majeure is resolved or triggers broader AI data-center debt repricing.
Source →Long-end Treasury yields at highest since 2007
@KobeissiLetter reports the 10Y note yield above 5.10% (highest since July 2007) and the 30Y at 5.44% (highest since June 2007), with IG credit at 5.7%, high yield at 7.5%, private credit at 8.3% and money market funds at 3.6%, framing it as 'higher for longer' creating income opportunities.
10Y Note Yield is currently trading above 5.10%, its highest level since July 2007.
Multi-decade-high long yields reset discount rates for equities and raise the bar for risk assets, while making credit and duration-sensitive income strategies competitive with equities for the first time in years.
Watch Whether 10Y and 30Y yields hold above 5% or break back below, and whether credit spreads widen as yields rise (stress) or stay tight (income bid).
Source →