The Market Strength System V4 build is finished. Now comes the harder part. Proving it. For the last couple of weeks I have been rebuilding the system in shadow while V3 continued running normally in production. The construction phase is now complete. V4 includes the new architecture I wanted: Adjusted price RS RS5 and RS10 tactical strength Absolute vs relative performance RS velocity and acceleration Cross sectional percentiles Historical state tracking Transition and sequence analysis Theme breadth Theme diffusion and concentration Swing geometry Volume intelligence Volatility lifecycle And the infrastructure required to preserve market states through time instead of looking only at today's screen. But I want to make one distinction very clear. Build complete does NOT mean production ready. Now I move from development into confirmation. This is where I stop adding things and start asking whether the complete system actually behaves the way it was designed to behave. The next stage is about running V4 through real market cycles in shadow. New market data. New source cycles. New immutable snapshots. Repeated state transitions. Different market conditions. Different sectors and themes. Different winners. Different failures. Then I compare what V4 sees against what the existing V3 system sees. I want to confirm: Data remains clean. Security identity remains correct. Historical states remain immutable. The recurring process works repeatedly, not just once. New modules do not contaminate each other. The same inputs reproduce the same outputs. Changes in classifications can be explained. The system handles new and disappearing stocks correctly. And most importantly, the additional information actually improves the screening process. That last part matters. I did not spend all this time building V4 just to create a more complicated dashboard. More data is not automatically better data. More signals are not automatically better signals. V4 has to demonstrate that the additional information gives me something useful that V3 does not. Until that evidence exists: V3 remains production. V4 remains shadow. No automatic promotion. No assumption that newer means better. The build is finished. Now the system has to earn its place. Price first. Relative strength second. Theme confirmation last. My market journal, not financial advice.

$RMBS KEY READ-THROUGHS FROM RAMBUS Q3 2026 EARNINGS CALL Rambus’s results strengthen the case that AI infrastructure spending is broadening beyond accelerators into CPU-based servers, conventional memory modules and increasingly complex silicon designs. The most actionable signal is the expected acceleration in product revenue to 34% year-over-year growth in Q4 despite minimal MRDIMM contributions. However, the call also exposes important differences between module demand, memory-bit demand, supplier profitability and actual end consumption. Customers are preserving module counts while reducing module capacity, semiconductor input costs are pressuring Rambus’s margins, and earlier ordering is improving backlog visibility without resolving uncertainty about downstream inventories. The strongest positive implications concern conventional DDR5 interface suppliers, server processors and advanced semiconductor IP. The strongest negative implications concern aggressive MRDIMM adoption assumptions, unconstrained server-DRAM content forecasts and valuations that treat precautionary ordering as equivalent to final demand. Countries below indicate corporate domicile; magnitude refers to the affected business and consolidated earnings exposure, not a forecast share-price movement. MEMORY INTERFACE SEMICONDUCTORS: CONVENTIONAL DDR5 CONTENT EXPANSION IS ALREADY MONETIZING, WITHOUT REQUIRING MRDIMM (READ-THROUGH 1) Affected companies: Montage Technology (688008: China) and Renesas Electronics (6723: Japan). Direction: positive. Magnitude: high strategic relevance for Montage’s concentrated interconnect business; meaningful for Renesas’s memory-interface operations but more diluted at the consolidated level. The critical evidence is Rambus’s expectation of approximately $130 million in Q4 product revenue, up 34% year-over-year, with management attributing the growth primarily to existing RCD generations and companion chips. Seraphin described the outlook as “34% year-over-year growth based on the core, I would say, RCD and companionships.” MRDIMM remains “very modest.” This is evidence that the conventional memory-interface opportunity is generating substantial revenue before the more speculative architectural transitions become material. The transmission mechanism is a combination of more populated memory modules, higher-speed RCD generations and increased supporting-chip content. Management highlighted the progression from 8 to 12 to 16 memory channels. A move from 12 to 16 channels increases potential channel count by 33%, although realized module demand depends on population rates and platform mix. RCDs, power-management devices, SPD hubs and temperature sensors can benefit from this expansion without requiring