Daily Alpha · X
· Post-Market Alpha · by Buzzberg Research
X argued with itself over the AI complex: chip dip-buyers met a permanent-regulation bear case, refiners got the sharpest product-shock work, and software and cyber was the day's clearest rotation.
Themes on this desk
AI hardware
Dip-buying against a permanent-regulation bear case; PCB content per rack up 69-222% and Nvidia reserving probe-card and test-socket capacity.
Rates
10-year through 5% with positioning set for one-and-done; the dot plot is the risk and Brent above $100 keeps the hike case alive.
Oil and products
Repairs 3-5 weeks versus official 'running soon'; IEA cut 2026 demand by 940k bpd; refiners priced for margin, not volume.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →Gavin Baker: frontier labs 'pacing' means more compute on alignment, lower margins
Baker argues Anthropic and OpenAI will 'pace' the frontier by spending more time and more compute on alignment, monitoring and evals, and that labs choosing to pace likely spend slightly more on compute at the cost of lower margins.
on alignment, monitoring and evals. The frontier labs that choose to “pace” likely spend slightly more money on compute at
If correct, safety pacing is a margin headwind for frontier labs but a demand tailwind for compute suppliers; contradicts the view that safety slows compute spend.
Watch Frontier lab disclosures on compute spend and gross margin, and continued accelerator demand despite safety rhetoric.
Source →Salesforce up 32% in a month as software outperforms chips
bubbleboi noted Salesforce is up 32% in a month, quoting a post that software stocks outperformed chip stocks by the greatest margin in history.
Salesforce is up 32% in a month.
A record one-day software-versus-semis spread plus a sharp single-name move suggests rotation into software; it is a price observation, not a stated position or forward call.
Watch Whether software's relative outperformance versus semiconductors persists over subsequent sessions or mean-reverts.
Source →East-West pipeline outage framed as product-market shock, not crude shock
TheValueist argues the Saudi East-West pipeline shutdown is more consequential for refined products than a conventional Saudi crude-export disruption because it threatens both crude flows to international markets and feedstock supply to a large refining system outside the Strait of Hormuz; a sustained interruption would favor operationally reliable U.S. refiners with substantial diesel production, secure crude access and export capability.
$VLO $MPC $PSX $DK $CLMT INVESTMENT ASSESSMENT The East–West pipeline shutdown is potentially more consequential for refined products than a
If correct, the marginal barrel at risk is middle distillate rather than crude, which would transmit into diesel cracks and U.S. refiner realizations rather than simply crude flat price.
Watch Confirm with documented Saudi refinery run cuts and product-loading reductions; invalidate if pipeline flow is restored enough to keep the Red Sea refinery cluster fed.
Source →AI safety going mainstream is a permanent regulatory regime shift that lowers compute value
jasonschips argues AI safety went from a niche San Francisco concern to a mainstream cable-news story in ten days, that this is a permanent shift in the Overton window making AI 'a regulated one' rather than an innovative technology, and that because a lab seen accelerating becomes the target of regulation, every lab will train less than it otherwise would — so compute is worth less than in a race-to-superintelligence world, with the demand loss landing on hardware while labs shift to capital-efficient inference.
By contrast, in a slowed, regulated industry, hardware economics are not maximized.
A structural bear case for AI hardware demand that does not depend on any specific regulation passing, implying lower training-compute intensity and multiple compression across the semiconductor supply chain.
Watch Evidence of actual training-run scale cuts, training-to-inference compute shifts, or passage of binding legislation (e.g., the Sanders/Casar bill); continued record training-cluster capex would falsify the thesis.
Source →Nvidia, Palantir, Booz Allen and Pentagon limit Anthropic model use
Multiple wires report Nvidia, Palantir, Booz Allen Hamilton and the Pentagon will limit their use of Anthropic's AI models, following the weekend debate over slowing AI development.
BREAKING: NVIDIA, Palantir, Booz Allen Hamilton and the Pentagon have said will limit their use of Anthropic models.
Enterprise and defense buyers pulling back from a frontier lab is a demand signal for competing model providers and a reputational hit to Anthropic, and it shows AI procurement decisions are now politically entangled.
Watch Whether Anthropic confirms lost contracts or revenue impact, and whether rival model providers publicly announce wins from these accounts.
Source →