Sakana AI Gains Access to Sumitomo's 100,000 Customers

The announcement spells out clearly how the three divided the work. Sakana AI draws on its research and development capability in models and solutions to turn customer problems into concrete use cases, and covers everything from workflow design through technical review, evaluation and actual deployment into operations. SCSK takes on data preparation, integration with existing systems, and quality assurance that includes information security and governance, while Sumitomo Corporation supplies the operating sites and carries the work out into industry and society. The announcement describes SCSK's share as the "integration capability that bridges the gap between AI algorithms and business implementation."

What the three identified as the problem was not model performance. The announcement states that what matters to a company is not the adoption of a particular AI in itself, but choosing AI suited to the purposes of its management, business and operations and connecting it to concrete results. It goes on to say the aim is "to connect AI to the transformation of businesses and services rather than stopping at operational efficiency." The diagnosis is that adoption has already become common and the remaining problem is making it keep running inside an organization.

The work begins in four areas. The first is real projects in mission-critical fields such as finance and manufacturing, where demands for reliability and information security are strict. The second is cybersecurity, where the companies said that as part of SCSK's frontier AI effort they will study and develop services that use an orchestration AI model built by Sakana AI to support everything from diagnosing a vulnerability to fixing it. The third is joint development of enterprise AI products and services, and the fourth is expansion into social infrastructure.

Read as a question of institutions, the most interesting sentence in this partnership is the one about avoiding dependence. The three wrote that it is important not to depend excessively on any particular technology or model, and to keep evaluating and revising how AI is used as purposes and conditions change, while the company retains control over its important data and operations. It is a declaration that AI adoption will be treated as a continually renewed relationship rather than a one-time contract, and that the control sits with the adopting company rather than the supplier.

Read from the position of labor, the same sentence reads a little differently. Three of the four opening areas involve moving inside existing operational systems, and the targets are places like finance, manufacturing and social infrastructure, where people have worked for a long time. SCSK's share, the data preparation, the integration with existing systems and the quality assurance, ultimately amounts to redesigning how work is done at those sites. When a company that sells technology joins hands with a company that holds the sites, what actually gets negotiated is not model performance but who keeps making which decisions.

The companies' histories also explain the combination. Sakana AI is a research and development company founded in Tokyo in 2023 with David Ha as chief executive, Ren Ito as president and chairman, and Llion Jones as chief technology officer. It has put forward The AI Scientist and a multi-agent orchestration foundation model as its own technology. SCSK runs PROACTIVE, an AI-native next-generation ERP, along with the cloud-native infrastructure NebulaShift and the AI agent platform NebulaCore. Sumitomo Corporation said it established a digital and AI strategy in October 2025 and made it a priority initiative.

The announcement also sets out a plan not to stop here. The three said they will develop the knowledge gained from real projects in finance, from cybersecurity, and from enterprise product development into services and business models that other customers can use as they are. They said they will widen this into broader industrial areas such as manufacturing and social infrastructure, and will also consider working with other companies that hold the necessary technology or expertise depending on the project and the business theme. If the approach the three build first hardens into the industry's default form, the side that set the frame ends up setting the place of the companies that arrive later.

The announcement METAL reviewed does not state which customer or which project the three will take on first, nor what the equity arrangement or contract size looks like. Sakana AI's official account said only that the three companies' technical capability, integration capability and business base will be multiplied together to accelerate the implementation of AI in society. What can be confirmed at this point extends as far as the division of roles and the opening areas.

Moves to tie a domestic AI company to an industrial base have appeared in several places this year. METAL reported earlier that Samsung Electronics led a 3 billion euro investment in Mistral, and if that approach binds through capital, this partnership binds through routes to market and implementation staff. The same judgment underlies both: that the wall a model-building company cannot clear alone is not the technology but the operating site.

Sakana AI is in the middle of moving from a company that sells research results to a company that goes into industrial sites, and it chose a trading house and a systems integrator as the passage. Two things are worth confirming in this structure: whether a service using the orchestration model in cybersecurity actually ships as a product, and how many of those 900 companies and 100,000 customers put their names to a first project.