proportional growth in DRAM capacity per module. Renesas’s disclosed portfolio spans these categories, while Montage’s operating exposure is heavily concentrated in interconnect chips. Montage provides unusually strong independent corroboration. Its October 20 preliminary announcement projected Q3 interconnect-chip revenue growth of approximately 49%, identified higher shipments of DDR5 interface and supporting chips, and anticipated further demand growth in Q4. Interconnect chips represented approximately 98% of its preliminary quarterly revenue, although that category is broader than RCDs alone. Rambus therefore confirms an already-disclosed industry expansion rather than providing the initial evidence for a Montage earnings surprise. The margin comparison is particularly important. Rambus attributed its approximately 60.4% product margin primarily to supplier-cost pressure, while Montage projected an interconnect gross margin of 68.0% to 68.5%, improving both sequentially and year-over-year. Product mixes and accounting definitions differ, so the figures are not a clean competitiveness comparison. Nevertheless, they demonstrate that Rambus’s margin pressure should not automatically be extrapolated across the entire interface-chip group. The near-term catalyst is Q4 shipment conversion and the next disclosures on higher-generation RCD mix and companion-chip adoption. The longer-duration driver is rising interface content per server as platforms add channels and memory speeds increase. The cross-portfolio implication is favorable for qualified interface suppliers across China, Japan and the US, but the call does not establish regional share shifts or identify a beneficiary of any specific customer allocation. MEMORY INTERFACE SEMICONDUCTORS: MRDIMM ADOPTION SHOULD BE MODELED MORE CONSERVATIVELY THAN THE ADDRESSABLE-MARKET NARRATIVE (READ-THROUGH 2) Affected companies: Renesas Electronics (6723: Japan), Montage Technology (688008: China) and Micron Technology (MU: US). Direction: negative for aggressive adoption and product-mix assumptions. Magnitude: moderate for MRDIMM-specific revenue expectations; limited for immediate consolidated earnings where current contributions remain small. Management’s clearest qualification was that “the ramp will be shallower in our opinion than the other generations of products.” The reasons were platform-specific introductions, uncertainty about the proportion of systems adopting MRDIMM rather than conventional DIMMs, customer qualification requirements and unfavorable memory economics. Rambus’s strong Q4 growth explicitly does not require a substantial MRDIMM contribution. The transmission mechanism is delayed monetization of richer module content. MRDIMM introduces multiplexing and additional buffering functionality, creating opportunities for higher interface-chip revenue and differentiated high-bandwidth memory products. However, suppliers realize that opportunity only when compatible processors ship, system vendors qualify the configurations and customers select MRDIMM over conventional alternatives. A large addressable market cannot substitute for those adoption steps. Renesas has direct exposure through its MRCD and multiplexed data-buffer portfolio, alongside supporting PMICs and other module components. Its July announcement placed production availability for its third-generation MRDIMM chipset in H2 2027. Rambus’s commentary does not contradict that schedule or establish a delay to Renesas’s announced plan. It does reduce confidence in scenarios that assume rapid, broad-based adoption across platforms before customer economics and qualifications are established. For Montage, the implication is to separate the already-strong conventional DDR5 franchise from incremental MRDIMM upside. For Micron, it argues for caution around the timing of premium server-module mix benefits rather than a reduction in aggregate memory demand. Revenue can remain strong in conventional modules while the richer architecture develops more slowly. The competitive question is unresolved. An analyst noted that a competitor had secured the initial MRDIMM platform, but Rambus’s response emphasized market timing rather than demonstrating that it had displaced that supplier. Consequently, the call provides no basis for assuming an immediate competitive loss at Renesas or Montage. Equally, early participation in a platform does not guarantee that the entire market will adopt MRDIMM rapidly. The near-term trading implication is reduced justification for capitalizing a steep MRDIMM ramp into 2027 estimates without shipment evidence. The longer-duration opportunity remains intact. Relevant catalysts are production qualifications, disclosed module shipments and actual adoption across successive CPU platforms during 2027. This is a timing caution, not evidence of canceled platforms or a formal reduction to peer guidance. DRAM PRODUCERS: MODULE GROWTH IS NOT MEMORY-BIT GROWTH, AND HIGH PRICES ARE ALREADY CHANGING CONFIGURATIONS (READ-THROUGH 3) Affected companies: Micron Technology (MU: US), SK hynix (000660: South