Daily Alpha · X
· Post-Market Alpha · by Buzzberg Research
Anthropic's filed numbers, the 30-year yield record, Meta's Muse distribution push and open-weight substitution set the agenda; disclosed positioning clustered in AI power, optics and advanced packaging.
Themes on this desk
Anthropic IPO economics
The leaked filing's $42bn net loss and $518bn of forward cloud commitments sit against a reported $2tn target, with about $161bn of Broadcom TPU leases and a quarter of revenue from two customers.
Meta and Muse momentum
Bulls argue Muse's distribution across 3.6bn users and Meta's $200bn revenue base let it give agents away, with North America ARPU possibly topping $400 by year-end; Altman calls Muse a nice product but says he is not threatened.
Memory pricing
TrendForce sees blended HBM ASPs up 121% year on year with 2027 supply constrained and 8-Hi carrying a 10-20% per-Gb premium, while server DDR5 RDIMM pricing sits above $2,000 per unit into Micron's print.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →El-Erian frames the lock-in-versus-sidelines dilemma as every Treasury maturity except the 2-year sits above 5%
Mohamed El-Erian notes that with the exception of the 2-year, every maturity on his Bloomberg screen is now above 5%, and poses the question of whether investors should lock in the highest nominal and real yields in years or stay sidelined amid a structural supply/demand imbalance and volatility as fast money fills the void.
With the exception of the 2-year, every maturity on this Bloomberg screen is now above 5%.
This is a framing of a genuine allocation decision point: high absolute yields create a valuation floor argument for duration, but the supply/demand imbalance and fast-money participation argue for wider term premia and continued volatility.
Watch Whether long-end auctions show improving non-dealer participation, which would validate the lock-in case, or continued weak demand, which would validate staying sidelined.
Source →Photon Capital: AI flows broadening into advanced packaging/test, long AMKR and ASX
The author argues strength in $AMKR and $ASX shows AI investment flows broadening into advanced packaging and test; $AMKR expanded Arizona advanced packaging/test investment to $12B after customer commitments exceeded Phase 1 capacity, with Q2 revenue $1.90B (+26% YoY) on record data-center Computing revenue, while $ASX posted Q2 revenue +17% and EPS +149% YoY on AI/HPC packaging demand. The key signal is the two moving together, with attention shifting downstream from GPUs to HBM and packaging/test bottlenecks.
- The strength in both $AMKR and $ASX suggests AI investment flows are broadening into advanced packaging and test.
Identifies the next bottleneck in the AI supply chain (advanced packaging/test) and names two listed pure-plays, with $AMKR offering US onshore exposure and $ASX large-scale AI/HPC OSAT exposure — a differentiated way to play AI capex beyond GPUs.
Watch Whether AMKR and ASX continue to move together rather than one being an isolated move; AMKR's $12B Arizona buildout progress and further HBM/packaging capacity commitments.
Source →Anthropic's $252B non-cancelable compute and $161B TPU lease obligations
bubbleboi reacts with alarm to a quoted Reuters report that Anthropic carries $252 billion in non-cancelable compute obligations across Microsoft, Google and Amazon, plus $161.2 billion in Broadcom TPU lease obligations that are largely non-cancelable.
Anthropic has $252 billion in non-cancelable compute obligations across Microsoft, Google and Amazon, and $161.2 billion in Broadcom TPU lease
Non-cancelable multi-year obligations of this size convert hyperscaler and Broadcom AI revenue into contracted, but counterparty-concentrated, backlog — a durability question for the AI capex trade if Anthropic's revenue ramp disappoints.
Watch Anthropic revenue disclosures versus the contracted obligation schedule; any renegotiation, impairment, or disclosure of these leases in hyperscaler/Broadcom filings.
Source →Corporate pivot to open-weight AI models threatens closed frontier lab pricing power
Corporate America is shifting structurally toward open-weight AI models, with executive mentions up 6x over the past year as businesses bypass expensive frontier APIs; AT&T routes 40% of AI workloads through open-weight models targeting 70% while cutting coding and operational costs 56-80%, and Match Group and PNC are routing workloads away from proprietary frontier APIs.
Corporate America is orchestrating a massive structural pivot toward open-weight AI models, with executive mentions spiking 6x over the past
If enterprises can cut AI costs by more than half using open-weight models, the pricing power and revenue moats of closed frontier labs (OpenAI, Anthropic) come under pressure, while cost-saving adopters like AT&T, Match Group and PNC see margin benefit.
Watch Whether open-weight adoption rates continue rising and whether frontier labs respond with price cuts; a plateau in open-weight migration or successful premium lock-in by closed labs would invalidate the threat.
Source →US SPR at lowest since 1982, 27th straight weekly decline
US Strategic Petroleum Reserve inventories fell about 800,000 barrels last week to 283.8 million barrels, the lowest since 1982, marking a 27th consecutive weekly decline and down more than 132 million barrels from the high; the Trump administration is offering up to 40 million barrels from the SPR, which could drive inventories below 250 million barrels.
BREAKING: Inventories in the US Strategic Petroleum Reserve fell by ~800,000 barrels last week, to 283.8 million barrels, the lowest
A depleted SPR removes a key supply buffer, raising the sensitivity of crude prices to any geopolitical disruption and reducing the US's ability to cushion oil spikes.
Watch Whether the 40 million barrel SPR offering is actually drawn and whether inventories break below 250 million barrels.
Source →