The movement to release AI technology free of charge is accelerating, led primarily by major U.S. tech companies. Behind this lies a strategy to establish standards favorable to one’s own company early on and design the competitive playing field itself. Japanese companies tend to participate only after completing technical validation, highlighting the risk of falling behind in standardization discussions.
Major technologies of the past—the internet communication protocol “TCP/IP,” the smartphone OS “Android,” and the cloud management technology “Kubernetes”—achieved explosive adoption through “openness,” the practice of publishing specifications and design blueprints. That wave is now sweeping into the AI domain. OpenAI has released open AI models such as “gpt-oss,” Google has released “Gemma,” and Anthropic donated the “MCP (Model Context Protocol)” standard—which connects AI to external systems—to a nonprofit organization.
At first glance, free release appears to be an act of giving away proprietary technology. But an official at the “Agentic AI Foundation,” a nonprofit aiming to standardize AI agent technology, points out that there is a shrewd corporate strategy at work.
The Agentic AI Foundation was established by the Linux Foundation in December 2025. Its founding members are OpenAI and Anthropic, with approximately 250 companies and organizations participating, including Google, Microsoft, and Hitachi. According to the foundation’s research, the proportion of enterprises using open-source AI surged from 65% in 2024 to 89% in 2025. Generative AI-related projects published on the source code sharing platform “GitHub” also increased by 98% from 2023 to 2024.
The speed of MCP adoption is particularly striking. The software development kit (SDK) for implementing MCP surpassed 120 million cumulative downloads in roughly 20 months after its release in November 2024. Compared to Kubernetes, the cloud management software, which took three years to reach a similar scale, the speed of MCP’s spread stands out.
Manik Sultani, co-founder and CTO of the Agentic AI Foundation, explains that the trend toward openness has accelerated further since the emergence of AI agents. This is because AI agents cannot deliver value without integration with external systems such as business systems and the internet.
“Standalone AI agents are already running in production environments. The next wave will be the ‘Internet of Agents,'” he said at a press briefing. His view is that as system integration across vendor boundaries advances, open standards and open source will hold the key to system expansion.
Why Voluntarily Give Up Competitive Advantage?
As openness progresses, the technology becomes commoditized and less likely to serve as a source of competitive advantage. So why do companies still release their proprietary technology? Noriaki Fukuyasu, Vice President for Japan at the Agentic AI Foundation, explains that it becomes easier to understand when product development is viewed through a “three-layer structure.”
The bottom layer is “Shared Cost.” This refers to elements that anyone can procure, like bolts and nuts, where development costs are shared. The top layer is “Value”—the unique value that constitutes a company’s competitive advantage. In between lies the “Cost” layer, which includes the base software and servers needed to deliver Value.
According to Fukuyasu, the idea of shifting this Cost layer into the Shared Cost layer is the fundamental driver behind the push for openness. Taking AI agents as an example, companies face challenges such as “different data structures” and “mismatched security requirements,” and must pay costs to resolve them. If these elements are standardized, companies can concentrate management resources on the Value layer.
“Rather than adapting to something that has already been established as a global standard, you create a playing field that is favorable to you. Then other companies will have to build their businesses within that field,” Fukuyasu said. Investment in openness is a cost increase in the short term, but if one’s own technology is adopted as the standard, the competitive environment itself can be designed to one’s advantage.
Standardization also leads to strengthening the Value layer. For example, if a company whose strength lies in high-quality data opens up areas other than data—AI models, management tools, integration systems, etc.—it can concentrate the competitive focus on data.
The “Participation Timing” Problem for Japanese Companies
Regarding Japanese companies’ efforts, Mazin Gilbert, Executive Director of the Agentic AI Foundation, acknowledges their participation and cooperation but points out issues with the timing of their involvement in standardization discussions.
U.S. companies tend to participate in discussions from the early stages of rule formation, even when production deployment or commercialization is still far off. In contrast, Japanese companies “relatively often participate only after first evaluating and adopting the technology and confirming its effectiveness and practicality,” he analyzes.
“Both approaches have their rationale. However, when it comes to standardization activities, once a standard is set, it has a significant impact on the subsequent ecosystem, so there is great value in participating and joining discussions from an early stage,” Gilbert emphasized.
That said, there are signs of change. According to Gilbert, the timing at which Japanese companies join standardization discussions is getting earlier, with Hitachi and others already participating in the Agentic AI Foundation, and further increases are expected. “The trend of Japanese companies increasing their presence in global open technology communities—not just as users but as contributors involved in standard-setting—will accelerate further going forward,” he projected.
Enterprise System Integration Expanding Around MCP
MCP, which connects AI to external systems, has become a symbolic presence in the openness movement. Just as smartphones and peripherals are connected via USB-C connectors, MCP is a standard that enables different AIs and business systems to exchange information in a defined manner. The structure involves the party providing enterprise data or business functions building an “MCP server,” while AI agents retrieve information and execute actions through an “MCP client.”
After Anthropic released MCP in November 2024, OpenAI, Google, and Microsoft added support in their own products, transforming it into an industry-wide common standard not limited to any specific model. In December 2025, it was donated to the Agentic AI Foundation under the Linux Foundation, and development now proceeds in a form detached from any single company’s control.
Adoption in major AI and development tools is also progressing. ChatGPT, Gemini, Microsoft Copilot, Cursor, and Visual Studio Code already support MCP. Additionally, “Agent2Agent (A2A),” an interoperability standard between AI agents released by Google, is also being developed under the leadership of the Linux Foundation. While MCP plays the role of connecting AI to data and tools, A2A is a standard that enables collaboration between AI agents themselves; the two are complementary rather than competitive.
The benefit of MCP for enterprises is the ability to reduce dependence on specific AI products. If internal databases and business systems are connected via MCP, it becomes easier to swap or use multiple AI models and agents in combination depending on the purpose. The burden of developing connection functions each time is also reduced.
However, MCP support does not by itself guarantee secure connections. AI agents are capable not only of viewing documents but also of modifying data and executing operations in external systems, so operational design is separately required—such as which MCP servers to allow, how much authority to grant to users and agents, and whether to require human approval for critical tasks. The official MCP specification also recommends explicit user consent and the right to refuse sensitive operations, verification of server reliability, and management of access permissions and execution logs.
Responses from related companies are also expanding. GitLab is beta-providing an MCP server that allows AI tools to access project information, issues, and development and deployment tasks. Megazone Cloud has implemented MCP server registration and management functions in its enterprise AI platform “AIR Studio V2,” focusing on centralized management of server deployment, versions, and per-user permissions. Oracle is also expanding MCP support in its AI agent studio and autonomous AI database, enabling the use of external systems and approved functions of Oracle databases without developing individual connection programs.
The competitive axis for enterprise AI platforms is shifting from whether MCP is supported to “validation of connection targets” and “management of permissions, versions, and execution history.” While open standards provide a common connection pathway, who operates that pathway and how securely is expected to become the differentiating factor going forward.
The openness of AI technology is not merely goodwill or idealism—it is a strategic action to design the competitive environment to one’s own advantage. Once a standard is established, latecomers have no choice but to build their businesses within that framework. Whether Japanese companies can shift their position from “users” to “standard-setters” will determine their competitiveness in the AI agent era.