Microsoft and EY to Accelerate Enterprise AI Adoption with Over $1 Billion Investment

Microsoft and consulting giant EY have announced a partnership that will invest more than $1 billion (approximately 150 billion yen) to move client companies' AI projects from the pilot stage to large-scale deployment.

In an interview, Errol Gardner, EY's Vice Chair of Global Consulting, noted that "it's at the large-scale deployment stage that clients truly see a return on their investment." Enterprise AI initiatives have often stalled at the proof-of-concept (PoC) stage, and the transition to production environments has been a challenge across the industry.

The two companies also plan to develop joint solutions in areas such as AI agents and workflow automation.

Trump Administration Commits $2 Billion to Quantum Computing as IBM Builds a New Foundry

The Trump administration has revealed plans to award IBM a $1 billion (approximately 150 billion yen) grant to support the construction of a foundry for manufacturing quantum computing chips.

In conjunction, IBM will establish a new company, Anderon, and invest $1 billion of its own to manufacture the processors. The move is part of a broader strategy to foster the U.S. quantum computing industry and expand its technological leadership, against a backdrop of growing expectations for quantum computing in AI as well.

Taiwan Makes First Bust of Nvidia AI Chips Smuggled to China

Taiwanese authorities have sought the detention of three individuals for illegally exporting Nvidia AI chips to China. This is the first time Taiwan has cracked down on semiconductor smuggling.

The suspects allegedly made false declarations about AI servers manufactured by U.S.-based Super Micro Computer and attempted to export them to China, Hong Kong, and Macau. The United States has restricted exports of high-performance AI chips to China since 2022, and the case highlights ongoing attempts to circumvent these controls.

AI chip supply controls remain one of the major flashpoints in U.S.-China tensions, and this crackdown could influence future regulatory tightening.

Tencent Open-Sources "Hy-MT2," a Translation Model Supporting 33 Languages

Tencent has released Hy-MT2, a model family specialized in multilingual translation. It comes in three sizes — 1.8B, 7B, and 30B-A3B (MoE) — and supports translation across 33 languages.

Notably, the 7B and 30B-A3B models outperform DeepSeek-V4-Pro and Kimi K2.6 in fast-thinking mode. The lightweight 1.8B model is also said to beat commercial APIs such as those from Microsoft and Doubao overall.

For on-device deployment, extreme 1.25-bit quantization via AngelSlim is also available, keeping the storage requirements of the 1.8B model down to 440MB. The company will also participate as a partner in the WMT26 video subtitling translation task.

Baidu Releases High-Performance Training Framework "LoongForge"

Baidu has open-sourced LoongForge, a high-performance training framework that supports a wide range of AI models, including LLMs, VLMs (vision-language models), VLAs (vision-language-action models), and Wan.

It aims to address challenges in training large-scale models, such as improving distributed training efficiency and memory optimization. The project has been drawing attention on Hacker News, underscoring continued investment by Chinese tech companies in open-source AI infrastructure.

Hitting 110 tok/s on a Consumer GPU — ik_llama.cpp Optimizations Draw Attention

A report on Reddit's LocalLLaMA community is drawing attention: running the Qwen3.6 35B A3B model on an RTX 4070 Super (12GB VRAM) at 110 tok/s.

While stock llama.cpp delivers around 90 tok/s, using the optimized fork ik_llama.cpp achieves a 23% speedup. It leverages the Multi-Token Prediction (MTP) feature to reach this speed while maintaining a 128K context length.

With its strength in CPU offload optimization, ik_llama.cpp is becoming a compelling option for comfortably running large models near the VRAM limit on consumer hardware.