MSA (Memory Sparse Attention) represents our significant exploration in the field of long-term memory. It stands as the first end-to-end long-term memory framework for large models to genuinely achieve a 100M context length. Interestingly, as the memory length scales from 16K to 100M, the model's performance score decreases by a mere 9%, demonstrating highly robust scalability. Main contribution： 1，We propose MSA, an end-to-end trainable, scalable sparse attention architecture with a document-wise RoPE that extends intrinsic LLM memory while preserving representational alignment. It achieves near-linear inference cost and exhibits < 9% degradation even when scaling from 16K to 100M tokens. 2，We introduce KV cache compression to reduce memory footprint and latency while maintaining retrieval fidelity at scale. Paired with Memory Parallel, it enables high-throughput processing for 100M tokens under practical deployment constraints, such as a single 2×A800 GPU node. 3，We present Memory Interleave, an adaptive mechanism that facilitates complex multi-hop reasoning. By iteratively synchronizing and integrating KV cache across scattered context segments, MSA preserves cross-document dependencies and enables robust long-range evidence integration. 4，Comprehensive evaluations on long-context QA and Needle-In-A-Haystack benchmarks demonstrate that MSA significantly outperforms frontier LLMs, state-of-the-art RAG systems and leading memory agents. Welcome to feedback: We are looking for passionate talents to join our team! If you are interested in our work and vision, please don't hesitate to send us an email at evermind@shanda.com.

- Lightmatter officially introduced Passage L20 CPX on September 17 and joined the Open CPX MSA, describing the product as the first bidirectional Open CPX optical engine for AI infrastructure. - Passage L20 CPX uses 1311 nm and 1331 nm wavelengths to carry transmit and receive traffic over the same single-mode fiber, cutting fiber and connector requirements by roughly 50% versus conventional UniDi architectures. - The module is designed for 6.4 Tbps Tx plus 6.4 Tbps Rx, or 12.8 Tbps aggregate bandwidth, using 32 optical lanes per direction at a 212.5 Gbps PAM4 line rate with more than 500 meters of SMF reach. - In Lightmatter's 512-GPU scale-up pod model, the architecture reduces fiber count from roughly 131,072 to 65,536 while eliminating approximately 16,000 connectors and more than 200 miles of fiber. - Lightmatter estimates that BiDi could reduce total scale-up interconnect network spending by roughly 15%, although this figure comes from internal modeling and will vary by topology and deployment configuration. - Importantly, this is not yet a volume-production product. Passage L20 CPX evaluation kits are expected to ship to customers in Q1 2027, and the disclosed specifications remain preliminary. - The announcement follows Lightmatter's June entry into NVIDIA's NVLink Fusion ecosystem and its August launch of an OCP CPO architecture effort targeting AI systems scaling from 72 to more than 1,024 nodes.

CHILLING: A Frisco, Texas, high school MSA used a classroom as a prayer hall and posted “more to come.” ⠀ The pictures are from Liberty High School in Frisco ISD. Students stood on rugs between desks and a smart board. Girls in hijabs stood to the side. ⠀ The club caption read, “First meeting of the year inshallah more to come....” ⠀ The same campus had already been in a fight over a now-deleted Liberty MSA intro video. Students called the room “a space for Muslims to pray,” “not a hang out spot,” and said “only Muslims will be allowed in.” ⠀ That video also said girls were not to speak and were to hand their khutbahs to boys. Frisco ISD said the space is a supervised “quiet room,” not a faith-only facility, and that prayer is supposed to fall in non-instructional time unless a parent signs off. ⠀ Liberty High is not a first-time fight. In 2017 then-Attorney General Ken Paxton’s office wrote the district after reports that a classroom was being treated as an Islamic prayer room that shut out other students. ⠀ In February 2026 the Frisco board voted not to add a formal prayer period under Senate Bill 11. Counsel said student religious clubs already exist on every high school campus and cited Texas Education Code 25.091 and 25.154. ⠀ The Muslim Students Association was founded in 1963 at the University of Illinois. Brotherhood-linked organizers were among the early builders. The 1991 Explanatory Memorandum, entered as evidence in the Holy Land Foundation case, listed the MSA among “our organizations.” ⠀ That memo called Brotherhood work in America a “civilization-jihadist process.” Texas later designated the Muslim Brotherhood and CAIR as foreign terr*rist organizations under state law. ⠀ Texas now holds an estimated 400,000 to 500,000 Muslims, with about 150,000 in Dallas-Fort Worth. Frisco, Plano, McKinney, and Irving have been adding mosque capacity as families move into those school districts. ⠀ Other North Texas chapters have already tested the same model. Wylie ISD saw a “Why Islam” lunch booth. Humble ISD MSAs organized mosque visits. Mesquite chapters used school PA time for Ramadan greetings. ⠀ The Middle East Forum counts almost 800 Islamic educational projects nationwide. Liberty High’s own club account is now showing the classroom version. Are you awake yet?!?!?! ⠀ #Texas# #Frisco#

