404K SEMI-AI Evening Brief 2026-07-29 — Financing Constraints Spread: Cloud Capex Scrutiny Extends to Memory and Equipment Chains
目录
Pre-Market Highlights
Full AI/Semiconductor Supply Chain
AI Models/Applications and Capital Expenditure
CSP/Cloud Capex and AI Data Centers
GPUs/CPUs/ASICs and Server Networking
HBM/DRAM/NAND, Advanced Packaging, and Testing
Optical Communications, Passive Components, and Power Semiconductors
Internet/Platforms
Software/SaaS
Consumer Electronics/Smart Vehicles
Overview
404K | 2026-07-29
Pre-Market Highlights
AI infrastructure demand shows no clear signs of cooling; what has changed is the market’s validation sequence. Investors are first assessing whether cloud providers can continue financing capital expenditures, and then whether chip, memory, testing, and optical-interconnect orders can be realized. The divergence between semiconductor earnings and share prices indicates that a single quarter’s results are no longer sufficient to support valuations independently.
Divergence within the supply chain continues. SK hynix, Teradyne, Corning, and Bloom Energy delivered positive signals on demand or earnings, while spot memory and advanced packaging are also improving. On the downside, cloud providers’ free cash flow, project returns, the lumpiness of equipment spending, and incremental capacity are all beginning to be priced in.
The focus in software and platforms is shifting to how AI is monetized. Enterprises are reallocating workloads among long-term contracts, on-demand usage, and open models. Tax software and SAP, meanwhile, show that whether AI revenue can cover sustained investment will determine whether margin pressure is temporary or structural.
Full AI/Semiconductor Supply Chain
AI Models/Applications and Capital Expenditure
OpenAI and Anthropic
1) As of July 2026, the two labs had combined annual recurring revenue of approximately $120 billion, while the scenario involving 18 to 22GW of incremental capacity would still require nearly $800 billion in additional revenue. Compute constraints must therefore continue to be validated by paid demand.
2) Anthropic shifted to purchasing physical books in 2024, spending millions of dollars to acquire and scan millions of books into its database. Reportedly, purchasers could place single orders ranging from thousands of books to as many as 1 million. Lawful format conversion reduces copyright risk, but identification of rare editions, public archiving, and long-term access remain counterevidence.
“Even if the long-term direction is correct, the 2027 construction timeline may still precede actual demand.”
Moonshot AI is accelerating both fundraising and commercialization. Reportedly, the company raised $3.5 billion at a $35 billion valuation, above its initial target of $1 billion to $2 billion. Annual recurring revenue rose from $200 million in April to $300 million in June, while daily sales increased by at least 6x after the launch of Kimi K3. Faster revenue growth is improving the financing narrative, but the higher valuation also raises the threshold for subsequent execution. The next metrics to watch are enterprise payments, overseas revenue, and cash burn.
Enterprise model procurement Contract-based purchasing is giving way to more flexible workload routing. Enterprises are combining frontier and open models based on task complexity, cost, latency, and control requirements, while switching between self-hosted and managed platforms. When on-demand pricing is higher, cloud platforms benefit in the near term, but predictable demand remains better suited to reserved capacity, and long-term contracts will not disappear. Key metrics are the share of on-demand usage and cost per task.
“Enterprises are increasingly combining frontier and open models, routing workloads according to task complexity, scale, cost, performance, latency, and control requirements.”
Generative AI adoption in India Goldman Sachs estimates that generative AI can perform 9% to 17% of tasks undertaken by the non-farm workforce; 8% to 12% of jobs face displacement, while 42% to 48% are more likely to be augmented. Under the base case, annual labor-productivity growth increases by approximately 0.4 percentage points over the next 10 years. Software development, customer service, and knowledge-process outsourcing will be affected first, with realization dependent on the adoption sequence, job reallocation, and power supply for data centers.
