Reports
AI-generated structured vendor updates
Meta Invests $9.17B in Canada AI Data Center, Iris AI Chip Mass Production Begins MTIA Roadmap
Meta announced a $9.17B AI data center in Canada with 1GW capacity, and its first in-house AI chip Iris will mass produce in September, kicking off the MTIA four-generation roadmap. Meta targets 14GW compute by 2027, using 6-month chip iterations to challenge NVIDIA's annual cadence and reduce GPU dependency.
PrismML's 1-bit Compression: 27B Qwen Model Runs Fully on iPhone 17 Pro in 4GB
PrismML compressed a 27B-parameter dense LLM (Qwen 3.6) to 4GB, running fully on iPhone 17 Pro. Using native 1-bit quantization (weights as {-1, +1}), it achieves >92% compression, 8x faster inference, and 75-80% energy reduction. This challenges Apple's sparse architecture, potentially shifting edge AI from cloud-reliant to device-native.
Huawei Ascend 10K-Card Cluster Goes Live, UnifiedBus Protocol Pools All Resources
Huawei launched an Ascend 10,000-card AI cluster in Shaoguan, Guangdong, and showcased the Atlas 950 SuperPoD with its proprietary UnifiedBus interconnect supporting 8,192 NPUs at 16.3 PB/s. Huawei Cloud also entered the Gartner 2026 Cloud AI Infrastructure Leaders quadrant, reinforcing its push for a self-contained AI ecosystem.
Samsung GAIA AI PC Chip Samples with Memory-Centric NPU, Targeting 50 TOPS
Samsung launches GAIA AI PC processor with 4nm process and memory-centric NPU, integrating LPDDR5X controller with NPU for near-memory computing, achieving 40% energy efficiency improvement and 50 TOPS. Certified for Microsoft Copilot+ PC, Lenovo to adopt in Q4 2026.
AMD's Experimental Topological Ghost Protocol Boosts MI300X Inference 10x
AMD introduces experimental Topological Ghost Protocol (TGP) on MI300X GPUs, achieving 431 tokens/sec with 100% success in high-concurrency inference, 10x improvement over standard vLLM. TGP uses KV-cache recycling and segmented state management, still experimental but potentially redefining AI inference benchmarks.
OpenAI Reopens with GPT-oss Models: Apache 2.0 License Hides Cloud Offload Control
OpenAI launches GPT-oss-120b and GPT-oss-20b under Apache 2.0 license, capable of running on a single 80GB GPU. However, a built-in cloud offload mechanism routes complex queries to proprietary models, masking a strategic control point shift behind the open-source facade.
Google Gemini 3.5 Pro Rebuilds from Scratch: 2M Token Context Window Reshapes AI Frontier
Google DeepMind targets July 17 for Gemini 3.5 Pro, a full architectural rewrite of its pretraining stack to overcome deficits in math reasoning, SVG generation, and image quality. Specs include a 2M token context window, Deep Think reasoning layer, and multi-step autonomous workflows, though unconfirmed by Google.
SambaNova完成11亿美元融资估值110亿美元：推理芯片新格局确立
...
NVIDIA Rigel Core: Single-Threaded CPU as the New Control Plane for Agentic AI
NVIDIA unveils Rosa CPU architecture with custom Rigel core (Arm v9.2), targeting single-threaded performance for Agentic AI workloads, paired with Feynman GPU (1.6nm, 50 PFLOPS) in 2028. This shifts CPU design from core-count scaling to serial-latency optimization, directly challenging AMD EPYC and Intel Xeon dominance.
MediaTek and Alibaba Cloud Deploy Tongyi Qianwen LLM on Dimensity Chips
MediaTek partners with Alibaba Cloud to deploy a small version of the Tongyi Qianwen LLM on Dimensity 9300/8300 mobile platforms, enabling offline multi-turn conversations. This move aims to capture edge AI inference control via NPU optimization and SDK integration, directly challenging Qualcomm.
