North America — United StatesWolfspeed expands its 200 mm silicon-carbide materials portfolio with commercially available premium n-type power substrates. 200 mm n-type SiC substrates; commercial availability; incremental capacity and capex undisclosed.
North America — United StatesU.S. International Trade Commission opens patent investigation covering DRAM devices from Micron and products from HPE, Lenovo and Supermicro. 337-TA-1523; target completion date to be set within 45 days; no merits determination.
North America — United StatesCoherent launches PhotonLink integrated-optics platform as it expands 6-inch InP and specialty-fiber manufacturing capacity. 6-inch InP expansion; >10 CPO, >10 NPO and >5 chip-to-chip engagements; Q4 2026 revenue-ramp target.
Global — Taiwan / United StatesSynopsys secures TSMC A14 EDA certification and expands 2 nm, HBM4, CoWoS and co-packaged-optics enablement. TSMC A14 certification; 2 nm IP; PCIe 7.0, HBM4, CoWoS and COUPE enablement.
Global — Taiwan / United StatesCadence expands TSMC partnership across A14, N2P/N3P and wafer-scale 3D-IC systems containing hundreds of chiplets. A14 tool certification; N3P/N2P silicon-proven IP; wafer-scale 3D-IC support for hundreds of chiplets.
Global — Belgium / TaiwanIC-Link by imec expands TSMC partnership to provide managed-cloud access to advanced-node chip design and manufacturing support. TSMC OIP Cloud Alliance membership; mixed-partner EDA; 2.5D/3D packaging and custom wafer services.
Global — Germany / TaiwanSiemens and TSMC extend semiconductor-design collaboration into AI-driven DRC fixing and advanced 3D-IC verification. Calibre, Aprisa and Solido AI agent; SoIC verification; CoWoS-L certification work ongoing.
Europe — France / GermanySiPearl delivers first Rhea1 processor samples to Bull for integration into the JUPITER exascale supercomputer. First Rhea1 CPU samples; BullSequana XH3000 integration; JUPITER system; not mass production.
Europe — NetherlandsMorphotonics closes more than €40 million Series B to scale nanoimprint manufacturing technology for optical components and CPO. Expanded Series B >€40 million (about US$46 million); round originally opened in 2024.
Global — Netherlands / United StatesSMART Photonics and GlobalFoundries combine silicon and indium-phosphide manufacturing in open-access photonics foundry service. Open-access silicon/InP photonics service; lasers and modulators; qualification milestones to follow.
Europe — Belgiumimec demonstrates 100 GHz Ge/Si photodetector and net 400 Gbps optical link on a 300 mm silicon-photonics platform. 100 GHz; 5 V; 1.8 A/W; net 400 Gbps; claimed 3 dB sensitivity improvement; 300 mm platform.
Southeast Asia — MalaysiaAT&S and Marvell sign long-term agreement supporting additional advanced IC-substrate capacity at Kulim. Long-term Marvell agreement; Kulim advanced IC substrates; incremental volume and investment undisclosed.
Southeast Asia — SingaporeNexstrom raises US$12 million to scale equipment for growing 2D semiconductor materials on commercial 300 mm wafers. US$12 million seed; US$15 million total funding; 300 mm platform; commercial readiness targeted for 2030–2035.
Southeast Asia — SingaporeSemiconductor companies commit more than S$500 million to Singapore R&D projects over five years. Industry R&D >S$500 million (about US$390 million), 2026–2030; separate S$800 million and S$500 million government programmes.
China — ChinaAlibaba unveils Zhenwu V900 AI accelerator, targeting Q1 2027 mass production and clusters of up to 500,000 cards. V900: claimed 3× M890 performance; 216 GB; 1,200 GB/s; FP8/FP4; Q1 2027 production target.
Japan — JapanEdgeCortix launches RAIDEN scalable AI-chiplet architecture for physical AI with up to four dies and 256 GB memory. RAIDEN X4: four dies; 3.36 PFLOPS FP4; 256 GB; 1.54 TB/s die-to-die; 6.4 Tb/s scale-out.
