Executive Highlights
Canada The Government of Canada committed CAD $195 million through the Strategic Response Fund toward Xanadu’s CAD $893 million Project OPTIMISM in Toronto.
United States / South Korea SK hynix broke ground on its West Lafayette, Indiana AI-memory advanced-packaging facility, with more than $4 billion planned investment and mass production of next-generation HBM targeted for H2 2029.
United States / Global Marvell reported record fiscal Q2 2027 revenue of $2.739 billion, up 37% year over year.
United States Diodes Incorporated completed its $250 million cash acquisition of ElevATE Semiconductor.
United States / Global NVIDIA reported fiscal Q2 2027 revenue of $96.221 billion, up 106% year over year, while Data Center revenue reached $89.0 billion, up 117% YoY.
United States / Global AWS and NVIDIA announced plans for AWS to deploy 2 million additional NVIDIA GPUs during 2027–2028 across Blackwell Ultra, Rubin and Rubin Ultra platforms.
United States / Global Semtech reported record fiscal Q2 2027 net sales of $341.9 million, increasing 17% sequentially and 33% year over year, with adjusted gross margin reaching 54.5%.
India / Germany / Global Infineon announced the acquisition of C2i Semiconductors, a Bengaluru-based specialist in intelligent digital power management for AI data centers.
Canada Commits CAD $195 Million to CAD $893 Million Advanced Photonics Manufacturing Project
August 28, 2026 | Canada
The Government of Canada committed CAD $195 million through the Strategic Response Fund toward Xanadu’s CAD $893 million Project OPTIMISM in Toronto. The project is designed to expand R&D and establish domestic manufacturing capabilities for advanced photonic and semiconductor components used in fault-tolerant quantum computing, including heterogeneous photonic-chip integration, photonic integrated-circuit packaging, wafer-level semiconductor testing and measurement, and quantum-module assembly. The project is expected to create 275 jobs. This is a funded manufacturing-capability project rather than existing commercial semiconductor capacity; no monthly wafer capacity, process node or wafer size was disclosed.
Source: Government of Canada
Strategic Watch
The real milestone is not the CAD $893 million project value, but whether Canada can convert photonics R&D into repeatable manufacturing capability. Watch qualification of wafer-level testing, heterogeneous integration and photonic packaging, followed by evidence of commercial-volume customers. Until those steps occur, this remains capability creation rather than meaningful semiconductor capacity.
GreyRadius Insight
Quantum manufacturing could create a new strategic layer alongside conventional logic and memory. Countries building packaging, testing and photonic-integration know-how now may gain disproportionate leverage if quantum hardware scales commercially. Canada’s opportunity is therefore less about competing directly in leading-edge silicon and more about owning specialized manufacturing capabilities that are difficult to replicate.
SK hynix Breaks Ground on More Than $4 Billion U.S. HBM Advanced-Packaging Facility
August 27, 2026 | United States / South Korea
SK hynix broke ground on its West Lafayette, Indiana AI-memory advanced-packaging facility, with more than $4 billion planned investment and mass production of next-generation HBM targeted for H2 2029. The project combines advanced HBM packaging with an R&D testbed developed with Purdue University, is expected to support about 7,000 direct and indirect jobs, and involves more than 100 ecosystem partners. U.S. CHIPS support includes $458 million in grants and up to $500 million in loans. This is advanced-packaging capacity rather than a new DRAM wafer fab; advanced wafers are initially expected to be manufactured in South Korea before packaging and testing in Indiana.
Source: SK hynix
Strategic Watch
Watch how much of the HBM value chain ultimately moves into Indiana. Initial wafers will still come from South Korea, making packaging, testing, R&D and supplier localization the first indicators of genuine U.S. capability. The bigger inflection point would be tighter integration between U.S.-based AI customers, HBM packaging and future memory development. With memory tightness expected to persist, execution timing will matter as much as announced investment.
