Chapter 09 / 16
Network and Photonics
网络、互连与光子层只有在架构承诺、认证和产品收入出现时才能越过观察门。
High-Signal Viewpoints This Week
| Source / account | Traceable source | Core view | Strategy formula | Classification | Radar implication |
|---|---|---|---|---|---|
| Serenity / @aleabitoreddit | X profile/post snippets; TradingKey reconstruction; Johnson Lee methodology note | The highest-return hunting ground is not "AI beneficiaries" but obscure physical chokepoints in AI infrastructure, especially photonics, InP, CW lasers, SiPh, testing, robotics components, and rare-earth magnets. | Start with certain terminal demand -> identify architecture migration -> isolate the non-routable material/component -> verify supply cannot expand quickly -> pick the focused oligopoly supplier before social consensus. | Early bottleneck discovery when evidence is original; now partially consensus-following because the Serenity trade itself is crowded. | Add an explicit "architecture-migration bottleneck" rule, but penalize names whose only edge is "Serenity mentioned it." |
| Serenity-derived SIVE discussion | TradingKey, X snippets, Sivers/Jabil, Sivers/GF | SIVE is framed as a high-power DFB laser / external light source bottleneck for 1.6T, CPO/LPO and AI optical interconnect. | Copper power/thermal limits -> optical interconnect -> CPO/LPO needs external lasers -> validated design-in creates switching cost -> scarce InP/DFB capacity creates pricing power. | Was early bottleneck discovery; as of this week, social consensus and valuation heat are rising fast. | Add SIVE to observation pool, not core pool, until independent share, capacity allocation, revenue conversion, and valuation math are proven. |
| TheValueist | X snippets on Broadcom AI XPV, NBIS, MRVL, storage | AI infra value is migrating from single-chip winners to financed platforms, power access, custom XPUs, networking, storage, and site execution. | Compute demand -> capital vehicle / contracted power / custom silicon -> bottleneck shifts from component availability to deployable financed capacity. | Mix of consensus-following and early infrastructure-bottleneck mapping. | Add "capitalized compute platform", "contracted power", "active MW vs contracted MW", and "financing vehicle" to watch terms; do not let high headline GW substitute for monopoly score. |
| FinNewsNow / data center power repost ecosystem | X snippet plus DCD source | A South Carolina data center project was cancelled because utility power could not be sourced in time. | If project economics fail at interconnection stage, the scarce asset is time-to-power, not just GPUs or buildings. | Early bottleneck evidence when tied to specific cancelled projects; becoming consensus at theme level. | Raise evidence_velocity for onsite power, grid interconnection, switchgear, transformers, and behind-the-meter suppliers. |
| Storage investor / media cluster | MarketWatch, TechRadar, Tom's Hardware, X snippets | HDD/NAND shortage is no longer hidden; AI inference and cloud storage are pulling supply away from consumers and PCs. | AI data growth -> exabyte storage demand -> constrained HDD/NAND supply -> pricing power for drive makers/controllers -> but trade becomes crowded after target hikes and vertical stock moves. | Real bottleneck, now consensus-following / cycle beta. | Keep STX/WDC/SNDK/SMI in watch_for_pullback/cycle_beta, not underdiscovered, unless a narrower controller/interface oligopoly is proven. |
| Penguin Solutions / CXL commentary ecosystem | Company release and CXL article; X snippets | Inference is memory-bound; CXL KV-cache and memory pooling can reduce GPU idle time and unlock long-context / agentic AI inference. | Inference workload -> KV cache memory wall -> CXL pooled memory tier -> validated system suppliers and interface silicon benefit. | Early bottleneck discovery if focused on CXL controllers/retimers/interface chips; product vendors alone may be execution options. | Upgrade CXL/KV-cache terms; require evidence of design wins, attach rate, and margin capture before core inclusion. |
| Chinese A/H Serenity spillover: 绿的谐波 | 21jingji report | Serenity-attributed post triggered A-share humanoid robot rally; thesis was harmonic reducers, actuator components, and China scale manufacturing. | Physical AI terminal demand -> joint BOM share -> domestic component oligopoly -> mass production leverage. | Social/liquidity-driven theme chase unless independently validated. | Put 绿的谐波 and humanoid component names in observation/pump-risk pool; do not mix robotics supply-chain thesis into AI data-center core radar without separate framework. |
Observation Pool, Not Core Pool
- SIVE / Sivers Semiconductors: high-signal photonics bottleneck candidate, supported by Jabil and GF collaborations, but social heat and valuation expectations are high. Need independent proof of share, gross margin, capacity rights, and product revenue ramp.
- AAOI / Applied Optoelectronics: repeatedly appears in Serenity-derived photonics discussions, but needs a narrow monopoly layer beyond "optical networking demand."
- POET / Celestial AI / Ayar Labs ecosystem names: strategically important to CPO / optical compute narratives, but public equity capture and supplier economics differ by company.
- PENG / Penguin Solutions: useful CXL product evidence; core inclusion requires proof that Penguin captures margin versus being a systems integrator.
- MRVL / Marvell: strong AI connectivity and custom silicon narrative; likely more consensus now. Treat as platform winner/watch_for_pullback unless a specific unavoidable layer is isolated.
- NBIS / neocloud and powered data-center names: contracted power is valuable, but equity value depends on active MW, utilization, customer credit, and financing terms.
- 绿的谐波: humanoid/physical AI component thesis is plausible, but this week's move looks social-liquidity driven; separate from AI data-center radar.