Global — South Korea / United StatesNavitas takes US$5 million strategic stake in Magnachip to support qualification of high-voltage SiC technology in South Korea. US$5 million; 1,461,988 shares at US$3.42; 1,200 V, 2,300 V, 3,300 V and higher SiC technology.
Memory, Materials & Packaging
Wolfspeed expands its 200 mm silicon-carbide materials portfolio with commercially available premium n-type power substrates.
27 September 2026 | North America — United States
Wolfspeed announced commercial availability of a new 200 mm n-type SiC substrate, expanding its 200 mm materials portfolio. The company says the material improves defect density, micropipe count and wafer-shape stability, characteristics intended to improve downstream device yield as manufacturers move toward larger dies and higher-voltage SiC devices. This is a commercial-material availability milestone, not a new fab or capacity announcement; Wolfspeed did not disclose incremental wafer-production capacity or capital expenditure. The development matters because substrate consistency and defect control remain important manufacturing constraints as SiC production migrates from 150 mm toward 200 mm wafers.
The next commercial test is whether better substrate quality translates into higher saleable device yield across customer epitaxy and fabrication processes. Watch lot-to-lot consistency, qualification time and the premium customers will pay. If the premium exceeds the value of recovered dies, technical improvement will not automatically accelerate adoption.
GreyRadius Insight
SiC sourcing should be evaluated on cost per qualified die, including scrap, rework and delivery variability. A consistent substrate can support better economics even at a higher wafer price. Device makers should run matched-lot trials and negotiate quality specifications before committing volume; the opportunity for suppliers lies in proving predictable output rather than competing on wafer diameter alone.
Policy & Supply Chain
U.S. International Trade Commission opens patent investigation covering DRAM devices from Micron and products from HPE, Lenovo and Supermicro.
23 September 2026 | North America — United States
The USITC instituted investigation 337-TA-1523 following an August complaint from Netlist alleging patent infringement involving certain DRAM devices and products containing them. Respondents include Micron Technology, Micron Semiconductor Products, Hewlett Packard Enterprise, Lenovo Group, Lenovo US and Super Micro Computer. Netlist requested a limited exclusion order and cease-and-desist orders. The commission explicitly states that opening the investigation represents no determination on the merits; a target completion date must be established within 45 days. Any eventual exclusion remedy could affect U.S. memory and server sourcing, but there is no such restriction at this stage.
The relevant trigger is an enforceable remedy, not the opening of the investigation. Track the procedural timetable, scope of covered products and any settlements. Memory exposure can propagate into server availability, so procurement teams should identify which systems depend on the disputed components before a legal decision forces a rushed response.
GreyRadius Insight
A component-level patent dispute can become a system-level supply risk. Buyers should map affected memory to server configurations, review supplier indemnities and assess how long alternative components take to qualify. This preparation preserves options without assuming an import restriction already exists. The decision is whether contingency costs today are justified by the operational exposure of each workload.
Memory, Materials & Packaging
Coherent launches PhotonLink integrated-optics platform as it expands 6-inch InP and specialty-fiber manufacturing capacity.
21 September 2026 | North America — United States
Coherent launched PhotonLink, integrating InP lasers, VCSEL arrays, silicon photonics, micro-optics, specialty fiber and detectors for co-packaged optics (CPO), near-packaged optics (NPO) and chip-to-chip AI connectivity. The company disclosed more than 10 CPO customer engagements, more than 10 NPO engagements and more than five chip-to-chip engagements, with anchor customers and long-term agreements already secured for CPO and NPO. PhotonLink revenue is expected to start ramping in Q4 2026. Importantly for manufacturing, Coherent is expanding its 6-inch InP and specialty-fiber capacity; its existing manufacturing base has shipped more than 300 million InP lasers and one billion-plus InP/GaAs photodetectors.