AI capex validation The demand debate has shifted from “whether to build” to “who will pay and when returns will materialize.” The U.S. plans to add 37GW of data-center capacity in 2027, versus approximately 44GW of total capacity in 2025. If inference demand, utilization, and revenue per GW fall below expectations, cloud-service price reductions, higher depreciation, and weakening free cash flow will emerge before chip orders are revised downward.
“Hyperscalers face a prisoner’s dilemma: slowing collectively could protect cash flow, allow time to learn from hardware iterations, and reduce asset-obsolescence risk, but any individual company that slows first could lose access to power, chips, talent, model partners, and cloud customers.”
CSP/Cloud Capex and AI Data Centers
Cloud-capex buyers Sources of funding have become a leading indicator for orders. For 5 major cloud providers, spending on property, plant, and equipment as a percentage of operating cash flow has risen from 40% for the period ended March 2024 to 75% over the latest 12 months; Amazon stands at 102% and Oracle at 174%. Once capital expenditure exceeds internally generated cash, bond yields, equity-financing windows, and asset sales will all flow through to semiconductor orders.
“A company that once repurchased its own shares is now selling its own shares to purchase property, plant, and equipment.”
Google spent $44.9 billion on property, plant, and equipment during the quarter, resulting in a $5.9 billion cash shortfall after the expenditure, and raised $49.6 billion through instruments including common stock and mandatory convertible preferred shares carrying a 6.25% annual dividend. Demand has not disappeared, but capex has shifted from a corporate-treasury issue to a capital-markets issue. Interest expense, potential equity dilution, and depreciation will collectively raise the return threshold. The next question is whether trailing-12-month free cash flow and share repurchases can recover.
Bloom Energy Q2 revenue reached $1.0654 billion, up 165.5% year over year and 41.8% quarter over quarter. Full-year revenue guidance was raised to $3.9 billion to $4.2 billion, while operating cash flow turned positive at $226.4 million. Rapid deployment of power capacity is giving the company access to AI data centers, and Nebius has also shifted from combustion-based equipment to Bloom Energy. Approximately $20 billion of backlog, customer prepayments, and actual installation progress still require verification through regulatory filings.
Celestica Citi said demand remains above supply. The company raised FY2026 sales guidance from $19 billion to $20.5 billion and expects at least $34 billion in FY2027. Citi raised its FY2027 EPS forecast by 45% to $20, maintaining its Buy rating and $415 price target. Growth is driven by 800G, AI compute, and white-box solutions, while the principal risks are customer concentration, a capex pullback, and pricing competition.
GPUs/CPUs/ASICs and Server Networking
NVIDIA The current debate is not whether GPU demand is strong, but the threshold required for further growth. If the AI trade continues at a similar intensity, annual revenue would need to reach $640 billion for the company to grow another 100%. A 22x forward P/E appears undemanding, but it requires end-user paid demand, cloud-provider returns, and financing capacity all to materialize in tandem. The formal incorporation of financing into commercial arrangements also indicates that the cost of funding GPU purchases has become an industry variable.
MediaTek Morgan Stanley expects 3Q26 revenue to decline approximately 5% quarter over quarter, but 3nm TPU revenue to ramp significantly from 4Q26, with 2027 TPU revenue guidance potentially raised from $7 billion to $12 billion to $12 billion to $15 billion. Supply-chain checks indicate 3 million 3nm TPUs in 2027. The 2nm product is expected to enter mass production in 2028, but remains constrained by EMIB-T yields, CoWoS costs, and warpage in large dies.
Accton Technology Celestica’s strong guidance provides cross-validation for the switch supply chain: 800G continues to scale, while 1.6T shipments have begun and are expected to continue into 2027. Morgan Stanley maintains its Overweight rating, top-pick designation, and NT$3,800 price target. The factors that will actually determine order realization are component supply, the pace of migration between switch specifications, and penetration of custom AI-accelerator modules.
