Reports
AI-generated structured vendor updates
DeepSeek Plans $35B 1GW AI Data Center, Shifting from Rented to Owned Infrastructure
DeepSeek plans a 1GW AI data center in Inner Mongolia with an estimated $35 billion investment, shifting from renting chips from Alibaba to owning dedicated infrastructure. The project leverages low electricity costs and cool climate, aligning with China's Eastern Data Western Computing strategy.
DeepSeek Claims China Will Phase Out NVIDIA in One Year with Huawei Ascend 950 and Self-Developed Tile Language
DeepSeek founder claims China will phase out NVIDIA within a year, with Huawei Ascend 950 SuperPod matching GB200/GB300 performance. DeepSeek is developing its own inference chips and Tile Language to reduce CUDA dependency, marking a critical shift from import substitution to an independent AI compute ecosystem.
Alibaba Cloud Unveils Agent-Native Suite and Open-Source SAIL Stack to Rival CUDA
At WAIC 2026, Alibaba Cloud launched its Agent-Native cloud suite (AgentLoop, AgentTeams, AgentRun, TokenWorks), open-sourced the T-Head SAIL AI software stack, and unveiled the 2.4T-parameter Qwen 3.8-Max-Preview model and Zhenwu M890 supernode, marking a comprehensive push into agent-native cloud and open-source AI ecosystem.
NVIDIA and Wistron Launch US-Based GB300 Production, Shifting AI Manufacturing Landscape
NVIDIA partnered with Wistron to open a $700M factory in Texas, achieving L6 integration of the GB300 Grace Blackwell Ultra system. The first US-made unit contains 1.5M parts, weighs 2 tons, and costs $4M. This milestone accelerates US AI capex localization from 5% to 30%, reshaping the global AI supply chain.
NVIDIA Vera Rubin Driving Performance Per Watt, Lowest Token Cost for Partners Worldwide
...
Moonshot AI Launches Kimi K3: 2.8T Parameter Open-Source MoE Model at $3/$15
Moonshot AI unveils Kimi K3, a 2.8T parameter open-source MoE model with 896 experts (16 active), native vision, and 1M context. API pricing at $3/$15 input/output per million tokens undercuts rivals. Open weights release July 27. GPU capacity exhausted within 2 days. Arena score 1679 tops Fable 5.
NVIDIA Vera Rubin Platform and Dynamo 1.0 Disaggregate Inference, Shift Focus to Intelligence per Dollar
NVIDIA unveils Vera Rubin platform with a 7-chip stack (Vera CPU, Rubin GPU, NVLink 6, etc.) and Dynamo 1.0 inference disaggregation. A single NVL72 rack packs 72 GPUs/36 CPUs with 1.6 PB/s bandwidth, achieving up to 7x inference performance. The new 'intelligence per dollar' metric signals a shift from training to inference cost competition.
Alibaba Launches 2.4T Parameter Qwen3.8-Max MoE Model with 0.2x Pricing
Alibaba released Qwen3.8-Max-Preview, a 2.4 trillion parameter MoE multimodal model with 1M context window. It launched Qoder platform and Token Plan with aggressive discounts up to 0.2x, significantly reducing inference cost. The company claims it is second only to Anthropic Fable 5, marking China's AI entry into dual-track of parameter arms race and open-source competition.
Huawei Atlas 950 SuperPoD & 灵衢2.0: A Systemic Pivot in China's AI Compute from Chip to Cluster
At WAIC 2026, Huawei publicly demonstrated the Atlas 950 SuperPoD, a 1024-ascend NPU card cluster, and unveiled the 灵衢2.0 high-speed interconnect protocol. This signals a strategic shift in China's AI infrastructure from single-chip to system-level leadership, creating a closed-loop ecosystem that directly challenges NVIDIA's NVL series dominance.
Microsoft Replaces OpenAI/Anthropic with In-House MAI Models to Cut Costs and Reduce Dependency
Microsoft has started replacing OpenAI and Anthropic AI calls in Excel and Outlook with its in-house MAI models, handling tens of thousands of prompts weekly. The move aims to cut costs and reduce dependency on Anthropic, signaling a strategic shift toward internal AI models and impacting the AI vendor ecosystem.
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.
When AI Invents the Attack: Browser-Native Ransomware
...
Anthropic Claude Goes Exclusive on Azure, Microsoft Locks AI Model Distribution via GB300
Anthropic's Claude models are now generally available on Azure Foundry, powered by NVIDIA GB300 NVL72 clusters with over 4600 Blackwell Ultra GPUs. Initial models include Opus 4.8 and Haiku 4.5 with prompt caching and extended thinking. Microsoft gains exclusive enterprise distribution, strengthening its competitive position against AWS and Google Cloud.
OpenAI GPT-5.6 Aggressive Pricing and 1.5M Context Window Targets Agent Era
OpenAI reportedly launches GPT-5.6 with 1.5M token context window, aggressive pricing at one-third of Claude Fable 5, and improved agent reliability. This move capitalizes on Anthropic's forced downtime and addresses internal alignment issues.
Microsoft Shifts Copilot Cowork to Usage-Based Pricing, Eyes DeepSeek for Cost-Efficiency
Microsoft transitions Copilot Cowork to usage-based billing (Copilot Credits) and considers integrating fine-tuned DeepSeek V4 or open-source models as low-cost alternatives, hosted on Azure. This move addresses high costs from intensive usage and signals a multi-model strategy.
US Export Controls Halt Anthropic's Fable/Mythos: AI Geopolitical Precedent Set
The US Commerce Department suspends access to Anthropic's Fable 5 and Mythos 5 models for all foreign nationals, including Anthropic's own foreign employees, citing national security. Models are taken offline immediately. Anthropic dispatches executives to Washington for negotiations, marking a potential turning point for AI export controls.
Huawei's LogicFolding: 3D Stacking Rewrites AI Chip Rules
Huawei's Tau Scaling Law and LogicFolding architecture boost transistor density by 55% and power efficiency by 41% via vertical logic stacking, targeting 1.4nm-class by 2031. Ascend 920/910C chips are now used for DeepSeek V4-Pro post-training, signaling real-world AI workload deployment and challenging Nvidia's dominance in China.
OpenAI buys Ona: Control point shifts to persistent AI agent runtime
OpenAI acquires cloud infrastructure startup Ona to integrate its persistent execution environment into Codex, enabling AI agents to run independently for hours or days in enterprise-owned clouds. This addresses security, governance, and audit requirements, signaling OpenAI's shift from model provider to full-stack AI platform.
NVIDIA Blackwell Sweeps MLPerf: NVLink and NVFP4 Redefine AI Training Economics
NVIDIA Blackwell dominates MLPerf Training 6.0, submitting across all seven benchmarks including MoE workloads. GB300 NVL72 delivers up to 1.6x faster training than GB200, with fifth-gen NVLink unifying 72 GPUs as one giant GPU. NVFP4 low-precision training and massive scale (8,192 GPUs) set new industry standards.