SoftBank and OpenAI's USD 30B Asia AI Gamble

The biggest bet in Asian AI infrastructure just got a name and a location

SoftBank and OpenAI have announced a USD 30 billion joint venture to construct what will become the largest AI data centre campus in Asia. The site: Hokkaido, Japan's northernmost main island, chosen for its cold climate, abundant open land, and proximity to renewable energy sources. The ambition is explicitly imperial in scale, and neither party is being subtle about it.

SoftBank CEO Masayoshi Son has described the project as the "Manhattan Project of AI infrastructure". That framing tells you everything about how the partners view both the stakes and the urgency. Whether you find it inspiring or alarming may depend on how close you live to a proposed turbine or geothermal well.

"The Manhattan Project of AI infrastructure." - Masayoshi Son, CEO, SoftBank

What the SoftBank-OpenAI Joint Venture Is Actually Building

The Hokkaido campus is designed to function as a regional AI compute hub serving both East and Southeast Asia. With over 100,000 GPUs on site, this is not a facility built to serve one company's internal workloads. It is infrastructure built to sell capacity: to enterprises, to governments, and to the wave of AI startups across the region that currently have no viable path to serious compute at home.

The power strategy is central to the project's credibility. The facility will draw from wind and geothermal energy sources, both of which Hokkaido has in meaningful abundance. Japan's northern island sits on the Pacific Ring of Fire, making geothermal a practical rather than merely aspirational option. The renewable commitment is partly environmental posturing and partly hard economics. Electricity costs in Japan are among the highest in the OECD, and locking in low-cost green power over a long horizon is simply rational.

Japan's national government has moved quickly to smooth the path, offering tax incentives and fast-tracked permitting. This is consistent with Tokyo's broader strategy of positioning Japan as a trustworthy, politically stable host for sensitive AI infrastructure. It is a deliberate contrast to the regulatory uncertainty that has clouded investment in some other markets.

This project connects directly to the broader surge in Asia-Pacific enterprise AI and sovereign investment that is reshaping how regional businesses access and deploy AI capability. The Hokkaido facility is the most visible expression of that trend so far, but it is far from the only one.

The Asia-Pacific AI Infrastructure Race

The SoftBank-OpenAI campus does not exist in a vacuum. Across the region, the AI data centre race is accelerating at a pace that would have seemed implausible three years ago. The competitive landscape is now sufficiently crowded to constitute a genuine strategic contest, not merely parallel corporate decisions made in isolation.

• Malaysia: Google is developing a major data centre presence, attracted by Kuala Lumpur's pro-investment posture and the country's improving power grid.
• Indonesia: Microsoft has partnered with Tokopedia, leveraging local market access and the archipelago's enormous internet-native population.
• South Korea: Samsung and Naver are co-developing facilities that combine Korean semiconductor manufacturing expertise with domestic cloud scale.
• Japan: The SoftBank-OpenAI venture represents the most capitalised single project, but Japan also hosts multiple smaller hyperscaler builds by AWS, Google, and Microsoft independently.

Analysts project Asia-Pacific data centre investment to exceed USD 100 billion by 2028. That figure, already striking, is likely conservative given the speed at which new projects are being announced. The energy demands of these facilities are also generating their own set of challenges, as explored in our coverage of innovative approaches to the data centre energy crisis.

Asia-Pacific data centre investment is projected to exceed USD 100 billion by 2028. - Industry analyst consensus

What This Means for Asia: Strategic and Geopolitical Stakes

The concentration of AI infrastructure investment across Asia-Pacific is not purely commercial. These facilities represent compute sovereignty: the ability of nations and their companies to train, run, and iterate on frontier AI models without dependency on infrastructure controlled by foreign entities in foreign jurisdictions. Japan's decision to actively court the SoftBank-OpenAI venture is a policy choice as much as an economic one.

Vietnam has already moved to codify its AI ambitions legislatively. Vietnam's enforcement of Southeast Asia's first standalone AI law signals that regional governments are increasingly treating AI as a matter of national strategy, not merely private sector opportunity. The Hokkaido campus will be watched closely by regulators and policymakers across ASEAN as a template for what large-scale, government-facilitated AI infrastructure looks like in practice.