PhotonLink’s expected revenue ramp will test whether engagements have progressed into qualified, repeatable orders. Watch customer acceptance, contracted volumes and the utilisation of expanded InP capacity. CPO, NPO and chip-to-chip applications have different adoption paths; combining their engagement counts can obscure where near-term revenue is actually emerging.
GreyRadius Insight
AI optics shifts value toward suppliers that can coordinate light sources, detectors and integration into a dependable system. For entrants, the strongest route is a qualified component or process tied to an anchor customer. Capacity plans should follow the application with demonstrated demand, while accounting for packaging yield and test complexity that can erode the apparent margin advantage of advanced optics.
Foundry & Advanced Nodes
Synopsys secures TSMC A14 EDA certification and expands 2 nm, HBM4, CoWoS and co-packaged-optics enablement.
23 September 2026 | Global — Taiwan / United States
Synopsys and TSMC expanded their collaboration across A14, 2 nm-class IP, advanced packaging and silicon photonics. Synopsys announced certified EDA flows for TSMC A14, expanded 2 nm IP and tape-out milestones including PCIe 7.0 and HBM4, and further design enablement for multi-die systems using TSMC CoWoS. The companies are also extending support for TSMC-COUPE co-packaged optics and agentic-AI engineering workflows. This is a design/manufacturing-readiness milestone rather than wafer production: it expands the tool and IP ecosystem customers need to tape out future AI/HPC silicon on TSMC's most advanced processes.
A14 certification expands the design pathway, but programme readiness still depends on IP availability, package integration and customer tape-outs. Watch whether HBM4, PCIe 7.0 and CoWoS milestones align on the same product schedule. A delay in any one interface can postpone a complete AI system even when the logic design is ready.
GreyRadius Insight
Leading-edge chip planning is becoming a coordination problem across multiple suppliers. Leadership should approve a node transition only after the memory, packaging and verification plan supports the target launch date and economics. Reserve engineering time for cross-interface validation and compare the expected system benefit with the additional development cost; smaller geometry alone is an incomplete investment case.
Foundry & Advanced Nodes
Cadence expands TSMC partnership across A14, N2P/N3P and wafer-scale 3D-IC systems containing hundreds of chiplets.
23 September 2026 | Global — Taiwan / United States
Cadence announced tool certification for TSMC A14, new silicon-proven IP for N3P and N2P, and a UALink implementation demonstrated on N3P. Its TSMC enablement now spans N3, N2, A16 and A14, alongside agentic-AI design flows. Particularly significant for advanced packaging, Cadence says its TSMC 3DFabric support extends to wafer-scale systems containing hundreds of chiplets. The announcement is therefore relevant both to leading-edge logic design readiness and the manufacturing ecosystem needed for increasingly large heterogeneous AI systems.
Wafer-scale integration increases exposure to chiplet defects, assembly losses and difficult-to-test interfaces. Watch demonstrated customer systems, repair or redundancy strategies and the maturity of package-wide sign-off. Tool certification helps teams start designs; it does not establish that systems containing hundreds of chiplets will achieve attractive production yield.
GreyRadius Insight
Chiplet strategies should optimise delivered system cost rather than the number of dies assembled. Reusing qualified blocks can reduce redesign, but each interface introduces verification, power and yield considerations. Before scaling an architecture, teams should model how a failed die or bonding step affects the cost of the entire assembly and define who owns recovery, testing and reliability across suppliers.
Foundry & Advanced Nodes
IC-Link by imec expands TSMC partnership to provide managed-cloud access to advanced-node chip design and manufacturing support.
23 September 2026 | Global — Belgium / Taiwan
IC-Link by imec joined TSMC's Open Innovation Platform Cloud Alliance, becoming what imec describes as the first managed-cloud service supporting mixed-partner EDA flows for silicon design. IC-Link combines secure cloud design infrastructure with TSMC advanced-node access and manufacturing support. Its broader manufacturing services cover advanced CMOS ASICs, integrated photonics, 2.5D/3D packaging and custom wafer processing. The significance is improved advanced-node access for fabless companies and other designers that lack their own full design/manufacturing infrastructure.
