SoftBank Pours $5B Into Japan's Rapidus AI Chip Bet

SoftBank Group has committed a massive $5 billion investment in Rapidus, the Japanese semiconductor startup racing to produce cutting-edge AI chips. The deal marks one of the largest single investments in Asia's semiconductor sector and signals SoftBank's deepening bet on AI infrastructure beyond software and services.

The investment positions Rapidus — a company barely 2 years old — as a serious contender in the global AI chip manufacturing race, currently dominated by TSMC in Taiwan and Samsung in South Korea. For SoftBank CEO Masayoshi Son, the move represents a strategic pivot toward owning a piece of the physical backbone that powers artificial intelligence.

Key Facts at a Glance

• Investment size: $5 billion from SoftBank Group into Rapidus
• Chip target: Rapidus aims to manufacture 2nm process node chips, among the most advanced in the world
• Timeline: Mass production targeted for 2027
• Government backing: Japan's government has already pledged over $10 billion in subsidies to Rapidus
• Technology partner: Rapidus has a licensing deal with IBM for its Gate-All-Around (GAA) transistor technology
• Strategic goal: Reduce Japan's — and SoftBank's — dependence on TSMC for advanced AI semiconductors

SoftBank Doubles Down on AI Hardware Infrastructure

SoftBank's $5 billion commitment to Rapidus is not an isolated move. The Japanese conglomerate has been systematically building an AI empire that spans the entire technology stack, from chips to cloud to applications.

In recent months, SoftBank has announced plans to invest up to $100 billion in AI infrastructure in the United States through its Stargate joint venture with OpenAI and Oracle. The company also acquired Arm Holdings — the chip architecture firm whose designs power virtually every smartphone on the planet — in a deal that has proven prescient as AI workloads expand beyond data centers.

The Rapidus investment adds a critical manufacturing layer to SoftBank's portfolio. Unlike Arm, which designs chip architectures but doesn't manufacture them, Rapidus aims to actually fabricate cutting-edge semiconductors. This gives SoftBank potential vertical integration — designing chips through Arm and manufacturing them through Rapidus.

'Son has been vocal about his belief that artificial superintelligence is coming within a decade,' noted semiconductor industry analysts. This investment suggests he wants to control the hardware pipeline that makes it possible.

Rapidus: From Zero to 2nm in Record Time

Rapidus was founded in August 2022 with an audacious mission: to bring Japan back to the forefront of semiconductor manufacturing. The company's name — derived from the Latin word for 'fast' — reflects its aggressive timeline to achieve what typically takes decades of incremental progress.

The startup has several key advantages working in its favor:

• IBM partnership: Access to IBM's 2nm GAA nanosheet technology, proven in research labs at Albany, New York
• Government support: Over $10 billion in committed Japanese government subsidies, making it one of the most heavily state-backed chip ventures in history
• Talent recruitment: Aggressive hiring from former Toshiba, Renesas, and Intel engineers
• Greenfield advantage: Building a brand-new fab in Chitose, Hokkaido, with no legacy equipment or processes to work around

However, skeptics remain. Manufacturing at the 2nm node is extraordinarily difficult. Currently, only TSMC and Samsung have demonstrated the ability to produce chips at 3nm scale, and even they face significant yield challenges. Intel, despite decades of manufacturing experience and billions in investment, has struggled with its own advanced node transitions.

Rapidus essentially needs to leapfrog several generations of manufacturing know-how in just a few years — a feat no company has ever accomplished.

Why Japan Is Betting Big on Semiconductor Independence

Japan's push to revive its semiconductor industry is driven by both economic ambition and national security concerns. In the 1980s, Japan dominated global chip production, accounting for more than 50% of worldwide semiconductor revenue. Today, that share has fallen below 10%.

The geopolitical landscape has made this decline feel existential. With TSMC manufacturing roughly 90% of the world's most advanced chips in Taiwan — an island that faces ongoing tensions with China — governments worldwide are scrambling to diversify their semiconductor supply chains.

