SoftBank Bets $30B on Asia's Largest AI Data Center

SoftBank Group is committing approximately $30 billion to construct what it says will be the largest AI data center in Asia, marking one of the most ambitious infrastructure investments in the region's history. The Japanese tech conglomerate's bold move positions it as a direct competitor to US hyperscalers like Microsoft, Google, and Amazon in the rapidly escalating global race for AI compute power.

The planned facility underscores a dramatic shift in SoftBank's strategy under founder Masayoshi Son, who has pivoted the company from a diversified telecom and investment holding group into an AI-first enterprise. This investment dwarfs most individual data center projects announced globally and signals that Asia intends to close the AI infrastructure gap with the West.

Key Facts at a Glance

• Investment size: Approximately $30 billion, making it one of the largest single AI infrastructure commitments in Asia
• Scale: Planned to be the largest AI data center on the Asian continent
• Strategic shift: Represents SoftBank's full pivot from telecom conglomerate to AI infrastructure powerhouse
• GPU ambitions: Expected to house hundreds of thousands of advanced AI chips, likely from Nvidia and potentially custom silicon
• Timeline: Construction expected to ramp through 2025-2030 with phased capacity rollout
• Global context: Competes with $100B+ AI infrastructure plans from Microsoft, Google, Amazon, and Meta in the US

Masayoshi Son Goes All-In on AI Infrastructure

SoftBank's $30 billion commitment represents a defining moment for Masayoshi Son, who has repeatedly described artificial intelligence as the most important technology humanity will ever create. Son has spent the past 2 years restructuring SoftBank's portfolio and messaging around a singular AI vision.

The investment follows SoftBank's earlier moves to deepen its AI footprint, including its partnership with Nvidia and its role in the Stargate joint venture alongside OpenAI, Oracle, and the US government. Son has publicly stated that SoftBank aims to become the world's leading AI infrastructure company, not merely an investor in AI startups.

This data center project differs from SoftBank's previous investment strategy in a crucial way. Rather than writing checks to promising startups through the Vision Fund, SoftBank is now building physical infrastructure it will own and operate. This shift from financial investor to infrastructure operator mirrors what Amazon did with AWS two decades ago — betting that owning the underlying compute layer is more valuable than owning the applications built on top of it.

Why Asia Needs Its Own AI Compute Backbone

The global AI infrastructure landscape remains heavily tilted toward the United States. Microsoft, Google, Amazon, and Meta have collectively announced well over $200 billion in AI data center spending for 2025 alone, with the vast majority of that capacity located in North America.

Asia, despite being home to some of the world's largest technology markets — Japan, South Korea, India, and Southeast Asia — has lagged significantly in dedicated AI compute capacity. Most Asian enterprises currently rely on US-based cloud providers for their AI workloads, creating latency issues, data sovereignty concerns, and supply chain vulnerabilities.

SoftBank's mega-facility addresses several critical gaps:

• Data sovereignty: Asian governments increasingly require sensitive AI training data to remain within regional borders
• Latency reduction: Local compute infrastructure dramatically improves inference speed for real-time AI applications
• Supply chain resilience: Reduces dependence on US-controlled cloud infrastructure during geopolitical tensions
• Enterprise demand: Japanese and Asian corporations need localized AI infrastructure for proprietary model training
• National security: Governments across Asia want domestic AI capabilities independent of foreign providers

Compared to Equinix and Digital Realty, which operate large data center portfolios across Asia, SoftBank's planned facility would be purpose-built for AI workloads from the ground up. This means optimized power delivery, advanced liquid cooling systems, and networking architectures designed specifically for GPU cluster communication — capabilities that retrofitted traditional data centers struggle to match.

The GPU Arms Race Intensifies

At the heart of any AI data center lies its GPU infrastructure, and SoftBank's $30 billion budget suggests an extraordinarily ambitious chip procurement strategy. Current pricing for Nvidia's H100 GPUs hovers around $25,000-$30,000 per unit, while the newer B200 chips command even higher premiums.