Broadcom Morgan Stanley expects MediaTek’s 2027 TPU shipments could approach Broadcom’s, but Broadcom may still retain approximately 80% of long-term revenue share due to its higher average selling prices and gross margins. This means ASIC competition depends not only on chip volumes, but also on per-chip value, packaging yields, and customer reliance on end-to-end design services. If average selling prices weaken, a volume advantage may not be sufficient to preserve revenue share.
SPARC pre-silicon verification Processor-security verification is advancing from “detecting leakage” to “identifying root causes.” The framework uses information-flow tracking, shadow logic, and statistical testing to flag secret-dependent switching activity, then maps leakage to hardware signals and software instructions. Commercialization, accuracy, and tool customers have not yet been disclosed; for now, the primary implication is rising demand for EDA security verification.
HBM/DRAM/NAND, Advanced Packaging, and Testing
SK hynix Q2 revenue reached KRW79.32 trillion, up 257% year over year and 51% quarter over quarter. Operating profit was KRW60.54 trillion, representing a 76% operating margin. HBM4 is already in mass production for major customers, with yields and quality approaching those of mature HBM3, and the company has completed long-term supply negotiations with more than 10 customers. It is advancing the transition to 321-layer NAND and developing 15nm DRAM. High profitability will attract supply, so the next focus is capacity for advanced products and capex discipline.
“HBM4’s competitiveness depends not only on achieving the required performance, but also on the ability to supply at scale with stable yields and consistent quality.”
Seagate Technology Emerging cloud providers are beginning to colocate data and compute, expanding hard-drive demand beyond traditional hyperscale-cloud customers to the neocloud segment. Management said that even some of the largest neocloud providers need substantial volumes of data transferred into their environments. The investment implication is that, as AI infrastructure matures, storage is no longer merely a back-end procurement item, but a colocated resource affecting compute utilization and data sovereignty.
Memory spot market Both DRAM and NAND prices have rebounded, but transaction volumes remain insufficient. The average spot price of DDR4 1Gx8 3200MT/s rose from $41.50 to $42.08, up 1.39%; the spot price of 512Gb TLC wafers increased 1.69% to $19.250. Inquiries and inventory restocking by a single module manufacturer can lift prices, but do not yet demonstrate that broad-based demand has taken over.
Teradyne Revenue reached $1.329 billion, up 104% year over year, with semiconductor-test revenue of $1.122 billion. HBM, advanced packaging, and high-capacity NAND are collectively increasing test intensity. Q3 revenue guidance falls back to $1.2 billion to $1.3 billion, suggesting equipment orders are more likely to arrive in concentrated bursts. Accounts receivable also rose to $1.11 billion and should be tracked alongside customer installations and collections.
“Demand for test equipment can arrive in concentrated bursts: customers may spend heavily upfront and then pause investment once sufficient installed capacity is in place.”
Micron FY2026 third-quarter revenue reached $41.4 billion, 4.5x the $9.3 billion recorded in the prior-year period, and the company guided to next-quarter revenue of $50 billion and EPS of $30.73. During the share-price pullback, its P/E nevertheless fell from 57x to 19x. DRAM revenue growth was driven more by pricing than shipments, while company capex rose to $7.1 billion. High profitability and a strong cycle have already been priced in; the next factors to watch are incremental supply, inventory, and bit shipments.
JCET Group plans to invest RMB7.8 billion in a new Shanghai facility serving AI compute, automotive chips, and next-generation memory. The bottleneck in advanced packaging has expanded beyond capacity alone to include warpage, local topography, and matching the thermal expansion of materials. Whether capacity expansion translates into yields will depend on high-density metrology, process simulation, and reliability validation. Capacity coming online before yields ramp will also weigh on returns.
“The core judgment is that flatness can no longer be described by a single overall bow measurement; it should instead be viewed as a local surface-topography issue that changes with material combinations, thermal history, and process steps.”