Korea) and Samsung Electronics (005930: South Korea). Direction: negative for unconstrained conventional server-DRAM content assumptions, while near-term pricing can remain positive. Magnitude: potentially meaningful within server DRAM, but not sufficient to establish a consolidated earnings downgrade. The most important quotation is: “it’s not that people decrease the number of modules, but they tend to decrease the capacity of each one of these modules.” Management attributed that behavior to memory economics. This is a direct indication that customers are protecting bandwidth-oriented configurations while reducing capacity to manage system cost. The transmission mechanism differs materially between interface-chip suppliers and DRAM producers. Interface-chip demand is substantially linked to the number and configuration of modules. DRAM-bit demand also depends on capacity per module. More modules can therefore support strong RCD shipments while generating much less memory-bit growth. Illustratively, 20% growth in modules combined with a 15% reduction in average capacity produces only 2% growth in installed bits. Those percentages are a sensitivity, not an estimate of the observed market. This is not a high-conviction outright bearish call on memory earnings. Lower capacity growth can coexist with strong pricing, richer product mix and rising profitability when supply remains constrained. The read-through is specifically negative for models that simultaneously assume unrestricted bit-content growth, rapidly rising prices and no customer adaptation. It also should not be transferred to HBM, whose configurations and demand drivers differ from conventional CPU-attached memory. The novelty is limited but useful. Micron’s September earnings call had already described strong server-unit growth accompanied by a modestly lower rate of content growth than previously expected because of tight memory supply. Rambus provides independent, module-level corroboration of that adjustment. It is therefore stronger evidence for the mechanism, not a newly discovered reason to cut Micron’s already-reported quarter. Near-term catalysts are SK hynix’s October 28 results and Samsung’s October 29 earnings discussion, particularly their commentary on conventional DRAM shipments, server configurations and pricing. Samsung’s preliminary profit guidance is already public; the incremental information is the underlying volume, mix and outlook discussion. For Micron, the relevant catalyst is the next update to fiscal 2027 shipments and content assumptions. The longer-duration implication is that memory spending and useful installed capacity may diverge materially during a tight supply cycle. Across a portfolio, interface-chip strength should not be treated as an interchangeable confirmation of DRAM-bit growth. Conversely, lower capacity per module should not be interpreted as evidence that server deployments or AI infrastructure investment are collapsing. SERVER PROCESSORS: AGENTIC AI IS EXPANDING CPU-ATTACHED MEMORY DEMAND RATHER THAN SIMPLY REPLACING CPU INFRASTRUCTURE (READ-THROUGH 4) Affected companies: Advanced Micro Devices (AMD: US) and Intel (INTC: US). Direction: positive. Magnitude: meaningful for EPYC and Xeon demand; moderate incremental consolidated read-through because pricing, market share, supply and other businesses remain important. Rambus identified orchestration, persistent workloads, tools management and memory management as drivers of higher CPU requirements. Seraphin stated that “the majority of the growth is coming from CPUs with, with standard memory subsystems.” He also described the CPU-to-GPU ratio as moving in Rambus’s favor. The transmission mechanism is additional CPU-based infrastructure around accelerated workloads. Agent execution requires coordination, data preparation, application logic and access to memory outside the accelerator’s local HBM. Those requirements can increase CPU deployments and favor platforms with greater memory bandwidth, more channels and stronger performance under memory-intensive workloads. AMD and Intel monetize the processor demand; their qualified memory ecosystems help determine how quickly that demand becomes deployed systems. The distinction between head nodes and the broader CPU estate is important. Rambus characterized LPDDR-based configurations as a smaller portion of the market and conventional DIMMs as the dominant source of growth. This weakens the assumption that growth in specialized accelerator-associated CPU configurations necessarily displaces the broader conventional server opportunity. However, standard DDR memory is not exclusive to x86. The call supports the end market served by EPYC and Xeon; it does not establish that Arm-based server processors are losing share. Nor does it establish an AMD-versus-Intel share outcome. Rambus’s 34% product-growth outlook also cannot be translated into 34% CPU shipment growth because memory channels, chip content and supplier share contribute to its revenue. The signal corroborates existing company commentary. AMD’s August release described