IBM AI History: Machine Learning, Watson & Enterprise Models

A century of pivots

IBM traces to 1911, when the Computing-Tabulating-Recording Company was incorporated to hold three smaller firms. Arvind Krishna became CEO in April 2020 and chairman the following January, succeeding Ginni Rometty, and pushed the company further from its hardware and legacy software roots toward hybrid cloud and consulting, a shift anchored by the 2019 acquisition of Red Hat for thirty four billion dollars. [1][2]

IBM's own AI models, the Granite family, run underneath its watsonx platform. Independent comparisons through 2026 place Granite behind the frontier labs on capability, which is consistent with how IBM now sells AI: less as a lab racing to the best model, more as the company that gets AI running inside regulated enterprises that need a vendor to be accountable to. [8]

The company that named machine learning

IBM's claim on AI history predates the field having a settled name. In July 1959, IBM researcher Arthur Samuel published a paper describing a checkers program that improved by playing itself, and the paper's title gave the field the term "machine learning," though the tidy one-line definition people attribute to Samuel does not actually appear in it. [4]

Two later demonstrations followed the same pattern: a machine beats a human at a game people used as a proxy for intelligence, and the achievement outruns any product built on it. Deep Blue beat Garry Kasparov at chess in 1997, and Watson beat Ken Jennings and Brad Rutter on Jeopardy in 2011. Watson in particular spent the following decade failing to become a durable business, a history worth keeping in mind when reading IBM's current AI pitch. [3][5]

Reselling the frontier

In August 2026, IBM put OpenAI's models inside its consulting business, wiring GPT-5.6, Codex and ChatGPT Work into IBM Consulting Advantage alongside a dedicated practice of certified consultants who deploy the tools with clients in regulated industries. [7]

Atlas interpretation: IBM did not need OpenAI's permission to build AI products; it has Granite for that. What it bought with the partnership was the ability to sell whichever model a client actually wants, backed by consultants who show up on site. What OpenAI bought was an enterprise sales force it has never built for itself. That trade, distribution for reach, is closer to IBM's actual role in this history than either of the two chess-and-trivia demonstrations above: not the company that builds the frontier model, but one of the companies that gets it into the building. [7][8]

Signing on to voluntary rules

IBM was also one of sixteen companies, alongside OpenAI, Anthropic, Google and Meta among others, that signed the Seoul frontier safety commitments in May 2024, agreeing to publish safety frameworks with risk thresholds and to hold back models that breach them. The commitments are voluntary and unenforceable, which is consistent with IBM's position in the field: present at the table where frontier labs make promises, without itself being one of the labs those promises are mainly aimed at. [6]