Managed-cloud access can shorten infrastructure setup, but its value should be measured in completed tape-outs and manufacturing handoffs. Watch licence terms, access to supported process kits, secure data handling and portability between EDA partners. Lower entry friction is useful only if designers can reach a supported, manufacturable implementation.
GreyRadius Insight
For smaller fabless firms, shared infrastructure can shift spending from fixed setup toward product development. The decision should include licence consumption, engineering support, IP rights and the cost of leaving the service. Choose a provider that connects design access to manufacturing and packaging milestones; otherwise reduced upfront cost may be offset by coordination delays later.
AI Chips & Custom Silicon
Siemens and TSMC extend semiconductor-design collaboration into AI-driven DRC fixing and advanced 3D-IC verification.
24–25 September 2026 | Global — Germany / Taiwan
Siemens announced continued TSMC collaboration across AI-powered EDA, advanced processes, silicon photonics and TSMC 3DFabric. The companies enabled an AI agent using Calibre, Aprisa and Solido to identify and fix design-rule violations across digital and custom IC flows. Siemens also expanded Calibre 3DStack support for inter-chiplet verification, including TSMC SoIC, while certification work continues for CoWoS-L. This is an EDA/manufacturing-enablement milestone rather than new physical capacity, but it targets the verification bottleneck created by increasingly complex 3D chiplet systems. Editorial qualification: Siemens’ page carries inconsistent date metadata; contemporary reporting places the collaboration in this week.
Track whether automated design-rule fixes reduce closure time while preserving verified sign-off quality. For complex 3D systems, faster local fixes are insufficient if they create new interface errors or downstream rework. The source-date inconsistency remains relevant: weekly timing is supported by contemporary reporting, while the Siemens page carries an earlier date.
GreyRadius Insight
The commercial case for AI-assisted EDA should rest on fewer engineering iterations and successful first-pass silicon. Start with auditable workflows, retain review of automated changes and compare total verification effort with the existing process. If productivity gains are repeatable, teams can redeploy scarce engineers to architecture and integration; buying an AI-enabled tool does not itself demonstrate that benefit.
AI Chips & Custom Silicon
SiPearl delivers first Rhea1 processor samples to Bull for integration into the JUPITER exascale supercomputer.
22 September 2026 | Europe — France / Germany
European fabless CPU developer SiPearl delivered the first Rhea1 samples to Bull, which began integrating the processor into its BullSequana XH3000 platform for the JUPITER supercomputer at Germany's Jülich Supercomputing Centre. The milestone is sample delivery and system integration, not mass production. Rhea1 is intended to become the first European-designed CPU deployed in a European-built supercomputing platform and forms the CPU complement to JUPITER's GPU partition. The development is commercially significant because it moves Europe's sovereign HPC processor effort from chip development into customer-system integration.
Sample delivery moves Rhea1 into customer integration, where application benchmarks, software readiness and stable operation become decisive. Watch published results and progress toward qualified system deployment. Early functionality should not be confused with broad availability, and sovereign procurement objectives still need evidence that workloads run reliably and economically.
GreyRadius Insight
Europe’s processor strategy now faces the adoption test: can local silicon sustain a supported computing platform that customers choose repeatedly? Buyers should assess software migration, operational support and workload performance alongside supply sovereignty. For ecosystem suppliers, the practical opening is in integration and application enablement around a real customer platform, rather than relying solely on the strategic appeal of European origin.
Investment & Infrastructure
Morphotonics closes more than €40 million Series B to scale nanoimprint manufacturing technology for optical components and CPO.
22 September 2026 | Europe — Netherlands
Dutch equipment developer Morphotonics closed an expanded Series B exceeding €40 million, equivalent to roughly US$46 million at contemporaneous exchange rates, to scale its large-area nanoimprint-lithography manufacturing technology. Investors include 3M Ventures, Innovation Industries, BOM, Invest-NL, EIC Fund, Ernij Next and the European Investment Bank. The company intends to expand manufacturing technology for AR waveguides while developing applications in co-packaged optics for AI infrastructure. Independent reporting notes that this is an expanded financing round originally opened in 2024, so the €40 million-plus total should not be interpreted as entirely new capital raised during this week.