Samtech displayed their Co-Packaged Copper (CPC) set-up in the CPX form factor at Taiwan OCP (See the open CPX MSA here: Bandwidth exits the 6.4T pluggable optical engine via a copper connector interface. There are 128 connector pins per module corresponding to 64 differential pairs (DPs) or 32 lanes of 200G. We expect the CPX form factor to present a compelling opportunity for many copper interconnect companies such as TE, Amphenol and Molex even as we see bigger optics TAM in scale-up networking.

#Tower# Semiconductor and #NewPhotonics# have begun high-volume shipments of laser-integrated, serviceable optical-engine PICs supporting 800G to 1.6T, with CPX MSA-compliant 6.4T NPO chipsets planned for volume production in 2027. Built on Tower's market-leading PH18DA silicon photonics platform, these scale-ready solutions integrate InP lasers with high-performance photonic circuits across FRO, LRO, LPO, XPO and NPO architectures- delivering greater bandwidth density, power efficiency and reliability.

Our U.S. Tour and OCP Global Summit 2026 Takeaways ☕️OCP – System-level inflection
• Decisive shift toward full-stack co-design across hyperscalers
• Nvidia extended networking roadmap new USFP pluggable standard (co-developed with Cisco). Enables 12.8 Tb/s modules
• Showcased PTFE for NVLink Board (inline with our PCB Spec view for 2H27 on Oct 5)
• Microsoft emphasized AI scaling as a system-engineering challenge
• MSFT Unveiled multi-GW Maia 300 deployments from 2027
• Meta maintained pragmatic stance on optics, piloting near-packaged solutions in scale-out domains and Retaining copper for high-density scale-up fabrics ☕️OCP – Optical scale-up and power integration
• Broadcom and OpenAI advanced OCI-MSA multi-rack optical scale-up, highlighted hybrid topologies: high-bandwidth copper for tensor parallelism + optical links for expert parallelism
• GEN2 and GEN3 timelines: 400G in 2027, 800G in early 2030s 
• Google shifted focus to power, positioning datacenters as grid-integrated resources
• 1 GW demand-response capacity with U.S. utilities
• LVDC roadmap: from ±400 VDC Mt Diablo racks to native 800 VDC distribution
• Power engineering now dictates deployment velocity as much as silicon performance
• AMD reinforced CPU-dense trajectory of agentic AI
• Announced Helion CPU racks reaching 32,768 EPYC cores
 🚌 U.S. Tour — Supply Chain and Compute Economics
• No near- to mid-term power-supply risk, Supported by gas-based generation and remote locations 
• Liquid-cooling infrastructure is the bottleneck
• Rack pricing continues to rise (GB300 US$5.5–6.5m; VR200 US$8m), reflecting strong compute demand
• Some believe Inference economics challenging: GPU utilization 20–30%, memory and cooling costs rising
• For agentic AI, model shifts CPU-to-accelerator ratios from 1:4–1:8 to 1:1 → doubling CPU TAM #OCP# #AI# #MSFT# #GOOG# #NVDA# #VR200#

- Tower Semiconductor ($TSEM) and NewPhotonics have begun high-volume shipments of laser-integrated silicon photonics optical engines, with current production products supporting 800G and 1.6T. - The key technology is $TSEM's PH18DA SiPho platform, which heterogeneously integrates InP devices to combine lasers, SOAs, modulators, and photodetectors on a single PIC. - NewPhotonics' NPG102 uses a 200G-per-lane architecture for 800G and 1.6T PICs, targeting optical architectures ranging from FRO/LRO/LPO and XPO pluggables to NPO. - The more important next step is the 6.4T NPC505 chipset, a CPX-MSA-compliant NPO solution scheduled for volume shipment in 1H27, targeting both scale-out and scale-up AI interconnects. - $TSEM says NewPhotonics is the first customer to bring heterogeneous laser-integrated optical engines on PH18DA into high-volume production, although specific customers and shipment volumes were not disclosed. > The most important phrase here is “high-volume shipment.” This suggests $TSEM's silicon photonics platform has moved beyond PDKs, tape-outs, and prototypes into actual volume manufacturing of laser-integrated PICs. > Integrating the laser, SOA, modulator, and detector around a single PIC can reduce component count and optical-engine assembly complexity. As AI interconnects evolve from pluggables toward NPO and eventually CPO, the value of a highly integrated, scalable SiPho foundry platform could increase materially. > From an investment perspective, the key point is that $TSEM is increasingly becoming more than a legacy analog foundry and is beginning to establish itself as a merchant SiPho foundry tied directly to AI optical connectivity volume production. With 800G/1.6T already shipping and 6.4T NPO targeted for 1H27, the $TSEM SiPho story is gradually shifting from future optionality to a production ramp story.