NVIDIA Denies Kyber NVL144 Delay, But 78-Layer PCB Bottleneck Exposes AI Hardware Physics Limit
NVIDIA officially denies reports of Kyber NVL144 rack delay to 2028, but SemiAnalysis revelations about a 78-layer ultra-high-density PCB midplane bottleneck and Rubin Ultra cancellation expose hard physical limits in signal integrity and manufacturing, opening a strategic window for AMD and Google.
Huawei Unveils Tao's Law V2: Kirin 2026 Boosts AI Inference 40% on Same Node
Huawei's He Tingbo releases Tao's Law V2, detailing Kirin 2026 metrics: 238 MTr/mm² transistor density (+55%), 41% power reduction at iso-performance, and 40% SRAM frequency increase. Without EUV lithography, co-optimization of architecture, circuit, and process delivers equivalent performance gains, proving system-level optimization as a viable alternative to Moore's Law scaling.
Anthropic Starts Custom AI Chip Development, Talks Samsung 2nm, Aims for Compute Independence
Anthropic has initiated its own AI chip development and is in talks with Samsung for 2nm foundry services. The move aims to reduce reliance on NVIDIA GPUs, optimize inference costs, and strengthen its technology moat ahead of a potential IPO. It joins OpenAI, Google, and others in the custom ASIC race, signaling a shift from software to hardware competition.
AMD Unveils Zen 6/7 CPU and MI400/500 GPU Roadmap, Targets NVIDIA Rubin with HBM4 and 2nm
AMD unveiled its Zen 6/7 CPU and MI400/500 GPU roadmap at its 2026 Financial Analyst Day, featuring TSMC 2nm process and HBM4 memory. The MI400 series boasts 432GB memory, 19.6TB/s bandwidth, and 40 PFLOPs FP4 performance, directly targeting NVIDIA's Vera Rubin architecture with an annual cadence to disrupt the AI hardware monopoly.
Anthropic's $15B Australia Bet: AI Infra Shifts to Energy Arbitrage
Anthropic plans to invest $15B to secure 1.4GW of data center capacity in Australia, aiming to activate 1GW by next year. This move bypasses US grid bottlenecks from local opposition and litigation, building a hybrid model of self-build, partnerships, and cloud leasing. It signals a shift in AI infra deployment toward energy and regulatory arbitrage.
Google Cloud Launches Blackwell GPU Confidential VM & Open-Source Prompt Encryption SDK, Redefining AI Security
Google Cloud upgrades its confidential computing portfolio with Blackwell GPU-based confidential VMs (Confidential G4 VMs preview), open-source Prompt Encryption SDK, and enhanced Confidential Space featuring Intel Trust Authority and Hopper GPU support, addressing TEE vulnerability CVE-2026-33697 to bolster AI inference and cross-organization training security.
Anthropic Launches Custom AI Chip: Vertical Integration to Control Inference Cost and Supply
Anthropic launched Claude Sonnet 5 and revealed a custom AI chip initiative, using Samsung foundry. This move aims to reduce dependency on NVIDIA, control long-term inference costs, and marks Anthropic's shift from a pure software company to a vertically integrated infrastructure firm.
OpenAI Winds Down Fine-Tuning API: A Strategic Shift in AI Customization Landscape
OpenAI plans to phase out its fine-tuning API by 2027, stopping new task creation but allowing inference on existing models. This forces startups relying on fine-tuning for differentiation to migrate to open-source models or RAG, reshaping the AI customization ecosystem.
OpenAI Ends Azure Exclusivity: Model Delivery Control Shifts from Microsoft to Multi-Cloud
OpenAI and Microsoft restructured their partnership in April 2026, ending exclusive Azure licensing and capacity commitments. OpenAI can now serve customers on any cloud; Microsoft retains right of first refusal and revenue share only on its platform. Driven by GPT-5.1's ~3 exaflops inference demand and FTC antitrust scrutiny.
NVIDIA Vera Rubin AI Platform Slated for July 2026 Shipments, Iterative Compute Upgrade
NVIDIA confirms its next-gen AI compute platform, Vera Rubin, will start shipping in July 2026 to major cloud providers like Microsoft and Google. The platform uses an advanced process node to boost AI training and inference performance, representing an iterative upgrade over Hopper and Blackwell without a fundamental architectural shift.