The expanded financing must translate into qualified equipment and customer production use. Watch application-specific throughput, replication consistency and adoption in optical manufacturing. The announced round total includes capital associated with the earlier round; it should not be used as a measure of entirely new weekly investment or immediate capacity.
GreyRadius Insight
Nanoimprint’s AI opportunity will depend on whether it lowers the cost of producing optical components at the quality customers require. Manufacturers should compare tooling, consumables, uptime and usable output with competing processes before committing. Investors should link the next funding or expansion decision to validated customer economics, because a promising CPO application is not yet evidence of repeat equipment orders.
Foundry & Advanced Nodes
SMART Photonics and GlobalFoundries combine silicon and indium-phosphide manufacturing in open-access photonics foundry service.
21 September 2026 | Global — Netherlands / United States
SMART Photonics and GlobalFoundries announced an open-access foundry offering combining GF's production silicon platform with SMART Photonics' high-volume InP photonic integrated circuits, including lasers and modulators. The service targets AI data centers, pluggable optical transceivers and other photonics applications requiring the benefits of both material systems. The manufacturing model is important because native InP light sources can be integrated with scalable silicon photonics rather than forcing customers to independently develop both process chains. The partners describe the offering as a commercial foundry service, although production availability and customer qualification should be tracked separately from the announcement.
The key execution risk sits at the boundary between silicon and InP manufacturing. Track supported design kits, integration methods, qualification and responsibility for test failures. Open access broadens customer participation, but reliable production depends on a clear handoff between the two process chains and sufficient visibility into combined yield.
GreyRadius Insight
A combined foundry offering can reduce the burden of assembling a photonics supply chain independently. Buyers should use that convenience to demand clearer accountability for interfaces, testing and delivery rather than assume integration risk disappears. Commercial advantage will come from shortening time to a qualified optical product while keeping yield losses and supplier dependency manageable.
Memory, Materials & Packaging
imec demonstrates 100 GHz Ge/Si photodetector and net 400 Gbps optical link on a 300 mm silicon-photonics platform.
21 September 2026 | Europe — Belgium
imec presented a Ge/Si avalanche photodiode fabricated on its 300 mm silicon-photonics platform that combines 100 GHz bandwidth, 5 V operation and 1.8 A/W responsivity across the O- and C-bands. The device demonstrated a net 400 Gbps optical link and a claimed 3 dB receiver-sensitivity improvement. imec says future work will address reliability across wider temperature/input-power conditions and integration with high-speed receiver electronics. This remains an R&D demonstration, not a qualified commercial component, but its 300 mm implementation is relevant to future manufacturability of AI optical interconnects.
The detector demonstration must progress through temperature reliability, optical-power tolerance and receiver-electronics integration. Watch whether the sensitivity benefit survives packaging and real operating conditions. Performance on a 300 mm platform is encouraging, but commercial readiness requires evidence beyond the demonstrated optical link.
GreyRadius Insight
The strategic value of improved receiver sensitivity is the flexibility it can create in the link budget: lower laser power or greater tolerance for optical loss. System designers should evaluate which trade-off creates more value for their architecture. Development partnerships should therefore measure complete-link energy and reliability, not only detector bandwidth; that evidence will determine where the component earns adoption.
Memory, Materials & Packaging
AT&S and Marvell sign long-term agreement supporting additional advanced IC-substrate capacity at Kulim.
22 September 2026 | Southeast Asia — Malaysia
AT&S expanded its relationship with Marvell through a long-term agreement to increase advanced IC-substrate capacity for next-generation AI and cloud infrastructure. AT&S identified Marvell as the additional customer behind its previously announced expansion of the Kulim, Malaysia manufacturing site. The project addresses increasing substrate complexity as AI devices grow larger and use more heterogeneous integration. AT&S did not disclose incremental substrate capacity or investment attributable specifically to Marvell in this announcement, so neither should be inferred; the important signal is that long-term customer commitments are supporting physical substrate-capacity expansion.