Semtech (SMTC US) Key Takeaway • Rating: Maintain Buy, Target Price $162 (22x FY28E EPS) • F2Q revenue $341.9mn (+33% YoY, +17% QoQ), beat consensus. Non‑GAAP EPS $0.71 (vs consensus $0.61) • Non‑GAAP gross margin record at 54.5% (+150bps QoQ), driven by data‑center • F3Q guidance tops market estimates: revenue $410mn; non‑GAAP gross margin 58.3%; EPS $1.02‑1.08 • AI‑related demand ~3x current supply capacity；growth bottleneck: SiGe / InP wafer & OSAT packaging constraints • 1.6T CopperEdge ACC entered mass shipment to major hyperscalers in F2Q (moved from sampling to material revenue) • 1.6T FiberEdge TIA‑driver initial shipments started; target >50% share in target 1.6T market by end‑FY27E • Post HieFo InP laser acquisition: advancing vertical integration of TIA, driver & laser chips for supply resilience • Deepened MSA standardization work for LPO / NPO; secured multiple design wins, expanding long‑term TAM • Pulled explicit segment YoY growth guidance due to supply uncertainty, but multi‑year hyperscaler AI demand outlook remains solid • Forecast FY27E / FY28E / FY29E EPS: $3.60 / $7.40 / $11 #SMTC# #Semtech# #1#.6T #Optical# #AIInfrastructure#

A visitor at CIOE in Shenzhen this morning asked a manager on the $ALAB stand whether the value of retimers gets eroded as the industry moves toward optics. I went and put the answer he got next to a specification the company is already helping to write. The answer, as he wrote it down: "The number of high-speed links is going to increase exponentially. Whether it's copper, re-timed optics, or linear optics, all of them are going to increase exponentially. If you think this is crazy right now, 2027 by all accounts is supposed to be where it goes even faster after that. So you cannot go wrong by investing in any of these. You just have to invest in at least one of them." The question was about value per link. The answer is about the number of links. Those are different quantities and only one of them was asked about. I am not going to make much of that on its own. It is a stand manager, unnamed, no title given, at a trade show, and none of that is company guidance. What is worth something is that Astera Labs has already put its name to the thing that decides the answer. Astera Labs is a contributor member of the Open CPX MSA, formed in March this year. What that group is writing, in the words of Credo's own release when it joined on 31 August: common specifications covering "mechanical, thermal, electrical, optical, and management interfaces" for "an optimized optical engine with a defined pluggable socket and electrical connector system." A defined pluggable socket is the whole argument in five words. Once the socket is specified, what goes in it becomes swappable, and swappable is the mechanism by which value per unit gets competed away. That is not a prediction, it is what the MSA is for. The founding members say so directly: the stated goal is to reduce power, cost and latency. The roster is the part I would sit with. Founding members are Ciena, $COHR, $MRVL, Molex, Samtec and TeraHop. Contributors include $ALAB itself, $INTC, $QCOM, $LITE, Source Photonics, TE Connectivity, Accton and Viavi. $CRDO announced it had joined on 31 August and had still not appeared on the published list when I read it this morning. Two names are on neither list: $NVDA and $AVGO. That is the shape of it. The companies without a proprietary scale-up interconnect are standardising the socket between them. The two companies that do have one are not in the room. NVLink is NVIDIA's and nobody sells into it. An open socket is the tool everyone else uses to build domains big enough to compete with it, and the price of that tool is that nothing inside the socket stays scarce for long. So the honest version of the booth answer is that the manager is probably right about volume and the question was about something else. Link counts go with the size of the coherent domain, and in all-to-all topologies links grow faster than accelerators do. Value per link is a separate fight, and the specification his own company is contributing to is the venue for it. Worth saying plainly: absence from a roster is not opposition to a standard, and being in the room is not the same as losing pricing power. Astera is contributing rather than founding, which is a real distinction in how these bodies work. For scale, everything in this chain is up today, which is a CPI day and not a verdict on any of this. $ALAB 294.53, $COHR 310.43, $MRVL 238.11, $LITE 969.09, $CRDO 163.69, all mid-session. What would change my read: watch whether the first Open CPX specification puts the retiming function inside the socket or leaves it on the board outside. If re-timing lives in the module, the module vendors take that content. If it stays outside, the socket standardises the optics and leaves the signal conditioning where it is, and the booth answer holds for reasons better than the one he gave.