A co-investment program, not a newly pooled “joint fund”: KDDI and Google have not announced one new pot of capital. Their existing KDDI Open Innovation Fund and Google AI Futures Fund will invest together, deal by deal. The headline ceiling is $2 million per company, but valuation, ownership and other terms will be negotiated individually. A selected startup must issue equity and use Google AI materially in its business.

The night does not always end for an AI startup when the code finally works. It may end when the cloud bill passes the monthly budget. It may end when an experiment stops for want of an available GPU. Or when the model at the center of the product reaches a rival months earlier. In 2026, ideas are not the only scarce input in artificial intelligence. So are compute, current models, access to customers and the time to keep trying.

On July 28, KDDI and Google AI Futures Fund opened applications for a new AI Startup Support Program in Japan. According to the companies’ official program pages, selected businesses may receive up to $2 million in co-investment, as much as $350,000 in credits for Google AI tools and Google Cloud, early access to Google DeepMind models, use of KDDI’s advanced GPU infrastructure in Osaka–Sakai, and technical and business support.

At this edition’s reference rate of ¥163.69 to the dollar, $2 million is approximately ¥327.38 million and $350,000 is about ¥57.29 million. Those are Japan.co.jp conversions, not contractual yen commitments. Actual investment size, valuation, ownership, credit eligibility and access terms will depend on review and individual agreements.

Up to $2 millionCo-investment per company; about ¥327.38 million
Up to $350,000Credits for Google AI tools and Google Cloud
Five fieldsWork, knowledge, software, creativity and entertainment
No deadlineRolling applications; decisions begin from September

The Four Resources the Program Actually Offers

The first is capital. KDDI Open Innovation Fund and Google AI Futures Fund may invest together, up to a combined $2 million in a selected company. “Up to” does not mean every participant receives the same check. Investments will be considered on a rolling basis and shaped by the business, its fit with the program and negotiated terms.

The second is time—a lead measured in model access. The Japanese page cites Gemini, Nano Banana and Lyria, while Google’s materials also include Gemma. Together, those names cover general multimodal AI, imagery, music and relatively lightweight open models. The possibility of testing technology before broad release is not merely a chance to show off a new feature. It lets a founder redesign a product earlier, observe users earlier and approach launch day with a working commercial experience. Early access is a form of temporal capital.

The third is compute. Alongside Google Cloud credits, KDDI is offering trials of KDDI GPU Cloud and Gemini on Google Distributed Cloud at its Osaka–Sakai data center. Inference, large-scale search and evaluation, and adaptation using proprietary data all create recurring computing costs. A credit is not cash, but it can extend a startup’s runway by postponing part of that bill.

The fourth is people and market access. The program promises contact with researchers, engineers, product managers and designers from KDDI and Google, as well as opportunities for product development, collaboration and expansion. For a young Japanese business-to-business company, security review, procurement, legal approval and trust can be higher walls than the technology itself. KDDI—with enterprise sales, consumer touchpoints, networks and data centers—can stand on both sides of those walls.

ResourcePublished offerWhat a founder should verify
Co-investmentUp to $2 million per companyValuation, ownership, investor rights and the effect on the next financing
CloudUp to $350,000 in Google AI and Cloud creditsExpiry, eligible services, post-credit pricing and effect on gross margin
Advanced modelsEarly access to Gemini, Gemma, Nano Banana, Lyria and othersPreview changes, API continuity, price and data-use terms
Domestic AI infrastructureTrials of GPU Cloud and Gemini on GDC in Osaka–SakaiCapacity, service levels, production migration, data location and operating responsibility
Experts and distributionTechnical, product, design and business help, plus collaboration opportunitiesDuration, scope, intellectual property and whether customer introductions are guaranteed

This Is Not a Free Accelerator

The program is open to full-time founders and startups across Japan. Its FAQ says preference will be given to companies from pre-seed through Series A with fundraising or business traction, while also saying that firms at any stage may apply if they have a clear concept and execution plan. Applications have no closing date. Selection results are due to be communicated on a rolling basis from September.

The price matters. This is an equity investment: a selected company gives up shares. Material use of Google AI in the business is also a condition. Other models may be used and the arrangement is non-exclusive. Even so, when capital, models, cloud services and technical help come from the same partners, the choice of technology stack can no longer be separated neatly from the capitalization table.

An AI investor no longer sits only on the shareholder register. It can help determine the entrance to the model, the location of the GPUs, the monthly cost base and the road to customers—all at once.

That does not make the offer too dangerous to accept. It could be an unusually strong deal. But founders should examine it with the rigor of a financing round: Will the unit economics survive when credits expire? Can the product switch models? Will a feature built on preview access remain viable after public release? Who owns the intellectual property created in a joint project? Speed is valuable only if the company can keep moving after the subsidy ends.

From a Check to a Stack: How Venture Support Changed

Venture investment was once understood mainly as money plus advice. Internet companies needed funds to buy servers, hire salespeople and advertise. As public cloud services spread in the late 2000s, servers moved from capital expenditure toward metered operating cost. Cloud credits became a standard part of accelerator programs: a seemingly simple benefit that could also introduce a young company to the provider it might keep for years.