GreyRadius Insight
Semiconductor localization is moving beyond the question of where wafers are fabricated. HBM demonstrates that where memory is stacked, packaged, tested and integrated with compute can be almost as strategically important as where the DRAM wafer originates. The Indiana investment therefore represents a broader shift toward regionalizing the complete AI semiconductor production chain—not simply building more fabs.
Marvell Data-Center Revenue Reaches $2.17 Billion as Custom AI Silicon Accelerates
August 27, 2026 | United States / Global
Marvell reported record fiscal Q2 2027 revenue of $2.739 billion, up 37% year over year. Data-center revenue reached $2.172 billion, increasing 46% YoY and 18% sequentially, and accounted for about 79% of total company revenue. Management highlighted exceptionally strong AI-related bookings and expected acceleration in custom semiconductor revenue in the second half of fiscal 2027. Reuters reported full-year revenue expectations of approximately $12 billion for FY2027 and $18 billion for FY2028, supported by custom AI silicon demand.
Source: Marvell
Strategic Watch
The next indicator is whether custom AI silicon moves from a hyperscaler-specific opportunity into a sustained second engine of leading-edge semiconductor demand. Watch design wins converting into production volumes, particularly where custom accelerators are bundled with networking and connectivity silicon.
GreyRadius Insight
The AI semiconductor market is beginning to split into two parallel architectures: general-purpose accelerator platforms and increasingly customized hyperscaler silicon. That matters upstream. Foundries, HBM suppliers and packaging providers may eventually need to support a much wider portfolio of high-volume AI designs rather than optimize capacity around a small number of dominant GPU architectures. Custom silicon therefore increases both the addressable market and supply-chain complexity.
Diodes Completes $250 Million Acquisition of Semiconductor-Test Specialist ElevATE
August 27, 2026 | United States
Diodes Incorporated completed its $250 million cash acquisition of ElevATE Semiconductor. ElevATE is a San Diego-based fabless semiconductor company specializing in highly integrated analog and mixed-signal ICs for automated test equipment. The acquisition gives Diodes greater exposure to semiconductor testing as AI accelerators, chiplets, heterogeneous integration and advanced packaging increase device complexity and test requirements. No manufacturing capacity, wafer size or process-node information was disclosed.
Source: Diodes Incorporated
Strategic Watch
Watch whether ElevATE enables Diodes to capture higher-value testing requirements created by chiplets, HBM and heterogeneous packages. Revenue growth alone will be less important than new ATE design wins and evidence that test intensity per semiconductor package is increasing.
GreyRadius Insight
Advanced packaging creates a less visible economic consequence: the cost of discovering a defect late becomes much higher. When multiple expensive dies, memory stacks and interfaces are combined into one package, yield protection becomes increasingly valuable. Testing therefore shifts from a downstream manufacturing step toward an economic control point protecting the value already embedded in the package.
NVIDIA Confirms Vera Rubin in Full Production as Data Center Revenue Reaches $89 Billion
August 26, 2026 | United States / Global
NVIDIA reported fiscal Q2 2027 revenue of $96.221 billion, up 106% year over year, while Data Center revenue reached $89.0 billion, up 117% YoY. NVIDIA said its Vera Rubin platform is in full production, with systems reported running at CoreWeave, Google Cloud, Microsoft Azure, Oracle Cloud Infrastructure and Nebius. The company guided fiscal Q3 revenue to $108 billion ±2%, while assuming no Data Center compute revenue from China. Q2 gross margin was 75.0%.
Source: NVIDIA
Strategic Watch
Rubin production should be monitored as a supply-chain execution test rather than simply another GPU launch. The critical indicators are HBM availability, advanced-package throughput, networking deployment and the ability of hyperscalers to bring sufficient power and cooling online. A shortage in any one layer can prevent finished silicon from becoming deployable AI compute.
GreyRadius Insight
The semiconductor industry is moving from chip scaling to system scaling. NVIDIA can accelerate architecture cycles, but actual AI capacity now depends on synchronized expansion across foundry, HBM, packaging, networking and data-center infrastructure. This changes competitive advantage: the strongest ecosystem may not be the one with the best individual component, but the one capable of scaling the entire system without creating a new bottleneck.
