東京工科大学、NVIDIA Blackwell搭載のAIスパコン「青嵐」を公開！大規模AI研究を劇的に加速

日本のAI研究コミュニティに新たな風が吹き込みました。東京工科大学は2025年10月2日、米NVIDIAの最新GPUアーキテクチャ「Blackwell」を採用したAIスーパーコンピュータ「青嵐（SEIRAN）」を公開し、10月1日より本格稼働を開始したと発表しました。これは、日本の私立大学が導入するAIスパコンとしては国内最大規模を誇り、大規模なAI研究と次世代AI人材育成の拠点として大きな注目を集めています。

「青嵐」とは何か？その驚異的な性能と私大最速の所以

「青嵐」は、東京工科大学が掲げる「AI大学」構想の中核を担う戦略的インフラです。NVIDIA DGX B200システムを12台導入し、合計96個の最新型GPUを搭載しています。これらはNVIDIA Quantum InfiniBandの高速ネットワークで接続されており、システム全体のAI学習理論性能はFP8精度で驚異の0.9EFLOPS（毎秒90京回計算）に達します。 この性能は、日本の私立大学が保有するAIスパコンとしては最高速であり、世界のスーパーコンピュータランキング「TOP500」へのランクインも視野に入れています。 特筆すべきは、理化学研究所のスーパーコンピュータ「富岳」と同程度のAI処理性能を持ちながら、消費電力はおよそ176分の1に抑えられている点です。

NVIDIA BlackwellアーキテクチャがAIの未来を拓く

「青嵐」の心臓部を成すのは、NVIDIAの最新GPUアーキテクチャ「Blackwell」です。Blackwellは、生成AIや大規模言語モデル（LLM）など、計算負荷の高いAIワークロードのために特別に設計されており、前世代のHopperアーキテクチャと比較して、AI処理性能が飛躍的に向上しています。例えば、GPT-MoE 1.8Tモデルの学習性能で3倍、リアルタイム推論では15倍もの高速化を実現するとされています。

2080億個のトランジスタを搭載し、10 TB/sのチップ間相互接続によって2つのダイを統合する革新的な設計は、大規模AIモデルの処理において圧倒的な信頼性、可用性、サービス性を提供します。また、第2世代Transformer EngineとBlackwell Tensorコアは、LLMの学習と推論を加速し、FP4 AIを可能にすることで、高い精度を維持しながら次世代モデルのパフォーマンスとサイズを倍増させます。

大規模AI研究と人材育成への多大な影響

この強力な「青嵐」の導入は、東京工科大学のAI研究と教育に計り知れない影響をもたらします。学生は24時間体制で高性能なGPU環境に自由にアクセスできるようになり、大規模データと計算環境を駆使した実践的なAIスキルを習得する機会が飛躍的に増大します。 具体的には、以下のような多岐にわたるプロジェクトが計画されています。

• 独自の日本語大規模言語モデル（LLM）の開発と活用
• AIデジタルヒューマン制作プロジェクト
• NVIDIA Omniverseを用いたデジタルツインの構築
• 説明可能AI（XAI）による研究
• AI倫理とガバナンス検証環境の整備
• 大規模な流体シミュレーション

東京工科大学は、AIテクノロジーセンターを中心に、これらのプロジェクトを通じて最先端のAI技術を社会実装し、日本のAI分野におけるイノベーション創出のリーダーシップを目指しています。

地域社会との連携、そして未来へ

「青嵐」は学術研究に留まらず、地域社会への貢献も視野に入れています。東京工科大学は八王子市とAI/DX連携協定を締結し、「青嵐」を活用した自動運転バスやスマート農業の実証実験などを通じて、高齢化や人手不足といった地域の課題解決に取り組む計画です。 この官学連携は、AI技術が地域経済と市民生活の質の向上に直接貢献する好事例となるでしょう。

「AI大学」として、東京工科大学はAIの基礎から応用、そしてAI倫理に至るまで、幅広い教育を全ての学生に提供することで、AIを「正しくつくり、正しく使う」ことのできる次世代AI人材の育成を推進します。 「青嵐」の稼働は、日本のAI研究と教育を新たなステージへと押し上げ、来るべきAI社会を牽引する重要な一歩となることに間違いありません。

Japan’s AI Frontier: Tokyo Tech Unleashes ‘Seiran’ – A Blackwell-Powered Supercomputer That Redefines Research

In a monumental leap for artificial intelligence research and education in Japan, the Tokyo University of Technology (TUT) officially unveiled its groundbreaking AI supercomputer, “Seiran,” on October 2, 2025. This state-of-the-art system, powered by NVIDIA’s latest Blackwell GPU architecture, is set to become a cornerstone for large-scale AI innovation, solidifying TUT’s position at the forefront of the nation’s technological advancement. Seiran is already being heralded as the “fastest among private universities in Japan” and is poised to make its mark on the global stage, with aspirations to rank among the prestigious TOP500 supercomputers worldwide.

