Daily Alpha · X
· Premarket Alpha · by Buzzberg Research
Bubbleboi challenges a uniform HBM roadmap; Jukan highlights prospective memory financing and Serenity a conditional laser ramp. Baker adds a competition-based counterargument, while Citrini discloses an actual options purchase.
Themes on this desk
Architecture and demand
Bubbleboi’s lower-HBM, larger-domain thesis concerns architecture. Baker separately argues a frontier duopoly or open-source losses would hurt compute demand, even if development accelerates.
Refinery maintenance
Valueist shares Eneos’s 153,000-bpd Negishi maintenance shutdown and expected mid-November restart; lost throughput can outlast a crude logistics improvement.
Optical supply counterpoint
Serenity’s planned laser capacity ramp contrasts with PEquityResearch’s report of easing InP shortages. Neither statement quantifies delivered new supply.
Ticker heat
?
Top voices by smart followers and alpha score
Market Radar →Citrini bought TLT calls while warning the bond selloff could continue
Citrini discloses buying some TLT calls at the close, while acknowledging that a further bond selloff remains tempting to envisage. This is a disclosed purchase with explicit uncertainty, not an unqualified claim that yields have peaked.
Bought some TLT calls at the close today out of the fact that if this happens again I’d feel stupid
Implication: a duration-reversal position can coexist with concern that the adverse trend continues.
Watch Test whether long-duration bond performance validates or undermines the calls’ reversal thesis.
Source →Gavin Baker says a frontier duopoly would hurt compute demand
Baker argues an OpenAI–Anthropic duopoly would be terrible for compute demand whether the labs accelerate or pace development; he also says open source losing would be terrible for compute demand. This is his conditional counterargument to a quoted all-scenarios-bullish thread.
Funny and almost right. A scenario where OpenAI and Anthropic are a duopoly would be terrible for compute demand regardless
Implication: the thesis makes industry competition and open source adoption relevant to infrastructure demand beyond the pace of frontier training.
Watch Test whether competition and open source deployment broaden compute usage or concentration reduces it.
Source →Jukan highlights a possible hyperscaler memory JV
Jukan argues that a potential Hynix–Intel joint venture with memory-hungry hyperscalers is the underappreciated angle; he describes a possibility, not a signed venture.
could form a JV with hyperscalers hungry for memory supply. That's more important than the Hynix and Intel news itself.
Implication: direct hyperscaler participation could change how a memory supply project is financed and commercialized.
Watch Test whether the companies announce an actual venture, partners and committed memory supply.
Source →Serenity links Sivers laser expansion to tier-one potential
Serenity argues Sivers’100M-plus CW DFB laser capacity buildout places it in the tier-one laser-capacity conversation if the ramp executes. This is a conditional thesis, not a completed production milestone.
I think Sivers 100M+ CW DFB laser capacity buildout forced the industry to revisit its models and generated considerably more
The implication is potential scale expansion contingent on delivery.
Watch Test actual capacity ramp and commercial shipments against the planned scale.
Source →Bubbleboi highlights larger scale-up domains as a lower-HBM workaround
Bubbleboi shares GDP’s claim that Chinese AI hardware designs compensate for lower HBM per NPU with larger scale-up domains. This is the quoted hardware thesis, not a disclosed trade.
- low HBM for each NPU and large scale up domains to compensate for that - at least 2 years
Implication: more scale-up connectivity could partly substitute for memory per accelerator, challenging simple comparisons based only on HBM capacity.
Watch Test whether measured workload performance supports the claimed substitution between HBM per NPU and scale-up domain size.
Source →