Connect with us

Economy & Tech

Sharp, KDDI to Build New AI Data Center in Osaka

Four companies are joining forces to build one of Asia's largest AI data centers on Sharp's Sakai LCD plant site, equipped with approximately 1,000 servers.



Sharp headquarters (foreground) and Sakai Display Product on April 28, Sakai City, Osaka Prefecture. (©Kyodo)

On June 3, it was revealed that four Japanese and American companies — Sharp, KDDI, Super Micro Computer, and Datasection — agreed to begin discussions on constructing an artificial intelligence (AI) data center in Japan. Datasection specializes in AI-based contract system development.

The data center will be built on the site of the Sakai Display Product (SDP) factory in Sakai City, Osaka. SDP is a subsidiary of Sharp that produces liquid-crystal display (LCD) panels. However, its factory will cease production by September.

Sharp's pioneering LCD business became a major cause of the company's financial troubles due to declining demand. Now, Sharp is shifting its focus to the rapidly expanding generative AI sector in an effort to revive the company.

Need for Advanced Data Centers

From OpenAI's ChatGPT to Google's Gemini of the United States, conversational AI relies on massive language models trained on extensive text data. This training enables natural language generation, summarization, and dialogue, necessitating advanced data centers equipped with numerous servers.

Once completed, the AI data center on the SDP premises will be one of the largest in Asia.

The servers will be sourced from Super Micro Computer, a leading American server producer. They will be equipped with NVIDIA's new graphics processing units (GPUs) named Blackwell. Approximately 1,000 servers will be procured for the new data center.

Compared to its predecessors, Blackwell boasts significant performance improvements, including up to a 25-fold increase in power efficiency. Despite these advancements, operating the servers still requires substantial power. However, repurposing the facilities at the former SDP factory should secure an ample power supply.

Sharp and Sakai Display Product buildings at Green Front Sakai on April 2, 2016, Sakai City, Osaka Prefecture. (©Sankei by Tomoichiro Takekawa)

Sharp's Ambitious Plan

With forecasts predicting rapid expansion in the generative AI market, Sharp sees this as an opportunity for its resurgence.

Construction of Sharp's SDP factory began in 2009. However, the profitability of its LCD business decreased as demand dropped. It also faced intense competition from Chinese and South Korean companies. Consequently, Sharp gradually divested its shares in SDP. In 2016, Taiwan's Hon Hai Precision Industry acquired Sharp.

Under Hon Hai's leadership, SDP was re-subsidiarized in 2022, causing Sharp to fall back into the red. At a financial results briefing on May 14, 2024, Sharp announced it would shut down SDP's operations and repurpose it as a data center for AI. Sharp emphasized the necessity of "building data centers capable of handling the rapidly increasing AI processing demands."

The SDP factory will close without realizing its potential in the liquid crystal business. Now, as it transforms into an AI-focused data center, it remains to be seen whether it can capture the market trend this time around.

Potential Impact

Professor Atsushi Osanai at Waseda University's Graduate School views Sharp's shift to a new technological field positively. "Concentrating management resources on a single technology left the company vulnerable," he explains. "Sharp can expect synergistic benefits by integrating its communication-related technologies with data center operations."

Osanai further highlights that it would be advantageous for Japan's national security to have its own large data center instead of "relying on servers in other countries."

Having a large, centralized data center offers further advantages. "Delays in transferring information between servers can lower service quality," Osanai notes. "That means inter-server coordination is extremely important."

Regarding the location of the new data center, Osanai says, "I've heard that the AI data center will be one of the biggest in Asia. Its location is also ideal [in terms of size]."

Professor Atsushi Osanai.

Generative AI is employed across diverse applications like content generation, language translation, and chatbots. The AI chatbot GPT-4o, unveiled in May, enables seamless dialogue with a voice-to-voice feature. Similarly, Google introduced Gemini, a swift and lightweight generative AI chatbot. The capabilities and applications of generative AI are expanding at an astonishing pace.

According to estimates by research firms Global Information and MarketsandMarkets, the global generative AI market is poised to skyrocket from around $20.9 billion USD in 2024 to about $136.7 billion USD by 2030. This is nearly 6.5 times its current size. This surge is expected to be driven by companies relying on generative AI for service delivery and analytics.

AI Risks and Regulations

As generative AI becomes more accessible to everyone, concerns regarding its potential for misinformation and violating human rights have been raised globally. This has led various countries to take regulatory actions.

In May, the European Union (EU) implemented the world's first comprehensive AI regulation law. This law requires watermarking AI-generated content and imposes significant fines for non-compliance. Similarly, in the US, a presidential executive order issued in October 2023 mandates that "the most powerful AI systems share their safety test results and other critical information with the US government."

Meanwhile, Japan's AI guidelines for relevant stakeholders, released in April 2024, lack legal enforceability. To address this, a panel of experts convened on May 22 to discuss regulations aimed at ensuring the safety of generative AI. The panel is focused on assessing the necessity and specifics of regulations targeting major AI developers.


(Read the article in Japanese.)

Authors: Hiroto Kuwajima and Yohei Ushijima, The Sankei Shimbun