AWS Commits to Deploy 2 Million Additional NVIDIA GPUs During 2027–2028
August 26, 2026 | United States / Global
AWS and NVIDIA announced plans for AWS to deploy 2 million additional NVIDIA GPUs during 2027–2028 across Blackwell Ultra, Rubin and Rubin Ultra platforms. The commitment follows an earlier AWS plan to add more than 1 million GPUs beginning in 2026. The expanded relationship also includes NVIDIA Vera CPU infrastructure, Spectrum networking and NVLink Fusion architecture with custom high-bandwidth memory. Separately, AWS and NVIDIA plan secure U.S.-government AI infrastructure incorporating 100,000 GPUs. The 2-million-GPU figure is planned future deployment rather than installed semiconductor capacity.
Source: NVIDIA
Strategic Watch
Focus on deployment conversion rather than the headline GPU commitment. The key question is whether semiconductor and infrastructure capacity can remain synchronized across Blackwell Ultra, Rubin and Rubin Ultra through 2028. HBM allocations, advanced packaging, networking equipment and available data-center power will determine how much of the planned capacity becomes operational on schedule.
GreyRadius Insight
A two-million-GPU commitment effectively becomes a multi-year capacity signal for the semiconductor ecosystem. It gives suppliers stronger visibility but also raises concentration risk: a handful of hyperscalers increasingly influence foundry allocation, memory investment and packaging expansion years in advance. Semiconductor capacity planning is therefore becoming more tightly connected to hyperscaler infrastructure roadmaps.
Semtech Posts Record $341.9 Million Quarterly Revenue as AI Networking Demand Strengthens
August 25, 2026 | United States / Global
Semtech reported record fiscal Q2 2027 net sales of $341.9 million, increasing 17% sequentially and 33% year over year, with adjusted gross margin reaching 54.5%. Management reported accelerating bookings and record backlog, providing revenue visibility into the following fiscal year. Independent reporting highlighted adjusted earnings of $0.71 per share and a Q3 revenue expectation of about $410 million. The results indicate that AI semiconductor demand is expanding into networking, connectivity and signal-path silicon alongside compute accelerators.
Source: U.S. SEC
Strategic Watch
Watch whether record backlog converts into sustained growth across AI-data-center connectivity products rather than remaining concentrated in a limited number of programs. Increasing accelerator cluster sizes should make bandwidth, signal integrity and optical connectivity progressively more important.
GreyRadius Insight
AI semiconductor value is spreading outward from the accelerator. Every increase in compute density creates additional requirements for moving data between GPUs, memory and racks. This creates a second-order semiconductor opportunity in connectivity, optical interfaces and signal-path silicon. Investors and suppliers focused only on GPU volumes risk underestimating where incremental semiconductor content is accumulating.
Infineon Acquires Bengaluru-Based C2i to Strengthen AI Data-Center Power Delivery
August 24, 2026 | India / Germany / Global
Infineon announced the acquisition of C2i Semiconductors, a Bengaluru-based specialist in intelligent digital power management for AI data centers. C2i develops software-defined multiphase controllers, smart power stages and Substrate Integrated Voltage Regulator architectures designed for demanding processor power delivery. The transaction adds grid-to-core power-management capability and another AI-power R&D base in India. Financial consideration, manufacturing capacity, wafer size and process node were not disclosed.
Source: Infineon
Strategic Watch
Watch whether C2i’s digital controllers and SIVR architecture move from component integration into design wins around next-generation accelerator platforms. The most important signal will be power conversion moving progressively closer to the processor and package.
GreyRadius Insight
Power delivery is emerging as another physical limit to AI scaling. Higher accelerator density means performance can no longer be separated from how efficiently electricity is converted and delivered inside the system. This creates a strategic convergence between compute architecture, semiconductor packaging and power electronics—potentially expanding the AI semiconductor opportunity from GPUs and HBM into controllers, power stages, GaN/SiC devices and vertical power-delivery architectures.
