IBM Put Claude In Its Own Developer Tools. Does That Change Your Server Plan?
On October 7, 2025 IBM and Anthropic announced a partnership to put Claude inside IBM's software, starting with a new development environment built around AI from the ground up. By the middle of 2026 that work was most visible inside IBM Bob, IBM's agentic software development platform.
This site exists to answer a hardware question, so the useful thing to do with an announcement like this is to work out whether it is one. It is not, and understanding exactly why is the most valuable thing an AS/400 shop can take from it.
What was actually announced
Claude, Anthropic's model family, is being integrated into selected IBM software products. The first of them is IBM's AI-first integrated development environment, aimed at the full software development lifecycle including modernization work. IBM published an accompanying guide on building secure enterprise AI agents using MCP, introducing something it calls the Agent Development Lifecycle.
IBM reported that more than 6,000 early adopters inside IBM were using the environment, with productivity gains averaging 45 percent.
How to read a 45 percent productivity number
It is a large figure and it deserves neither dismissal nor a spreadsheet built on it. Three things are worth holding in mind at once.
The sample is real and it is big
More than 6,000 people is not a demo. A figure drawn from that many working developers is worth taking seriously.
It is IBM measuring IBM
The users were IBM employees, the tool was IBM's, and the gains were self-reported. That is not dishonest, it is simply internal evidence rather than independent evidence.
Early adopters are not average users
People who volunteer to try a new development environment first are, by definition, the people most likely to get value from it. The average across a whole organisation is normally lower.
The sensible use of the number is as evidence that the category works, not as a figure to put in your own business case. If you want a figure for your own business case, run your own small trial and measure it.
The words that are not in the announcement
IBM's release talks about application modernization at scale, automated system upgrades, and framework migrations. Any RPG shop reading that will hear its own situation described. But the release does not name IBM Power, IBM Z, or IBM i anywhere in it.
That absence is worth noticing rather than filling in. It is easy, and common, for a proposal to quote an IBM AI announcement and imply a hardware conclusion that the announcement never drew. If somebody puts the 45 percent figure next to a server quote, those two things came from different places.
The rule this proves
There is a line running through everything on this site: the AI tool you use and the machine that runs the AI are two different decisions, with two different bills.
Two kinds of AI decision
| A software decision | A hardware decision | |
|---|---|---|
| Looks like | An AI coding assistant, a chat tool, a vendor subscription | A model serving your own data at steady production load |
| Examples | IBM Bob, Claude in an IBM development environment, watsonx services | On-chip Matrix Math Acceleration, a Spyre Accelerator card |
| Main cost | Subscription and training time | Card, expansion drawer, Linux partition, Red Hat subscription |
| Where the data goes | Often out to a service | Stays on your own floor |
| Decided by | Development leadership and procurement | Infrastructure, capacity planning, and the data owner |
Both can be right. Mixing them up is how a coding assistant turns into an unnecessary capital request.
A developer tool used by six people does not need an accelerator card. It needs a subscription and somebody to look at the governance questions. The hardware conversation starts somewhere else entirely: when a model has to run inside your building, next to data that cannot leave it, answering steadily all day.
What to actually do about it
Take the partnership as a signal about direction rather than a purchase trigger. It tells you IBM is comfortable shipping another company's model inside its own products, which is a meaningful shift for a company that also publishes its own Granite model family. Both things are now true at once, and a sensible AS/400 shop can use either.
If AI-assisted development is what you want, that decision lives with tooling and governance, and our software sibling covers the detail. If running a model on your own hardware is what you want, start with the Matrix Math Acceleration already sitting in your Power processor and find out whether the workload is real before anything gets ordered.
Figures and product details here come from IBM's October 2025 announcement and its coverage. Partnerships and product packaging change. Confirm current availability and terms with IBM before committing to either the tooling or any hardware you were told it needs.
Sources
- https://newsroom.ibm.com/2025-10-07-2025-ibm-and-anthropic-partner-to-advance-enterprise-software-development-with-proven-security-and-governance
- https://techcrunch.com/2025/10/07/anthropic-and-ibm-announce-strategic-partnership
Related Tools & Vendors
IBM Granite Models
IBM's own family of open language models, and the cargo an accelerator card is built to carry. You can download them free, run them on your own hardware, and nobody meters the tokens.
Power 11 Matrix Math Acceleration
The AI acceleration most AS/400 shops already own and have never switched on. It is inside every core of the processor, it costs nothing extra, and it is the cheapest honest answer to a first AI pilot.
IBM Spyre Accelerator for Power
A 75 watt PCIe card that adds more than 300 TOPS of AI inference to a Power 11 server. It is the only IBM AI chip an AS/400 shop can actually order as an add-on, and it comes with a short list of things it refuses to work without.