South Korea's involvement of Samsung is particularly significant. Samsung controls a meaningful share of global advanced memory chip production, and co-locating AI compute infrastructure with chip design and manufacturing expertise creates vertically integrated capabilities that pure hyperscalers cannot easily replicate. This is the kind of structural advantage that compounds over time and could give Seoul a durable edge in the regional AI race.

For Southeast Asian nations watching from a distance, the question is increasingly whether they can attract comparable investment or whether they will remain consumers of compute capacity built and controlled elsewhere. The gap between nations with serious AI infrastructure and those without is widening faster than most governments anticipated.

Environmental Pushback and the Electricity Problem

The Hokkaido facility will consume as much electricity as a city of 500,000 people at full operation. That number deserves to sit alone for a moment. It is not a projection of a distant, speculative future. It is the operating envelope of a facility currently in planning, in a country that has been grappling with energy security since the Fukushima disaster forced the shutdown of much of its nuclear capacity.

Communities near proposed sites in Hokkaido are already pushing back. Concerns centre on the visual and ecological impact of expanded wind infrastructure, the land use implications of a campus at this scale, and the broader question of whether local populations receive proportionate benefit from projects of this kind. These are not fringe objections. They are the same concerns that have delayed data centre builds in Ireland, the Netherlands, and Singapore.

Japan's government has chosen to fast-track rather than slow-roll the permitting process. That is a deliberate trade-off. The economic and strategic upside is judged to outweigh local friction. Whether that judgement holds as construction progresses will depend significantly on how seriously the project's operators engage with affected communities, and how genuinely the renewable energy commitment translates into measurable local benefit rather than accounting offsets.

The human dimension of AI infrastructure expansion is worth tracking alongside the capital flows. As we have explored in coverage of AI's less visible effects on how people actually work and think, the downstream consequences of rapid AI buildout extend well beyond the server halls themselves.

What Comes Next for Asian AI Infrastructure

The SoftBank-OpenAI announcement will function as a signal flare. Expect competing announcements from hyperscalers and sovereign wealth funds across the region in the months ahead. The USD 30 billion figure sets a new benchmark for what serious AI data centre investment looks like in Asia-Pacific. Projects below that threshold will increasingly be characterised as supporting infrastructure rather than primary compute hubs.

The race also has direct implications for AI access across the region. A compute hub of this scale, if operated on genuinely open terms, could meaningfully lower the barrier to frontier AI development for companies across East and Southeast Asia. If it operates primarily as a captive resource for SoftBank and OpenAI's own priorities, its regional significance will be considerably more limited. The commercial terms of third-party access have not yet been detailed, and that omission matters enormously.

For businesses thinking about how AI infrastructure shapes their own competitive position, the opportunities emerging for smaller enterprises in an AI-rich environment are increasingly tied to the availability of affordable regional compute. What gets built in Hokkaido over the next five years will have consequences far beyond Japan's northern coast.

• Third-party access terms for the Hokkaido campus remain undisclosed and will be critical to its regional impact.
• Japan's government is expected to release updated AI infrastructure guidelines in 2025 that may set precedents for other APAC markets.
• ASEAN nations without comparable projects face a growing compute gap that could limit domestic AI development capacity.
• The geothermal and wind energy strategy will be scrutinised as a model for sustainable AI infrastructure across the Pacific Rim.

Frequently Asked Questions

Why is Hokkaido chosen for Japan's largest AI data centre campus?

Hokkaido offers cold ambient temperatures that reduce cooling costs, substantial land availability, and access to geothermal and wind energy. Japan's government has also provided tax incentives and fast-tracked permitting to attract the investment, making the site commercially compelling as well as technically suited to large-scale AI infrastructure.

What is the environmental impact of the SoftBank-OpenAI Hokkaido facility?

At full operation, the facility is projected to consume electricity equivalent to a city of 500,000 people. The project commits to renewable energy from wind and geothermal sources, but local communities near the site have raised concerns about land use and ecological impact that have not yet been fully resolved.

How does the SoftBank-OpenAI joint venture compare to other AI data centre projects in Asia?

At USD 30 billion and 100,000+ GPUs, the Hokkaido campus is the largest single AI infrastructure project announced in Asia-Pacific to date. Competing projects from Google in Malaysia, Microsoft and Tokopedia in Indonesia, and Samsung and Naver in South Korea are significant but smaller in both capitalisation and stated GPU capacity.