Semiconductors & Infrastructure

Microsoft announced its in-house AI accelerator, Maia 300, revealing plans to reduce reliance on NVIDIA while significantly improving power efficiency.

• •The production deployment of the third-generation in-house AI accelerator, Maia 300, is scheduled to begin in 2027.
• •The company has unveiled a strategy to scale the Maia family to multi-gigawatt levels.
• •It aims to increase generated tokens per unit of power by 40% compared to GPU-based systems.

AirTrunk secured a $1 billion green loan and introduced Japan's first liquid-cooling-capable capacity at Data Center TOK1.

• •AirTrunk announced the procurement of a $1 billion (approx. 155 billion yen) green loan at a media briefing held on the 7th.
• •Using the secured funds, the company introduced Japan's first liquid-cooling-capable capacity.
• •This initiative aims to advance cooling technology at the "TOK1" data center.

Microsoft announced its third-generation AI chip, Maia 300, featuring TSMC's 2nm process and HBM4, with plans for a 2027 launch as a core infrastructure for its services.

• •Designed specifically for inference, it improves power efficiency by 40% compared to GPU-based systems and achieves up to seven times the processing power of the previous generation.
• •It adopts TSMC's 2nm process technology and is configured with 288GB of HBM4 memory.
• •It will be deployed at a scale of several gigawatts and positioned as the primary inference infrastructure for Microsoft's own services.

The current situation where memory and storage shortages and price surges driven by AI demand are worsening in the consumer market from Q4 2025 through 2026.

• •Memory and storage shortages and price increases began to manifest in the consumer market during the fourth quarter of October to December 2025.
• •It is pointed out that the situation has not improved in 2026 and is instead worsening further.
• •Abnormal semiconductor shortages caused by AI-specific demand are negatively affecting product supply and prices in the consumer market.

Overview of Janas, a tool that executes large LLMs exceeding RAM capacity via SSD streaming, achieving faster inference than llama.cpp.

• •Qwen3-Next-80B (48GB) was processed at 37 tok/s on a 32GB laptop, significantly outperforming llama.cpp's 2.8 tok/s.
• •All 13 models compared in Q4_K_M quantization format recorded faster read speeds than llama.cpp.
• •Specific performance figures were presented: 24 tok/s without drafting and 62 tok/s for reading.

Intel announced the D270, a data center GPU for inference, enabling agent AI, long-context, and high-parallel serving at low cost.

• •It is a data center GPU for inference optimized for agent AI, long-context, and high-parallel serving.
• •It is designed to operate with practical power consumption and cost.
• •It features up to 480 GB of LPDDR5X memory.

Sakura Internet began offering "Sakura AI Engine Private Edition" on October 8, allowing enterprises to use desired generative AI models in a dedicated environment.

• •Sakura Internet began offering the enterprise-exclusive generative AI platform "Sakura AI Engine Private Edition" on October 8.
• •It features the ability to use user-selected generative AI models in a dedicated (private) environment.
• •It is a service assuming on-premises-like operation, considering corporate data security and compliance.

Agent37がAIエージェント実行用のサンドボックスを提供し、競合他社より10%以下のコストで月額0.37ドルを実現

• •AIエージェントを実行するためのサンドボックス環境を提供し、主要なサンドボックスプロバイダーと比較してコストを10%以下に抑えている
• •顧客の1サンドボックスあたりの平均支出は月額0.37ドルである
• •2月にOpenClawラッパー企業としてスタートし、その後複数のハーネスを追加して機能を拡張した

This article details the Azure Maia 300 architecture and explains Microsoft's efforts to make advanced AI features more accessible, affordable, and useful.

• •A detailed explanation of the Azure Maia 300 architecture is provided.
• •The goal is to make advanced AI features more accessible.
• •The aim is to improve the cost-effectiveness and utility of AI features.

Microsoftが自社開発の第3世代AIチップ「Maia 300」を発表し、クラウドにおけるAIワークロードのコスト効率を向上させるとともに、最新の推論モデルやエージェントタスクへの対応を強化した

• •Maia 300はMicrosoftの製品やサービスにおける「主要なAI推論システム」として設計されている
• •クラウド上でAIワークロードを実行する際のコストと効率性が改善されている
• •最新の推論モデルやエージェントタスクを処理する能力が向上した