The United States passed the CHIPS and Science Act, committing $52.7 billion to domestic chip manufacturing. The European Union launched its own European Chips Act with €43 billion in public and private investment. Japan's support for Rapidus fits squarely within this global trend.

For SoftBank specifically, dependence on TSMC poses a business risk. As the company builds out AI data centers and develops custom AI processors, having a domestic manufacturing partner could provide both supply chain security and cost advantages.

The Global AI Chip Manufacturing Race Intensifies

SoftBank's investment in Rapidus arrives at a moment of unprecedented competition in AI semiconductor manufacturing. The landscape is shifting rapidly:

TSMC remains the undisputed leader, with its Arizona fab expected to begin producing 4nm chips for Apple and other customers in 2025. The company's 2nm production in Taiwan is on track for 2025, giving it a significant head start over Rapidus.

Samsung has been aggressively pushing its foundry business, recently securing orders from major AI chip designers. The Korean giant is investing $230 billion over 20 years in semiconductor capacity.

Intel is attempting its own comeback through the Intel Foundry services division, though the company has faced repeated delays and recently announced significant restructuring.

Compared to these established players, Rapidus faces enormous challenges. Building a leading-edge fab from scratch typically costs $20 billion or more and requires thousands of specialized engineers. The $5 billion from SoftBank, combined with government subsidies, brings Rapidus closer to that threshold — but execution risk remains high.

The startup's 2027 target for mass production would place it roughly 2 years behind TSMC's 2nm timeline. Whether customers will choose Rapidus over proven manufacturers remains an open question.

What This Means for the AI Industry

The SoftBank-Rapidus deal has implications that extend far beyond Japan. For the broader AI ecosystem, several consequences are worth watching:

Supply chain diversification becomes more realistic. If Rapidus succeeds, AI companies will have another option for manufacturing advanced chips, reducing the industry's dangerous concentration on a single Taiwanese manufacturer.

Chip costs could decrease as competition in advanced manufacturing intensifies. More fabs competing for customers should drive pricing pressure, potentially lowering the cost of AI training and inference hardware.

Vertical integration in AI accelerates. SoftBank's strategy — combining Arm's chip designs, Rapidus's manufacturing, and its own AI data center buildout — mirrors the approach pioneered by companies like Apple and increasingly adopted by Google, Amazon, and Microsoft, all of whom now design custom AI chips.

For developers and businesses building AI applications, more competition in the chip manufacturing space ultimately means better access to computing resources. The current GPU shortage that has constrained AI development could ease as new manufacturing capacity comes online in the late 2020s.

Looking Ahead: Can Rapidus Deliver?

The critical question surrounding SoftBank's $5 billion bet is whether Rapidus can actually execute on its ambitious plans. Several milestones will determine success or failure in the coming years.

By late 2025, Rapidus needs to have its Chitose fab construction substantially complete and begin installing ASML extreme ultraviolet (EUV) lithography equipment — the $200 million machines essential for producing 2nm chips. Equipment availability alone could be a bottleneck, as ASML has limited production capacity and a long order backlog.

By 2026, the company must demonstrate working 2nm test chips with acceptable yields. This is arguably the most difficult milestone, as yield rates — the percentage of functional chips produced — typically start very low and take years to optimize.

By 2027, Rapidus aims to begin volume production for paying customers. Securing those first customers will be crucial. SoftBank itself could serve as an anchor customer for AI processors, providing Rapidus with guaranteed demand.

The investment carries significant risk, but the potential reward is equally substantial. If Rapidus succeeds, it would represent one of the most remarkable industrial achievements of the decade — and give SoftBank a competitive advantage in the AI hardware race that few companies could match.

For now, the $5 billion bet reflects a broader truth about the AI industry in 2025: the battle for AI supremacy is increasingly being fought not in software labs, but in semiconductor fabrication plants. And SoftBank is determined to have a seat at that table.