01
Samsung raises advanced foundry prices, with some new orders increasing by up to 15%
FoundryPricingAI Chips
Samsung Electronics has reportedly raised prices for selected new advanced foundry orders. Increases across 4nm, 5nm and 8nm processes vary by customer region and demand conditions, with some customers seeing increases of roughly 10%–15%.
The move reflects tightening advanced-node capacity driven by AI-chip demand. With TSMC’s advanced capacity remaining constrained, some customers are directing more orders toward alternative foundry sources including Samsung and Intel.
R&A View: The foundry market is moving from a capacity-expansion story into a phase where pricing and allocation are tightening at the same time. Programs involving HPC, networking, AI ASICs, server CPUs/GPUs and advanced packaging should confirm wafer allocation, quote validity, process nodes, packaging/test capacity and second-source feasibility earlier.
02
Marvell and Google expand custom AI-chip cooperation, increasing the potential value of the partnership
Custom ASICGoogleMarvell
Marvell and Google have expanded their custom AI-chip cooperation, with Google receiving warrants to acquire up to roughly USD 12.2 billion of Marvell shares. If agreed milestones are reached, Marvell has indicated the relationship could generate substantial revenue over the coming years.
The agreement shows hyperscalers continuing to use custom ASICs, memory and networking chips to reduce reliance on a single general-purpose GPU path while expanding their in-house silicon ecosystems.
R&A View: Custom AI silicon is becoming a second major track for AI infrastructure. Supply-chain teams should watch ASIC wafers, advanced packaging, HBM, SerDes, retimers, Ethernet, DSPs, switch silicon, PCBs, connectors and high-speed cabling.
03
Analog Devices beats expectations as AI drives power-management and analog demand
AnalogPower ManagementData Center
Analog Devices issued a quarterly outlook above market expectations as AI-related computing investment continued to support demand for power-management, signal-chain and data-processing components.
ADI’s portfolio spans data centers, industrial automation and automotive markets, so its performance is another sign that the AI cycle is spreading into traditional analog and power semiconductors.
R&A View: The value content in an AI server is not limited to GPUs and HBM. PMICs, DC/DC converters, amplifiers, ADC/DACs, isolators, timing devices, sensors and interface ICs may see stronger structural demand. Procurement teams should focus on specific high-specification part numbers rather than assuming the entire analog market is equally tight.
04
Cerebras launches CS-4 AI inference system as alternative accelerator competition intensifies
AI InferenceAcceleratorServer
Cerebras launched its next-generation CS-4 AI system, combining its WSE-3 Turbo chip with a Nexus server architecture to target higher-performance AI chatbot and large-model inference workloads.
The system continues to emphasize lower internal communication latency, fewer system components and faster deployment, reflecting a broader shift in AI competition from training performance toward inference throughput, efficiency and time-to-deploy.
R&A View: As the inference market expands, supply chains will support more non-Nvidia platforms. Each platform creates different combinations of HBM/DRAM, NICs, switch chips, optical modules, power, cooling, motherboards and connectors, while also adding software-maturity and customer-qualification risk.
05
AI-chip startup Etched reaches a USD 21 billion valuation as inference silicon attracts more capital
AI StartupInference ChipsFunding
AI-chip startup Etched reached a valuation of roughly USD 21 billion after a new USD 700 million funding round, a sharp increase in less than a month. The company says it has already secured more than USD 1 billion in customer contracts.
Etched is focused on specialized inference chips designed to lower token cost and improve efficiency, aligning with the market’s shift from training toward large-scale inference deployment.
R&A View: Capital is moving quickly into specialized inference silicon, but commercialization still depends on manufacturing scale, software ecosystem, customer concentration and foundry partners. For procurement teams, the more important factors are production BOMs, foundry nodes, packaging method, memory configuration and network architecture—not valuation alone.
06
U.S. data-center boom expands into power generation, cooling, cabling and industrial supply chains
Data CenterPowerCooling
AI data-center investment is increasingly pulling demand into traditional industrial manufacturing. Generators, cooling systems, steel, bearings, cabling, prefabricated walls and construction equipment are all benefiting directly from data-center expansion.
Some suppliers are expanding U.S. capacity and signing longer-term customer agreements to address order backlogs and continued AI-infrastructure growth.
R&A View: The AI supply chain has expanded from the semiconductor BOM to rack-to-grid infrastructure. Power, UPS systems, generators, cooling, pumps and valves, busbars, cabling, connectors, sensors and structural components can all become delivery bottlenecks. Server programs should evaluate system-level lead times alongside chip lead times.
07
Nvidia H200 chips reportedly begin entering China in small volumes, raising cross-border supply and compliance complexity
H200ChinaCompliance
According to a Financial Times report cited by Reuters, small volumes of Nvidia H200 AI chips have recently entered mainland China, with ByteDance and Tencent each reportedly receiving about 10,000 processors. Reuters said it had not independently verified the report.
At the same time, China continues to promote domestic AI-chip ecosystems and impose additional policy requirements around deployment locations and procurement routes for advanced computing hardware.
R&A View: Availability of advanced AI chips increasingly depends on licensing, end users, deployment location and cross-border routing—not just inventory. Programs involving H200, Blackwell or other controlled compute products should move export-control, end-use, end-user, logistics-path and customer-compliance documentation to the front of the procurement process.
08
SMIC raises selected wafer prices as AI demand remains strong and utilization stays high
SMICWafer PricingChina Foundry
SMIC said AI-related demand continues to support orders and that it has raised prices for some of its most constrained capacity. Second-quarter revenue exceeded USD 3 billion for the first time, while wafer shipments increased and utilization remained high.
SMIC is also continuing to expand 12-inch capacity and accelerate new-line ramp-up to address demand from domestic customers across logic, analog, power and other semiconductor categories.
R&A View: Samsung and SMIC raising prices during the same period is an important common signal: AI demand is pushing the foundry market from recovery toward selective tightness. Both mature and advanced nodes should be assessed by specific wafer process, mask set, packaging and customer qualification rather than through a simple mature-node versus advanced-node split.