Generative AI deepened that relationship. Many application companies do not train a foundation model from scratch. They combine external APIs with open models running on somebody else’s cloud. The investor, leading supplier, development platform and co-seller can therefore be the same large corporation. When a capital provider also influences the technical substrate, support becomes ecosystem building—and a strategy for cultivating future demand.

For Google, AI Futures Fund can both help talented founders and enlarge a developer economy around Gemini. For KDDI, the program can reveal businesses beyond telecommunications, cultivate customers for its Osaka–Sakai infrastructure and introduce new services to its enterprise and consumer networks. Deals should move fastest where the interests of the founder, Google and KDDI overlap.

KDDI’s 15-Year Journey: From Smartphone Apps to AI Infrastructure

KDDI’s startup work predates the generative-AI boom by more than a decade. It launched KDDI ∞ Labo in 2011 as smartphones and application stores were rapidly expanding. The accelerator helped young developers build services and opened some of a large company’s assets and customer relationships to outside ideas. In Japan at the time, that was an early corporate experiment in startup collaboration.

In 2012, KDDI formed the first KDDI Open Innovation Fund and moved from acceleration into direct investment. By April 2025, KDDI said the fund series managed about ¥40 billion and had invested in 150 companies. Including regional and environmental vehicles, the total had reached roughly ¥53 billion and more than 170 investments by June 30, 2026. The new program’s own page describes more than 175 portfolio companies.

In 2025, KDDI established a fifth, ¥5 billion fund focused on AI and deep technology and announced a broader ¥30 billion startup-investment plan that included commitments to overseas venture funds. The $2 million ceiling in this week’s announcement did not appear from nowhere. It sits on top of 15 years of accelerators, corporate venture capital, international partnerships and physical infrastructure.

KDDI and Google have a history, too. In 2024, KDDI offered startups generative-AI capabilities linked to Gemini through APIs under its αU on cloud service. In April 2025, KDDI and Google Cloud agreed to a strategic AI partnership and began examining the placement of Gemini on KDDI’s domestic infrastructure. The 2026 program connects those steps: APIs, domestic deployment and joint equity investment.

2011 — KDDI ∞ Labo begins as a corporate accelerator for the smartphone age.

2012 — The first KDDI Open Innovation Fund is established.

2022 — Japan adopts its five-year startup development plan.

2024 — KDDI offers startups Gemini-linked functions through αU on cloud.

April 2025 — KDDI and Google Cloud form an AI partnership; KDDI establishes Fund V.

May 2025 — Google formally introduces AI Futures Fund.

January 2026 — KDDI’s Osaka–Sakai data center starts operations.

July 28, 2026 — Applications open for the Japan support program.

From September 2026 — Selection decisions are scheduled to begin.

Why the Physical Place in Sakai Matters

The most characteristically KDDI part of the program is that its GPU offer is tied to a building in Sakai, Osaka Prefecture, rather than existing only as an abstract cloud balance. KDDI acquired the land and building of part of Sharp’s former Sakai manufacturing complex. By reusing its high-capacity power and cooling infrastructure, KDDI brought the Osaka–Sakai data center into service on January 22, 2026.

The facility has about 57,000 square meters of floor area. It supports AI systems including NVIDIA GB200 NVL72 racks with direct liquid cooling, connections of up to 100 gigabits per second, closed-network links and multicloud connectivity. KDDI says it is powered with 100 percent renewable energy. An industrial site built to manufacture display panels has been recast as machinery for training and running models.

Domestic location is not only about lower latency. Health care, manufacturing, finance and government can be sensitive to confidential information leaving Japan, foreign legal jurisdiction and third-party control. With Gemini on Google Distributed Cloud, KDDI is presenting an option for using generative AI while keeping and managing data within the country. “Sovereign AI” need not mean only a model built in Japan; it can also describe a foreign model operated inside a Japanese legal and physical boundary.

There is a tension in that phrase. Domestic facilities may improve control over data, but they do not equal complete technological independence. Google controls the model roadmap, licensing and updates. The GPUs are imported. Data location, model ownership, chip supply and system operation are separate layers, and a serious buyer must ask who controls each one.

Why Google Is Giving Away the Early Seats

Google formally unveiled AI Futures Fund in May 2025. The structure offers selected companies, at various stages of growth, early access to DeepMind models, collaboration with researchers and product teams, cloud credits and opportunities for direct equity investment. The older Google for Startups Cloud Program already advertised up to $350,000 in cloud credits for qualifying AI-first startups. AI Futures Fund adds proximity to the model builders and the possibility of Google becoming a shareholder.

As model competition accelerates, public launch is no longer the finish line; it is the beginning of distribution. A platform owner needs developers who will push its models into legal documents, comics, music, factories, schools and daily workflows—and return evidence of where the technology fails. Founders get capabilities early. Google gets applications, feedback and potential long-term cloud customers.