accelerating EPYC demand, while Intel’s July release reported strong Data Center and AI growth and highlighted agentic infrastructure deployments. Rambus adds a later observation from the memory subsystem, strengthening confidence that the CPU opportunity is supported by actual platform requirements rather than only processor-vendor positioning. The most immediate catalysts are Intel’s October 29 earnings and AMD’s November 3 earnings. The key variables are CPU demand, shipment capacity and the 2027 outlook, rather than headline data-center revenue alone. AMD’s data-center business also contains accelerators, while Intel’s consolidated outcome includes foundry and other exposures. Over the longer term, the structural positive is a broader CPU workload base; the limiting factors are utilization improvements, workload consolidation and offload to other processors. ENTERPRISE SERVER HARDWARE: PHYSICAL MEMORY-SUBSYSTEM DEMAND SUPPORTS THE REFRESH CYCLE, BUT DOES NOT VALIDATE HEADLINE REVENUE GROWTH AS UNIT GROWTH (READ-THROUGH 5) Affected companies: Dell Technologies (DELL: US) and Hewlett Packard Enterprise (HPE: US). Direction: positive for conventional server demand, with important limits on revenue and margin extrapolation. Magnitude: moderate at the server-business level and more diluted at the consolidated level. The supporting commentary is Rambus’s observation that module counts are increasing even as capacity per module declines, combined with its expectation that conventional CPU systems will generate most of the growth. These are physical configuration signals, not merely higher dollar spending on memory. The transmission mechanism is demand for complete CPU-based systems used in enterprise refreshes, orchestration and inference-related infrastructure. Dell and HPE can monetize that demand through servers, configuration options, support and related infrastructure. Their exposure is downstream through the server ecosystem; Rambus did not identify either company as the recipient of a particular order or establish a direct customer relationship. The non-consensus distinction is between genuine infrastructure demand and the magnitude of reported dollar growth. Rambus’s evidence makes an explanation based entirely on memory-price inflation less persuasive. Nevertheless, higher channel counts can increase modules per server without an equivalent increase in server units. The call therefore supports physical infrastructure activity but cannot independently establish OEM unit growth. This distinction matters against already-reported results. Dell’s September release showed traditional servers and networking revenue increasing 122% year-over-year. HPE also reported strong server growth, and its earnings discussion identified both robust demand and supply constraints. Rambus corroborates the market direction, but provides no basis for applying its own growth rate to those companies or treating their revenue increases as shipment increases. Margin transmission is also conditional. Higher memory prices can be passed through, absorbed or managed through configuration changes and procurement arrangements. HPE’s recent profitability improvement demonstrates why a blanket conclusion that memory inflation must compress OEM margins would be incorrect. Conversely, continued supply tightness can delay shipments and increase working-capital requirements even when pricing is effective. The near-term catalysts are the upcoming results covering the October 2026 fiscal quarters, especially unit commentary, traditional-server order conversion, gross profit per system and component availability. The longer-duration positive is a broader role for conventional servers within AI-enabled enterprise infrastructure. The principal cross-portfolio risk is that OEM dollar growth, component shipments and end-system deployment are different measures and should not be counted as independent confirmations of the same growth rate. SEMICONDUCTOR IP AND DESIGN SOFTWARE: CUSTOM SILICON CREATES EARLIER MONETIZATION OPPORTUNITIES THAN PRODUCTION CHIP SHIPMENTS (READ-THROUGH 6) Affected companies: Synopsys (SNPS: US) and Cadence Design Systems (CDNS: US). Direction: positive. Magnitude: meaningful for advanced interface and memory IP; modest to moderate at the consolidated level because these are broader businesses and competitive allocation remains unknown. Rambus reported additional Tier 1 next-generation HBM design wins, increasing interest in PCIe controller and switch IP, and deeper architectural engagement before standards are finalized. Management described customers as “designing increasingly differentiated architectures and tailored silicon optimized for their own environments.” The CFO placed silicon IP growth toward the upper end of the previously communicated 10% to 15% range. The transmission mechanism is increased demand for validated design building blocks and verification as customers develop more complex custom systems. HBM interfaces, high-speed connectivity, security and integration must be addressed before a chip can reach production. Synopsys