Optical Communications, Passive Components, and Power Semiconductors
WIN Semiconductors Q2 revenue reached NT$5.257 billion, up 15% quarter over quarter and 39% year over year, while capacity utilization rose to 65%. PD entered mass production ahead of schedule, and LEO revenue contribution reached double digits earlier than expected. JPMorgan expects AI optical communications to contribute a double-digit share of revenue in 2027 and raised its price target from NT$336 to NT$373. The CW laser project is expected to enter mass production by the end of 2027, while customer concentration and project timing remain risks.
Corning Core sales reached $4.74 billion, with optical-communications revenue of $2.07 billion, up 32%, and AI data-center sales nearly doubling. The agreement with Meta worth up to $6 billion, a multibillion-dollar partnership with Amazon, and capacity-expansion arrangements with NVIDIA have improved visibility. The company is extending into passive photonic components surrounding CPO and near-package optical engines, but its 2030 targets require multiple businesses to deliver simultaneously.
“This opportunity is more difficult to model, and its development trajectory will not be smooth.”
Marvell Technology Data-center architecture may shift from single hyperscale campuses toward distributed clusters connected through optical links. PAM4 handles intra-campus and rack-level connections, while coherent optics connects geographically dispersed facilities. The greater the constraints on power, water, and local communities, the more optical interconnects approach a system-level bottleneck, increasing the value contribution of suppliers such as Corning and Coherent.
Navitas Semiconductor Q2 revenue increased 22% quarter over quarter, while high-power revenue grew 50% year over year. The growth indicates that demand for high-power devices continues to expand, but the available data do not provide order, margin, or customer-mix details. The next question is whether AI data-center power demand can continue translating into gross profit and cash flow. If customer qualification cycles lengthen, revenue growth may also slow before demand expectations are revised downward.
TSMC Revenue and profit grew 36% and 77% year over year, respectively, while annual capex was raised to approximately $60 billion. At the same time, the company guided gross margin down from 67.7% to 65% to 67%, as incremental depreciation and the ramp of advanced processes begin to affect margins. Following the magnitude-7.1 Kumamoto earthquake, TSMC said JASM was unaffected. Tokyo Electron suspended operations, while HORIBA’s restart timing remains undetermined, requiring continued near-term monitoring of equipment-supply continuity.
Internet/Platforms
Google AI investment is beginning to test cloud revenue, capital expenditure, and financing capacity simultaneously. Its model platform can absorb enterprise workloads shifting from frontier models to open models, but equipment spending in the quarter has already created a cash shortfall. Investors are now more focused on Vertex AI usage, returns from the cloud business, and free cash flow than on model-release frequency alone.
Amazon AWS Bedrock allows enterprises to use frontier and open models in a managed environment. Claude is priced identically on Bedrock and Anthropic’s own API, while custom models are available only through provisioned throughput, indicating that the cloud platform’s value lies more in governance, deployment, and workload routing. Capital expenditure has reached 102% of operating cash flow, so validation of returns can no longer be deferred.
“Claude on Bedrock costs exactly the same as Claude on Anthropic’s own API—dollar for dollar.”
Microsoft Foundry supports enterprise multi-model routing while also confronting the choice between on-demand pricing and reserved capacity. If enterprises migrate more simple tasks to lower-cost open models, platform usage may still grow, but the pricing premium for frontier models will be reassessed; capital expenditure, token spending, and customers’ actual savings need to be evaluated side by side.
Meta Large cloud and platform companies continue to compete for power, chips, talent, and model partners. For Meta, slowing first may mean losing access to resources, while continued expansion would increase depreciation and pressure on free cash flow. The market’s focus has shifted from absolute capital expenditure to utilization of incremental compute capacity and whether advertising and model services can generate sufficient cash returns. Depreciation growth and advertising monetization will provide the earliest indications of investment quality.
“Whichever financing method is used, the market will reprice the risk accordingly. This is not manipulation—it is the market functioning efficiently.”
