The finances of the agreement were not disclosed, but both parties emphasize that the purpose is purely business: easier implementation of AI in corporate environmentswhere security and control over data are important.
For Anthropic, this is another element of the offensive on the corporate market after the launch of the Claude Enterprise plan and the rapid expansion of its customer base. In recent days, the company announced its largest implementation to date. Deloitte will provide Claude with over 470,000 employees around the world, while creating a Center of Excellence and a certification program for specialists for implementation in regulated sectors.
Mike Krieger, head of product at Anthropic, emphasizes that nearly 80 percent consumer use Claude comes from outside the USand the company is growing intensively in Europe and Asia. Industry sources also point to cooperation with Databricks, in which MCP becomes the glue for AI agents using the customer's data and tools. These moves are intended to build Anthropic's scale and credibility as a business AI provider.
Check also: The Bank of England warns of the growing risk of the AI bubble bursting
Claude fights for market share
IBM brings something to the puzzle that many AI startups lack: decades of relationships with IT departments and experience modernizing large, mission-critical systems. The company's own models, Granite, are already refined for many tasks related to legacy IT systemsincluding languages such as COBOL and RPG, and watsonx Code Assistant helps programmers carry out code transformations in the realities of mainframes and IBM. The Claude integration is expected to add more general conversational and planning “intelligence” to the mix, while Granite will remain a specialized workhorse for code modernization. It's a complementarity that it can work well in enterprises.
As for Claude, ChatGPT's rival, its strongest draw has become working with a lot of context and programming code. In 2025, Anthropic increased the context window in the Sonnet family fivefold to 1 million tokens, which in practice allows you to paste entire repositories (of the order of 75-110 thousand lines) or analyze dozens of long documents at once – without laboriously dividing this work into fragments.
For legal teams, R&D research and due diligence this means a shorter path from source material to conclusions, and for IT departments the ability to “read” entire application modules through the model and propose safe fixes and modernizations to the code.
The second leg is agents and integrations with the work environment. MCP, developed by Anthropic as an open standard, allows you to build Claude agents that have secure, two-way access to tools, databases and repositories and can perform tasks from A to Z, instead of just giving suggestions in the chat. In practice, enterprises use it to automate operational tasks, generate and enrich business content, create customer support systems and internal assistants for various policies. IBM's announcement that it will provide customers with an agent guide signals that this agent nature will be the axis of the partnership.
See also: China “nanoseconds” behind the US. They want to break Nvidia's dominance
AI tailored for corporations
The third pillar is security and implementation control. Claude Enterprise provides administration, SSO, roles and audits, and from summer 2025, companies can purchase Claude Code seats to cover both document analytics and the entire software development lifecycle in one solution. This is combined with IBM's corporate governance and is intended to increase the trust of risk and security departments, which for months have been expecting from the market not only the “smartness” of models, but also process maturity.
As a result, the competition with ChatGPT is entering a new phase. OpenAI has the name recognition advantage and a huge ecosystem, but Anthropic is getting better at leveraging distribution in the corporate world. It gains consulting partners, integration standards, and software vendors who know how to get past corporate firewalls. On the one hand, IBM retains its own Granite and Watsonx portfolio, and on the other hand, it provides clients with a second “hand” in the form of Claude.
If the market is indeed moving towards agents and domain data today, it is such pairs that may determine who will be able to implement AI where it really matters, and not just in pilots.
Author: Grzegorz Kubera, journalist of Business Insider Polska