The five themes for the Japanese program—work, knowledge, software development, creativity and entertainment—are revealing. They point beyond the efficiency of a telecommunications network toward Japanese strengths in content, professional knowledge and workplace practice. Nothing in the published terms excludes heavier fields such as robotics or drug discovery, but the center of gravity is clear: AI-native products that can be moved into the market quickly.

Japan’s ¥10 Trillion Goal—and the Missing Pipes for Growth

Japan’s government adopted a five-year startup development plan in November 2022. Its goal is to lift annual startup investment to ¥10 trillion by fiscal 2027, more than ten times the 2021 level. Longer-term ambitions include 100,000 startups and 100 unicorns. The plan combines talent, capital, exit markets and open innovation. It does not mean the government itself promised to spend ¥10 trillion; it is a target for the wider investment market.

But adding investment volume alone will not produce globally competitive AI businesses. Japan has strong university research, industrial workplaces, manga, music, games and specialized corporate data. It has often had thinner links between those assets and managers able to productize them, overseas sales channels, large growth rounds, compute and English-language networks. In a 2026 partnership with a U.S. venture-capital firm, KDDI itself said Japan possessed excellent technologies but still faced challenges in global networks and growth acceleration.

The new program tries to join those pipes. Google brings models and a global developer ecosystem. KDDI brings domestic customers, communications networks, data centers and investment experience. Two million dollars is nowhere near enough to train a frontier foundation model. But for a pre-seed-to-Series-A company searching for global product–market fit on top of existing models, capital plus compute may provide meaningful runway.

What Might Be Selected in the Five Fields

Five lenses for likely candidates
  • Work: Auditable agents that complete a process, not merely generate a paragraph.
  • Knowledge: Search and reasoning grounded in Japanese specialist documents, company data or research.
  • Software development: Tools that compress testing, operations and security as well as code generation.
  • Creativity: Image, music and video systems that also handle rights and compensation for creators.
  • Entertainment: Experiences that translate, personalize and distribute Japanese intellectual property worldwide.

The important question in selection is unlikely to be whether a company “uses AI.” It will be whether AI sits so deeply inside the product that the business could not exist in its present form without it. The requirement for material Google AI use points away from a thin chatbot wrapper and toward companies whose commercial value changes with the capabilities of the underlying model.

At the same time, a service optimized only for Japanese language and local business habits does not automatically become globally competitive. KDDI may help with domestic adoption, but international expansion is an explicit part of the proposition. Founders will have to explain the boundary between a problem that can only be solved well in Japan and technology that can travel from Japan to the world.

Success Will Not Be Measured by the Cohort Photo

At announcement time, the companies had not disclosed a target number of participants, an annual deal count or a total program budget. Applications remain open and investments will be considered continuously. That makes the progress of individual products and revenues more important than a theatrical photograph of a “first class.”

The useful figures one year from now will go beyond applications. How many participants retain healthy gross margins when the credits expire? How many KDDI trials convert to paid contracts? How many win overseas customers? Can their products survive a model change? Does the ownership structure help or complicate the next round? Did domestic GPU access become part of production, or remain a one-time demonstration?

Google and KDDI face a test as well. Can the speed of a large company’s screening match the speed of AI development? Can security and legal standards be protected without stopping a young company in meetings? Is “early” access genuinely early? Does collaboration lead to a distribution channel, or end in a press release?

The $2 million ceiling makes the headline. The program’s real value will be measured in how many months a young company wins back from waiting—for GPUs, lawyers, customers and permission to move.

Once the Door Opens, Whose Road Will a Founder Travel?

For a Japanese founder, this is a practical and ambitious opportunity. One selection process may bring investment, up to $350,000 in computing support, globally important models, domestic data-center capacity, technical talent and access to large-company customers. That combination could materially shorten the journey from research demonstration to commercial service.

But the open door has a direction. The company must use Google AI; it may run on Google Cloud or KDDI GPUs; both partners may become shareholders. The arrangement is non-exclusive, yet technical and financial gravity is real. A strong founder will take the acceleration while protecting model portability, post-credit economics, ownership of data and the customer relationship.

When KDDI ∞ Labo gave young developers a workspace, mentors and an entrance to mobile networks in 2011, startup support centered on a place to build. Fifteen years later, that place extends across a data-center floor, a model researcher’s calendar, a cloud billing account, an investment committee and a corporate sales network.

KDDI and Google say they can connect all of it. The work of the founders selected next will not simply be to pass through the door. It will be to build a road beyond it that still carries their own company’s name.

Reporting Notes and Principal Sources

This article is based on KDDI, Google and government or public-agency materials available by July 29, 2026, 11:16 a.m. JST. Yen conversions use this edition’s reference rate of ¥163.69 to the dollar and are Japan.co.jp calculations, not contractual yen amounts. Published ceilings do not guarantee investment or usage; ownership, credits and other terms depend on screening and individual agreements. We found no confirmed participant count or total program budget in the published materials.