Reddit KeyBanc initiated coverage with an Overweight rating and a $225 price target. It expects revenue of $3.24 billion, $4.3 billion, and $5.46 billion in 2026, 2027, and 2028, respectively, with EBITDA of $1.43 billion, $1.98 billion, and $2.73 billion. Community content can support both advertising and licensing to large models; the key question is whether the value of the data can generate incremental revenue before 2027.
“Reddit’s data is becoming more valuable to large language models and model training, creating a potential upside catalyst extending through 2027.”
AppLovin The number of e-commerce clients increased from approximately 600 in 4Q24 to approximately 9,600 in 2Q26, but average monthly gross spend per client declined from approximately $140,000 to approximately $30,000. Citi expects e-commerce to contribute more than half of overall revenue growth by 2028 and maintained its Buy rating and $710 price target; Gist may reduce app-store policy risk, but spending per client will determine the quality of the second growth curve.
“E-commerce client growth is strong, but spending per client is low; its growth contribution is expected to approach that of mobile gaming by 2028.”
SpaceX Consensus expectations for second-quarter revenue are $6.9 billion, with 12.1 million consumer Starlink users and 1.4GW of compute capacity at quarter-end; Morgan Stanley expects full-year capital expenditure of $48 billion and year-end compute capacity of 2GW. In the near term, the company must also absorb the first lockup expiration of more than 930 million shares on August 6; longer term, the focus is on Starlink adoption, Starship reusability, and the economics of the AI business.
“The market generally believes that the quarterly figures themselves may not necessarily change the investment narrative; management’s messaging, tone, and explanation of the businesses may matter more.”
ISBNdb Training-data procurement is becoming a platform business. The company has reportedly helped clients procure anywhere from thousands of books to as many as 1 million books, while concealing buyers through nondisclosure agreements; one bookseller’s weekly sales increased from fewer than 20 books to several hundred. Order growth is being driven by demand for human-authored text from before 2022, but mistaken destruction of scarce editions and the transfer of public knowledge into private models are long-term risks.
Software/SaaS
Wolters Kluwer The moat in tax software comes from records, workflows, and chains of accountability rather than content alone. Approximately 92% of the Tax business’s revenue is recurring, while CCH Axcess spans 16 modules, more than 10,000 U.S. firms, and 1.4 million users, with a renewal rate above 95%. JPMorgan maintained its Overweight rating and €87 price target; AI monetization will come primarily through price increases, module attachment, and processing volume. The next question is whether adoption of AI functionality can translate into revenue per customer.
“The core view is that AI is more likely to strengthen CCH’s competitive barriers than weaken them. Although tax rules themselves can be encoded by models, customers also own the underlying data.”
SAP Q2 current cloud backlog grew approximately 26% year over year at constant currency, accelerating by approximately 1 percentage point from Q1, but Joule testing environments, sovereign cloud, product marketing, acquisition dilution, and token spending collectively compressed margins. JPMorgan maintained its Neutral rating and cut its price target to €175; if AI revenue and customer feedback do not improve, elevated investment may shift from a one-off burden to structural pressure.
“If AI revenue has yet to materialize after sustained investment and customer feedback also shows no clear improvement, this may mean that SAP must maintain higher spending over the long term to respond to competition.”
Enterprise software workflows AI is easier to monetize within systems that incorporate permissions, citations, audit trails, and human approval. Tax software can automatically classify documents, extract data, populate forms, and route tasks; the value comes from increasing throughput and share of customer wallet, rather than selling a standalone AI license. The key metrics are module adoption, renewal rates, and revenue per customer. Unless adoption rises, the AI label will be difficult to convert into actual pricing.
“Replacement barriers across modules arise primarily from workflow customization, historical files, permission systems, review records, staff training, and upstream and downstream data connections, rather than differences in any single feature.”
IT services and global capability centers The six leading Indian IT-services companies reduced net headcount by approximately 64,000 after the hiring peak, while the rest of the technology-services sector added approximately 700,000 employees over 3 years to reach approximately 4.4 million. If AI initially leads to lower pricing, revenue and margins may come under pressure; if demand for data engineering, cybersecurity, and implementation broadens, expanded service scope may offset efficiency gains in individual tasks.