and Cadence have direct product exposure to these requirements, rather than merely sharing a broad AI label. The timing advantage is important. IP licenses and associated engineering activity can monetize during design, before the customer’s finished chip reaches meaningful shipment volume. Design spending can therefore remain strong even while system adoption is constrained by memory supply or platform qualifications. However, licensing terms differ, and design wins should not automatically be treated as recurring royalties on every future chip shipment. The opportunity is reinforced by previously disclosed peer performance. Synopsys’s August results described a return to year-over-year growth in Design IP and specifically identified memory-interface, wired-interface and security IP within the business. Cadence released its Q3 results on October 26, reporting strong double-digit growth across product categories and record backlog. Rambus is consequently corroborating an established design-demand cycle, not predicting Cadence’s already-announced quarter. There are clear limits to the read-through. Rambus’s wins could represent competitive displacement in individual programs, so they do not establish that Synopsys or Cadence won the same business. The 35% growth in Rambus’s contract-and-other line is also not an appropriate growth assumption for the entire IP industry, since Rambus’s IP revenue spans multiple accounting categories. The near-term catalysts are Synopsys’s fiscal Q4 results and subsequent 2027 outlooks, with emphasis on IP bookings, program breadth and integration intensity. The longer-duration positive is greater engineering content per custom system. Rambus’s ChipAgents collaboration does not establish displacement of incumbent EDA tools or a quantified reduction in industry design spending; neither conclusion deserves financial credit from this call. CROSS-PORTFOLIO SUPPLY-CHAIN RISK: EARLIER ORDERING IMPROVES PLANNING BUT DOES NOT PROVE STRONGER FINAL CONSUMPTION (READ-THROUGH 7) Affected companies: Rambus (RMBS: US), Montage Technology (688008: China) and Renesas Electronics (6723: Japan). Direction: negative for the evidentiary quality of backlog-based growth assumptions, not evidence of an imminent demand decline. Magnitude: moderate for visibility and valuation; potentially more material for focused interface-chip suppliers if ordering later normalizes. The decisive exchange concerned downstream inventories. Management acknowledged, “It’s very hard to determine,” before explaining that customers are placing orders earlier. Seraphin summarized the distinction as: “we don’t see people building inventory, but we see people placing backlog.” That is narrower than confirmation that downstream inventories are lean. The transmission mechanism is a change in order timing. Customers responding to longer supplier lead times can extend procurement horizons without increasing the same proportion of final consumption. Suppliers then see stronger backlog and may build inventory or reserve capacity. If customers subsequently shorten those horizons, bookings can weaken even while underlying end demand remains healthy. Precautionary ordering can therefore amplify both the apparent upturn and a later normalization. Rambus’s own actions illustrate the balance-sheet consequence. Inventory increased approximately $12 million during Q3, another increase is planned for Q4, and management is using liquidity to protect supply. This strategy has worked operationally: management stated that supplier constraints did not affect Q3 or the Q4 outlook. The negative read-through is therefore not a reason to assume an immediate supply-driven miss. It is a reason to distinguish current execution from the durability of the visible order book. The risk cannot be transferred mechanically across suppliers. Contract terms, cancellation provisions, customer deposits, product breadth and inventory ownership differ. In particular, memory producers’ disclosed longer-term supply commitments should not be treated as equivalent to Rambus’s unspecified product backlog. Nor does Rambus’s limited visibility prove that Montage or Renesas has the same customer behavior. The near-term catalyst is the Q4-to-H1 2027 transition, when seasonal patterns, inventory coverage and earlier orders begin to interact. Shipment conversion, cancellations and inventory efficiency will be more informative than backlog growth alone. Over the longer term, suppliers that can maintain availability without excessive inventory commitments should earn better returns on capital. The cross-portfolio implication is to avoid counting stronger CPU demand, higher module shipments and longer component backlogs as 3 fully independent signals when they can reflect different stages of the same procurement cycle.

$SOFI just delivered one of the most INSANE earnings reports I’ve ever seen. • Revenue beat by 9% ✅ • EPS beat by 6% ✅ • 2026 guidance raised ✅ • 2027 revenue & EPS guidance WAY above estimates ✅ And the market is still valuing this company at just 15x forward earnings.