Google Integrates Nvidia GPUs, Aalo Atomics Unveils Modular Nuclear Plant, and Pakistan to Power AI with Surplus Energy - Silicon Brief #13

Hi there,

In this edition of Silicon Brief, we're diving into the latest developments in AI chips and data centers. NTT has introduced a new AI chip designed for edge video processing, while Google has unveiled its Ironwood AI Chip, adding to the growing landscape of AI hardware innovations. Meanwhile, Cerebras Systems has launched a fast inference solution for the Llama 4 models, showcasing advancements in AI processing capabilities.

On the data center front, Meta is planning to build an AI-optimized data center in Ohio, and Stack Infrastructure has announced plans for a massive 450MW data center in Sydney. These developments highlight the ongoing expansion and technological advancements in data center infrastructure to support the increasing demands of AI applications.

Aalo Atomics has introduced the Aalo-1 reactor prototype, part of its fully modular nuclear plant, the Aalo Pod XMR, designed specifically for AI data centers. The unveiling coincides with the completion of its manufacturing facility in Austin, Texas. Pakistan plans to allocate surplus electricity to Bitcoin mining and AI data centers, as announced by Chief Advisor Bilal Bin Saqib. This initiative aims to address energy sector challenges and support the fintech sector. Bulgaria is planning to use its nuclear energy resources to attract investments in artificial intelligence and data centers, with the Kozloduy Nuclear Power Plant playing a central role in this strategy.

Wanclouds and Cytomate partner to enhance cloud and AI technology adoption in Qatar, supporting digital transformation efforts. The UALink Consortium announces the UALink 200G 1.0 Specification to improve communication in AI computing pods. SK Telecom collaborates with Giga Computing and SK Enmove to develop liquid cooling technologies for AI data centers. Informatica introduces AI-driven features in its Intelligent Data Management Cloud for enhanced data integration. Google plans to integrate Nvidia Blackwell GPUs and Google Gemini models into its cloud platform by Q3 2025. LG Electronics expands its AI data center cooling solutions with advanced liquid cooling systems.

Nvidia CEO Jensen Huang reportedly reached an agreement with the Trump administration to prevent export restrictions on the company's H20 AI chips. This follows a promise to invest in new AI data centers in the U.S.
The US government is investigating TSMC for allegedly violating export control regulations by supplying chips used in Huawei's AI processor, with potential fines exceeding $1 billion.
A draft bill in the U.S. Senate proposes emissions fees for data centers supporting blockchain and AI operations. The 'Clean Cloud Act' aims to set federal emissions standards for these facilities.
The Federal Energy Regulatory Commission has denied a request to reconsider its decision blocking Amazon's data center in Pennsylvania from increasing power use connected to Talen Energy's nuclear plant, citing concerns over power reliability and public costs.

KAYTUS has introduced KSManage V2.0, a platform designed to enhance data center management efficiency and reduce energy consumption. NxtGen Datacentre and Cloud Technologies has engaged Avendus to raise $300 million, aiming for a $900 million valuation. MAGIC Research has launched Fabric Hypergrid, a private AI platform that cuts computing costs by 90%, enabling organizations to use existing hardware while ensuring data privacy and security.

NTT has launched an AI inference chip for Edge video processing, targeting commercialization in 2025. SambaNova and Meta collaborate to offer Llama 4 models on SambaNova Cloud, enhancing performance with RDU chips. ZeroPoint Technologies AB and Rebellions partner to develop memory-optimized AI accelerators for improved AI workload efficiency.