Marvell’s long-term agreement strengthens the demand foundation for Kulim expansion. Watch qualification, delivery schedules and capacity utilisation rather than infer an undisclosed investment amount. Advanced substrates must meet increasingly demanding electrical and physical requirements, so installation progress alone is not a reliable indicator of saleable capacity.
GreyRadius Insight
Customer-backed expansion can reduce the risk of building capacity ahead of demand, but it may also increase exposure to a small number of programmes. Suppliers should align expansion with qualification gates and contractual volume visibility, then assess customer concentration. For new entrants, the more credible opportunity is a specific qualified process or input supporting anchor demand, rather than generic exposure to AI packaging growth.
Investment & Infrastructure
Nexstrom raises US$12 million to scale equipment for growing 2D semiconductor materials on commercial 300 mm wafers.
22 September 2026 | Southeast Asia — Singapore
Singapore startup Nexstrom raised a US$12 million seed round, led by Xora Innovation with Foothill Ventures and SEEDS participating, taking total funding to US$15 million. Nexstrom is developing equipment to grow transition-metal dichalcogenide semiconductor materials on 300 mm/12-inch wafers and has installed a 12-inch system; it says industry partners are already testing samples. The company told TechCrunch it had systematically scaled from 2-inch to 6-inch wafers, expected to complete 8-inch work by end-October 2026, and was targeting commercial-production readiness between 2030 and 2035. This remains technology development/customer qualification—not mass production—but addresses a central challenge in moving 2D transistor materials from laboratory samples into conventional fabs.
Scaling wafer diameter must preserve material uniformity, defect control and compatibility with established fab processes. Watch the intermediate wafer milestones and independent customer validation; an installed 12-inch tool is not proof of repeatable 300 mm output. The 2030–2035 commercial horizon makes milestone quality more informative than near-term revenue expectations.
GreyRadius Insight
For emerging semiconductor materials, the critical investment question is whether the process can enter an existing manufacturing flow at acceptable yield and cost. Stage capital and partnerships around measurable process compatibility, sample qualification and repeatability. This reduces the risk of funding a strong laboratory result that cannot survive commercial wafer-scale manufacturing constraints.
Investment & Infrastructure
Semiconductor companies commit more than S$500 million to Singapore R&D projects over five years.
24 September 2026 | Southeast Asia — Singapore
Semiconductor industry participants will commit more than S$500 million (approximately US$390 million as reported contemporaneously) to Singapore R&D projects over 2026–2030, according to A*STAR information reported during the week. That compares with more than S$400 million spent between 2020 and 2025, indicating a higher nominal commitment, although the two more-than figures do not establish an exact percentage increase. Singapore is separately implementing its S$800 million (approximately US$624 million, using the supplied report’s currency ratio) RIE2030 semiconductor flagship and a S$500 million (approximately US$390 million) National Semiconductor Translation and Innovation Centre, which the government says will start operations in 2027; these government programs should not be added to the industry's S$500 million commitment as if they were one investment pool. Target technologies include advanced packaging, silicon photonics, RF GaN and next-generation power semiconductors. Date qualification: the 24 September CNA broadcast places the event in this week; the detailed article followed on 28 September.
Track which projects receive industry co-funding, which produce transferable IP and which reach customer qualification. Keep the industry commitment separate from government programmes and compare like-for-like periods before interpreting growth. The measure of success is the conversion of research into technologies that manufacturers can adopt, not the aggregate funding announced.
GreyRadius Insight
Singapore’s opportunity lies in connecting research, shared infrastructure and industrial customers. Entrants should identify the programme that resolves their specific qualification bottleneck, confirm access and IP terms, and secure an application partner early. Public infrastructure can reduce the cost of experimentation, but commercial value depends on who will adopt the result and how it enters production.
