Daily Alpha · X
· Premarket Alpha · by Buzzberg Research
Rates, AI compute mix and Oracle's power-risk contract terms dominate verified posts, with Joseph Wang and Citrini providing the sharpest analytical frames.
Themes on this desk
Long-end rates
The 10-year broke above 5.20% for the first time in 19 years; Ackman argues hikes may now be inflationary because demand for intelligence and energy is rate-insensitive; Bianco notes the post-1971 10-year average is 5.91% versus a 5.18% clo
AI hardware tightness
BofA sees the memory super-cycle running through 2028-30 with no downturn; enterprise SSD lead times are 16 weeks versus a balanced eight; Nomura models 2027 1.6T optical DSP demand of 60-70m units against 40-50m supply; Macquarie sees a fi
Oracle data center
Barclays estimates Oracle's force majeure delays about $30bn of hardware capex; FT-sourced posts say Oracle guaranteed investor payouts even without power and that force majeure by itself does not suspend rent until delivery.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →Citrini: 'having/using AI' is not a solution; test customer renewal, pricing power, profit-pool replacement
Citrini says AI adoption will be the default for every company and is not itself a fix; the questions that matter are whether customers still renew, whether pricing holds, who owns the customer relationship, and whether the new strategy replaces old profits after new costs and cannibalization. He cites Kodak shifting to 54% digital mix by 2005 yet still filing Chapter 11 in 2012 after a 30% 2010 rally on turnaround signals.
Do customers still renew? Does pricing hold? Who owns the customer relationship?
Provides a falsifiable screen for separating incumbents that deserve to rebound from those that do not, rather than rewarding AI-adoption announcements.
Watch Whether companies citing AI adoption show stable renewal rates, pricing, and profit-pool replacement in subsequent quarters, or continue to lose the disrupted profit pool.
Source →Nvidia pushing glass substrates for CoPoS packaging
Author observes Nvidia appears serious about glass substrates and wants to make CoPoS work one way or another.
What’s interesting is that Nvidia seems pretty serious about glass substrates.
Glass-substrate adoption would shift value in advanced packaging supply chains and is a leading indicator for next-gen AI accelerator roadmaps.
Watch Supplier disclosures or Nvidia roadmap confirmations of glass-substrate CoPoS qualification.
Source →AI issuance is not crowding out Treasuries, per swap-spread evidence
@josephwang argues AI-related issuance is not crowding out Treasuries, citing the absence of more negative swap spreads and leveraged investor balance sheets absorbing supply.
That implies more negative swap spreads - which we don't see.
If true, the AI capex boom is not starving the Treasury market of capital, removing a systemic risk overhang for both credit and rates.
Watch Watch swap spreads for a sharp negative move, which would falsify the no-crowding-out thesis.
Source →Oracle bears power delivery risk on New Mexico data center under force majeure notice
Per FT reporting cited in the post, Oracle guaranteed payouts to investors on a massive New Mexico data center even if it fails to secure electrical power before the project comes online, with one person saying the power delivery risk sits with Oracle, not investors. Oracle sent the developer, a unit of Blue Owl Capital, a force majeure notice citing a denied permit key to its energy plans.
secured in time. “The power delivery risk sits with Oracle, not the investors,” one of the people said."
Contractual guarantees that survive a force majeure claim mean Oracle could carry rent and cost obligations on an unpowered asset, a direct hit to free cash flow and a test of how much delivery risk hyperscaler-adjacent landlords can shift to tenants.
Watch Confirm whether the force majeure notice is upheld and whether the permit denial is resolved; disclosure of additional charges or lease obligations in Oracle filings would validate the risk.
Source →ren_stocks: JPM model shows GPU-to-CPU ratio halving by 2028, bullish CPU names
ren_stocks cites J.P. Morgan's model showing the GPU-to-CPU ratio falling from 4.0x in 2024 to 1.9x by 2028, with TrendForce expecting one to two GPUs per CPU in agentic deployments and Lip-Bu Tan saying some frontier labs report four CPUs per GPU. He argues every agent orchestration step and sandbox runs on CPU, and Meta giving every Muse user an always-on VM supports the trend, concluding 'Bullish' on INTC, AMD, and ARM.
$INTC $AMD $ARM J.P. Morgan's model has the GPU-to-CPU ratio falling from 4.0x in 2024 to 1.9x by 2028.
If CPU attach per GPU roughly doubles, CPU vendors capture more of AI infrastructure spend even as GPU shipments grow, a differentiated read versus the GPU-centric consensus.
Watch Whether hyperscaler and server CPU attach ratios move toward 1:2 or 1:1 GPUs per CPU, and whether Intel, AMD, and Arm data-center CPU revenue accelerates accordingly.
Source →