“However, efficiency gains from AI may be passed on to customers through lower pricing, limiting the benefit to revenue and margins; if prices fall before business scope expands, IT companies’ revenue growth may still come under pressure.”
AI software pricing models Frontier labs are promoting pricing based on token usage and total-cost-of-ownership markups, while enterprises are more willing to lock predictable workloads into contracts or provisioned throughput. The winner may not be the model with the highest unit price, but the platform that can reliably route tasks across performance, cost, latency, and governance. The next metric is how quickly enterprise budgets shift from seat-based pricing to usage-based pricing.
“But this raises a question—why would enterprises do this, especially when they can anticipate their token/compute requirements? Contracts allow them to reserve capacity at a predictable cost.”
Thomson Reuters is Wolters Kluwer’s most direct full-platform competitor in the U.S. professional-services market. Competition extends beyond tax content to documents, research, review, and end-to-end workflows; if AI-native point solutions capture high-value stages, both incumbent platforms will need product updates to defend module attachment rates and pricing. The real issue to track is whether customers are willing to migrate critical audit trails out of their core systems.
Consumer Electronics/Smart Vehicles
MicroPort MedBot Revenue grew approximately 200% to 230% year over year in 1H26, overseas revenue increased by more than 450%, and gross margin improved by more than 15 percentage points. Toumai has secured approximately 300 cumulative commercial orders, including more than 240 overseas; approximately 1,500 overseas procedures have been performed across 52 countries and regions. Management maintained its full-year guidance of 200 installations, RMB1.1 billion in revenue, and breakeven. Order collections and overseas utilization are the next points of validation.
“MicroPort MedBot has reached a financial inflection point: it achieved profitability in 1H26, with revenue growing approximately 200% to 230% year over year and overseas revenue increasing by more than 450% year over year.”
“Toumai Pro is priced at approximately 70% of da Vinci, while Toumai SE is approximately 30% cheaper than Pro.”
Edge Medical Robotics Global cumulative installations or shipments are expected to reach approximately 160 units in 1H26, including approximately 100 overseas. As of May, cumulative procedure volume exceeded 20,000, with approximately 210 procedures per unit annually in China and approximately 100 overseas. Following approval in Australia, the company continues to pursue approvals in Japan and South Korea and to build a direct-sales team in Europe; the key is converting overseas installations into sustained procedure volume.
Intuitive Surgical As of 2Q26, the company had 11,710 systems installed globally and approximately 20 million cumulative procedures. It installed 395 systems in the latest quarter, while procedure volume grew 20% year over year. Its large installed base provides advantages in physician training, clinical data, and product iteration; Chinese companies are entering overseas markets through lower prices and remote surgery, but lower upfront equipment prices alone are insufficient to disrupt long-term barriers to use.
“Product capabilities, clinical value, remote surgery, physician-training networks, platform completeness, and continuous iteration matter more than the equipment’s initial selling price.”
AK Medical As of June 2026, the company had won tenders for 11 robotic systems, comprising 6 in China and 5 overseas; management targets 20 units and RMB40 million in robotics revenue for the full year. The company has already deployed more than 80 trial systems during the year and plans to increase procedure volume from 4,000 in 2025 to 10,000–14,000 in 2026. The key is converting trials into commercial orders.
Cornerstone Robotics The Sentire laparoscopic surgical system received EU CE certification and Singapore approval in May 2026, and its first installation in Singapore has been completed. The registrations cover minimally invasive procedures in general surgery, gynecology, thoracic surgery, and urology, among other specialties, but commercialization must still overcome barriers in physician training, distributor capabilities, and after-sales service. Certification does not equate to stable revenue. Approval is only the entry point; recurring procedure volume determines consumables returns.