A facility of this scale could potentially house between 500,000 and 1 million advanced GPUs, placing it among the largest single concentrations of AI compute power anywhere in the world. For context, xAI's Memphis data center — often cited as one of the largest GPU clusters — houses approximately 100,000 Nvidia H100 GPUs.

SoftBank has also been exploring custom AI chip development through its subsidiary Arm Holdings, in which it holds a controlling stake. Arm's chip architecture already powers the vast majority of the world's smartphones and is increasingly being adopted for data center workloads. A SoftBank data center running on Arm-based custom AI accelerators could create a vertically integrated AI compute stack that rivals anything offered by US hyperscalers.

The power requirements alone are staggering. A facility of this magnitude would likely require multiple gigawatts of electricity, necessitating dedicated power generation infrastructure. SoftBank has previously invested in renewable energy projects across Asia, and the data center is expected to incorporate solar, wind, and potentially nuclear power sources to address both energy costs and sustainability concerns.

How This Reshapes the Global AI Competitive Landscape

SoftBank's investment sends a clear signal to the global technology industry: Asia will not cede the AI infrastructure race to the United States. The move has significant implications for multiple stakeholders.

For US hyperscalers, SoftBank's facility represents the emergence of a credible regional competitor. While Microsoft Azure, Google Cloud, and AWS all operate data centers in Asia, none have announced a single facility approaching this scale in the region. SoftBank's local relationships, regulatory expertise, and deep ties to the Japanese government give it advantages that foreign operators cannot easily replicate.

For Nvidia, the project represents a massive potential customer at a time when demand for AI chips already far exceeds supply. SoftBank's order could be worth billions of dollars in GPU sales alone, further cementing Nvidia's dominance in the AI accelerator market.

For Asian startups and enterprises, the facility promises democratized access to AI compute resources that have historically been controlled by Western companies. Localized infrastructure means lower costs, faster iteration cycles, and the ability to train models on region-specific data without cross-border transfer complications.

What This Means for the AI Industry

The practical implications of SoftBank's $30 billion bet extend far beyond Japan. This investment validates several important trends that are reshaping the global technology landscape.

First, it confirms that AI infrastructure is becoming a geopolitical asset, not just a commercial one. Governments increasingly view domestic AI compute capacity as essential to national competitiveness, similar to how semiconductor manufacturing became a strategic priority after the global chip shortage.

Second, it suggests that the AI infrastructure buildout is far from peaking. Despite concerns about overinvestment in AI, major players continue to announce increasingly large commitments. The total global investment in AI data centers is projected to exceed $500 billion by 2028, according to industry estimates.

Third, the investment highlights the growing importance of the inference economy. As AI models become embedded in everyday applications — from autonomous vehicles to healthcare diagnostics — the demand for local inference compute will dwarf current training requirements. SoftBank's facility is positioned to serve both training and inference workloads at massive scale.

Looking Ahead: SoftBank's AI Empire Takes Shape

SoftBank's $30 billion data center commitment is not an isolated bet but part of a broader strategy to build an end-to-end AI empire. The company's portfolio now spans chip architecture through Arm Holdings, AI infrastructure through this data center project, and AI applications through its investments in companies like Perplexity AI and its involvement in the Stargate initiative.

The key question is execution. SoftBank's track record with the Vision Fund included spectacular failures alongside its successes, and building and operating a world-class data center requires fundamentally different capabilities than making venture investments. The company will need to recruit top-tier data center engineering talent, secure reliable power supplies, and navigate complex semiconductor supply chains.

Timeline-wise, industry observers expect ground-breaking to begin in late 2025 or early 2026, with initial capacity coming online by 2027-2028. Full buildout could extend into 2030 or beyond, with capacity added in phases as demand materializes.

If successful, SoftBank's mega data center could fundamentally alter the balance of AI power between East and West. It would give Asian companies a home-grown alternative to US cloud providers for their most demanding AI workloads — and give Masayoshi Son the AI infrastructure crown he has long coveted. The stakes could not be higher, and the entire global tech industry will be watching.