AI Chips & Custom Silicon
Alibaba unveils Zhenwu V900 AI accelerator, targeting Q1 2027 mass production and clusters of up to 500,000 cards.
22 September 2026 | China — China
Alibaba's T-Head semiconductor unit unveiled the Zhenwu V900 AI accelerator for training and inference. Alibaba says it delivers 3× the performance of the preceding M890; reported specifications include 216 GB memory, 1,200 GB/s inter-chip bandwidth and native FP8/FP4 support. Alibaba says more than 1,000 V900 devices can operate as a supernode and its broader architecture can scale to clusters containing as many as 500,000 accelerator cards. Mass production and commercial release are planned for Q1 2027, meaning the V900 is not yet a volume-production product; the foundry, process node, absolute FLOPS and power consumption remain undisclosed. Alibaba also set a target of more than 20 GW of global data-center capacity by 2032.
The Q1 2027 production target is the next meaningful gate for V900. Watch delivered hardware, sustained workload performance, power requirements and software maturity. Claimed performance relative to M890 and very large cluster-scale targets do not establish competitiveness against alternatives or confirm that the stated architecture has been deployed.
GreyRadius Insight
Accelerator procurement should compare the cost of completing a representative workload, including power, software migration and utilisation. A domestic platform can improve sourcing options, but buyers need evidence that it performs inside their operating environment. Use bounded pilots with explicit acceptance criteria before committing infrastructure around a roadmap whose manufacturing and energy characteristics remain undisclosed.
AI Chips & Custom Silicon
EdgeCortix launches RAIDEN scalable AI-chiplet architecture for physical AI with up to four dies and 256 GB memory.
24 September 2026 | Japan — Japan
Japan-headquartered fabless semiconductor company EdgeCortix unveiled RAIDEN, a scalable AI chiplet platform aimed at physical-AI deployments. Its flagship RAIDEN X4 configuration scales to four dies and is specified at up to 3.36 PFLOPS FP4 compute, 256 GB memory, 1.54 TB/s die-to-die bandwidth and 6.4 Tb/s chip-to-chip scale-out connectivity. EdgeCortix says the platform has already secured adoption in a customer product, providing a commercial signal beyond a laboratory demonstration. The announcement qualifies because it is explicitly a chiplet architecture with manufacturing, packaging and commercial implications, rather than an ordinary consumer-product launch.
Watch delivery of customer samples and the transition from design wins to qualified production programmes. RAIDEN’s peak FP4 specification must be tested against sustained performance, memory constraints and power limits for the target application. A customer design decision indicates interest; deployment schedules and reliability validation determine when revenue becomes repeatable.
GreyRadius Insight
Physical-AI buyers should evaluate the complete perception-to-action workload, including latency and energy under sustained operation. A common chiplet and software platform may reduce redesign across product variants, but scaling to four dies must earn its integration cost. Select the smallest configuration that meets the operating requirement, and make higher performance tiers contingent on application-level evidence.
Investment & Infrastructure
Navitas takes US$5 million strategic stake in Magnachip to support qualification of high-voltage SiC technology in South Korea.
21 September 2026 | Global — South Korea / United States
Navitas Semiconductor agreed to make a US$5 million strategic equity investment in Magnachip, purchasing 1,461,988 shares at US$3.42 per share. The investment builds on their July agreement under which Magnachip gained access to Navitas' GeneSiC technology for 1,200 V, 2,300 V, 3,300 V and higher-voltage power-semiconductor applications. Magnachip plans to port, qualify and internalize the licensed SiC technology at its South Korean fabrication facility, targeting grid infrastructure, energy storage, industrial electrification, automotive and other high-power applications. The distinction is important: the investment and technology agreement are completed/announced milestones, while successful fab qualification and volume production remain future milestones.
The equity stake aligns the partners but leaves technology transfer, fab qualification and reliable volume output ahead. Track qualification by voltage class, application reliability and customer commitments. High-voltage SiC adoption depends on a successful process and validated devices; the investment amount cannot be treated as new operational capacity.