Shurui Its single-port surgical robot completed its first commercial deployment in Munich, Germany, in May 2026 and performed its first pediatric robot-assisted laparoscopic procedure in Barcelona in June. Product differentiation provides an entry point into overseas markets, but with approximately 30 domestic tendered units and an average procurement price of RMB13.3 million, repeat overseas orders matter more than a single deployment. It also remains to be seen whether niche indications such as pediatrics can broaden hospital coverage.
TINAVI announced on July 16 that it planned to acquire Shanghai MicroPort Orthopedics through the issuance of new shares. The transaction is intended to combine surgical-robotics capabilities with orthopedic implants and an overseas commercial network, expanding product synergies and global coverage; the next points to watch are transaction completion, channel integration, hospital utilization, and consumables revenue. M&A; synergies ultimately need to translate into utilization per system and associated implant sales.
Harbin Sagebot As of August 2025, cumulative orders exceeded 60 units, including approximately 40 overseas, while global registrations covered more than 50 markets. Order volume has reached a certain scale, but the reported figure represents cumulative orders rather than systems fully installed; commercialization quality must be assessed through actual deliveries, procedure volume, cash collection, and overseas distribution capabilities. If channel expansion outpaces clinical use, collection cycles may lengthen.
Weijing Medical As of year-end 2025, cumulative orders remained in the low single digits, clearly placing the company at an early stage of commercialization; it had completed more than 150 registration-related procedures. The product has entered industry comparisons, but orders, installations, and commercial procedure volume have yet to reach scale. The next questions are whether the scope of registration, feedback from the first hospitals, and channel investment can generate repeat purchases.
Medtronic Hugo RAS had more than 70 systems installed globally as of year-end 2024 and approximately 10,000 cumulative procedures as of year-end 2025. The company has mature medical-device channels, but its surgical-robotics business still needs to catch up with Intuitive Surgical in clinical data and physician networks; following U.S. approval at year-end 2025, the pace of commercialization and utilization per system are the core metrics.
CMR Surgical Versius had 186 systems installed as of year-end 2024 and more than 45,000 cumulative procedures as of March 2026, while also completing more than 30 global registration-related procedures. Its scale lies between leading platforms and new entrants. Existing data indicate that the equipment can achieve clinical utilization, but further growth still depends on hospital expansion, physician training, consumables revenue, and approvals across markets.
SS Innovations As of April 2026, the company had installed more than 200 systems and completed more than 10,500 cumulative procedures; European CE certification is expected in 4Q26. Its scale remains materially below Intuitive Surgical’s. If it competes for equipment orders solely on price, margins and after-sales investment may come under pressure; the hard metric is whether it can continue increasing procedure volume per system.
Medicaroid Hinotori had approximately 100 systems installed as of February 2026 and approximately 15,000 cumulative procedures as of year-end 2025, and received European CE certification in July 2026. Industry comparisons have shifted from “whether a company has a product” to approvals, installations, and procedure volume; new platforms still need to demonstrate that physician training and clinical use can scale in tandem.
“Domestic demand is constrained by hospital capital expenditure and installation quotas, while overseas markets are currently the primary growth driver. The report maintained Buy ratings on both companies.”
Samsung Electro-Mechanics has notified customers that prices for all MLCCs will increase by 30% effective August 1, 2026. The company said growth in electronics demand has exceeded the supply chain’s capacity to absorb pressure; the increase will affect costs for smartphones, servers, AI hardware, and automotive electronics. The next question is whether delivery lead times also lengthen. If downstream customers reject the increase, its impact may be partially diluted through product mix.
Taiyo Yuden announced that it would adjust prices effective September 1, citing persistently high raw-material costs; previous cost reductions had yet to fully absorb the pressure. Successive price increases by two major suppliers indicate tightening supply and demand for passive components, but whether this develops into an industry-wide increase still depends on demand and the pace of capacity rebalancing. The strength of server and automotive-grade orders is the next point of validation.
“The coming months may bring higher component costs and potentially longer delivery lead times.”