Unleashing Unprecedented Power: Seiran’s Core Architecture

At the heart of Seiran’s formidable capabilities lies a robust configuration of 12 NVIDIA DGX B200 systems, housing a staggering total of 96 NVIDIA Blackwell GPUs. These powerful units are interconnected by NVIDIA’s Quantum InfiniBand high-speed network, all built upon NVIDIA’s recommended DGX BasePOD design to ensure optimal performance and scalability. This marks a historic first for a private university in Japan, being the inaugural institution to adopt an AI supercomputer utilizing the NVIDIA DGX series.

The theoretical AI performance of Seiran, specifically in FP8 precision, is an astonishing 0.9 EFLOPS (exaflops), translating to an ability to perform 90 quintillion calculations per second. This immense computational power is not just about raw speed; it also boasts remarkable energy efficiency. Reports indicate that Seiran’s AI processing performance is comparable to RIKEN’s Fugaku supercomputer, yet it achieves this with approximately 1/176 of the power consumption, underscoring a significant stride towards sustainable high-performance computing. Benchmark testing is slated for the near future, and if Seiran achieves 80% of its theoretical performance, it is fully expected to secure a spot on the coveted TOP500 list, a testament to its world-class capabilities.

NVIDIA Blackwell: The Engine of a New AI Era

The NVIDIA Blackwell architecture, unveiled in March 2024, represents a monumental leap in GPU technology, purpose-built for the insatiable demands of generative AI and accelerated computing. Each Blackwell GPU is an engineering marvel, packing an astounding 208 billion transistors and fabricated using a custom TSMC 4NP process. Its innovative design features two reticle-limited dies seamlessly connected by a 10 terabytes per second (TB/s) chip-to-chip interconnect, creating a unified and immensely powerful processing unit.

Key to Blackwell’s transformative power is its second-generation Transformer Engine. This engine is specifically designed to accelerate both the inference and training of large language models (LLMs) and Mixture-of-Experts (MoE) models. It introduces new precisions, including 4-bit floating-point (FP4) AI, which effectively doubles the performance and the size of next-generation models that memory can support, all while maintaining high accuracy. Furthermore, the fifth-generation NVLink interconnect allows Blackwell to scale its capabilities across up to 576 GPUs, providing a phenomenal 1.8 TB/s of total bandwidth per GPU and up to 130 TB/s in a 72-GPU NVLink domain, enabling unprecedented model parallelism across multiple servers.

Beyond raw performance, Blackwell also integrates NVIDIA Confidential Computing, a hardware-based security feature that safeguards sensitive data and AI models from unauthorized access. A dedicated RAS Engine ensures intelligent resiliency by continuously monitoring hardware and software health, proactively preventing downtime and optimizing system stability. These advancements collectively lead to a significant reduction in the training and operational costs of AI models, thereby accelerating research and development across diverse industries.

Accelerating AI Education and Pioneering Research at Tokyo Tech

Seiran’s deployment is a pivotal step in Tokyo University of Technology’s ambitious “AI University” vision. This initiative aims not only to cultivate students in information science disciplines but also to empower students across all departments to become adept in applying AI technologies. The supercomputer is primarily intended for educational purposes, providing an invaluable environment where students and faculty can engage in AI learning and research around the clock, with full availability expected from early November 2025.

The robust infrastructure of Seiran will enable cutting-edge research in various domains, including the development of unique large language models (LLMs), large-scale fluid simulations, and in-depth collaborative research with industry partners. This access to a high-performance GPU environment is considered extremely valuable for students, especially given the almost 100% utilization observed with the university’s previous mini-server equipped with NVIDIA A100 GPUs for computer science students in 2023.

Broader Impact: Driving Regional Innovation and Societal Advancement

Beyond academic pursuits, Seiran is also set to play a crucial role in regional development. Tokyo University of Technology has forged a strategic partnership with Hachioji City, where the campus is located, signing an AI/DX collaboration agreement. This collaboration aims to leverage Seiran’s immense computational power to address critical local challenges, such as ensuring the sustainability of regional agriculture amidst an aging population and labor shortages, maintaining traffic infrastructure, and promoting the digital transformation of administrative services. Plans include practical demonstrations of autonomous driving buses and smart agriculture, demonstrating the real-world applications of advanced AI. This initiative highlights a commitment to an “AI-Wellbeing” societal vision, where cutting-edge technology directly contributes to the well-being of the community.

The Future is Now: Seiran’s Role in Japan’s AI Landscape

The unveiling of Seiran powered by NVIDIA Blackwell is more than just a technological upgrade; it’s a strategic investment in Japan’s future AI capabilities. By providing unparalleled computational resources, Tokyo University of Technology is fostering a new generation of AI talent and driving innovation that will have far-reaching impacts across academia, industry, and society. As AI continues to reshape our world, Seiran stands as a beacon of progress, ready to accelerate the next wave of scientific discoveries and technological breakthroughs, not just for Japan, but for the global AI community.