GreyRadius Insight
A manufacturing partnership creates value when licensed technology becomes a product customers can qualify and reorder. Sequence fab conversion around target applications, reliability requirements and credible demand, while clarifying technology-transfer responsibilities. For buyers, a second sourcing path becomes useful after production validation; for investors, the important return driver is qualified output and utilisation rather than the existence of the strategic stake.
Market Data & Intelligence
Metric
Latest Value
Publication Date
Region
Executive Implication
Wolfspeed commercially launches premium 200 mm SiC substrates
200 mm n-type SiC substrates; commercial availability; incremental capacity and capex undisclosed
27 September 2026
North America — United States
For power-device buyers, wafer price is only one part of cost.
USITC opens Section 337 investigation involving Micron DRAM and server products
337-TA-1523; target completion date to be set within 45 days; no merits determination
23 September 2026
North America — United States
Procurement teams should map the affected components and contractual responsibilities now.
Coherent launches PhotonLink and expands 6-inch InP manufacturing for AI optics
Cloud access can lower infrastructure barriers for fabless entrants, but manufacturing readiness still requires validated IP, packaging choices and a credible qualification budget.
Siemens and TSMC expand AI-powered EDA and 3DFabric enablement
Calibre, Aprisa and Solido AI agent; SoIC verification; CoWoS-L certification work ongoing
24–25 September 2026
Global — Germany / Taiwan
AI-driven verification becomes commercially useful when it improves reproducible sign-off quality and engineering cycle time.
SiPearl delivers first Rhea1 CPU samples for Europe's JUPITER system
First Rhea1 CPU samples; BullSequana XH3000 integration; JUPITER system; not mass production
22 September 2026
Europe — France / Germany
Sovereign processor strategies reach a meaningful commercial gate when customers integrate and validate the silicon.
Morphotonics closes €40 million-plus financing to scale nanoimprint equipment
Expanded Series B >€40 million (about US$46 million); round originally opened in 2024
22 September 2026
Europe — Netherlands
Nanoimprint equipment could capture value as optical components become more complex, but an AI-market expansion needs customer-specific evidence on yield, repeatability and throughput.
SMART Photonics and GlobalFoundries create open-access silicon/InP photonics foundry service
Open-access silicon/InP photonics service; lasers and modulators; qualification milestones to follow
21 September 2026
Global — Netherlands / United States
Combining native InP light sources with silicon photonics can simplify the supplier map for AI optics.
imec demonstrates 100 GHz Ge/Si APD on 300 mm silicon-photonics platform
100 GHz; 5 V; 1.8 A/W; net 400 Gbps; claimed 3 dB sensitivity improvement; 300 mm platform
21 September 2026
Europe — Belgium
A 300 mm platform offers a route toward manufacturability, but system adoption needs reliable receiver performance and packaging economics.
AT&S identifies Marvell as customer backing expanded Malaysian IC-substrate capacity
Long-term Marvell agreement; Kulim advanced IC substrates; incremental volume and investment undisclosed
22 September 2026
Southeast Asia — Malaysia
Long-term customer agreements can make substrate expansion more financeable and less speculative.
Nexstrom raises $12 million for 300 mm 2D-semiconductor manufacturing platform
US$12 million seed; US$15 million total funding; 300 mm platform; commercial readiness targeted for 2030–2035
22 September 2026
Southeast Asia — Singapore
The long 2030–2035 commercial horizon makes intermediate qualification evidence more useful than production claims.
Singapore semiconductor industry commits more than S$500 million to five-year R&D program
Industry R&D >S$500 million (about US$390 million), 2026–2030; separate S$800 million and S$500 million government programmes
24 September 2026
Southeast Asia — Singapore
Singapore is strengthening the bridge between research capability and industrial adoption.
Alibaba unveils Zhenwu V900 AI accelerator with Q1 2027 mass-production target
