IBM stands up an OpenAI practice a year after Anthropic, and stays pointedly model-agnostic

IBM has signed a partnership with OpenAI to widen enterprise customers' access to OpenAI models and tools, and IBM Consulting will stand up a dedicated practice around OpenAI products ^[1][5]. The reason to pay attention is not the logo pairing: it lands less than a year after an analogous IBM partnership with Anthropic, and IBM frames both as part of a strategy of working with multiple model suppliers rather than only its own ^[4][6].

The mechanics are ordinary systems-integrator mechanics. Over the coming months IBM plans training and certification for tens of thousands of consultants, and says most of that headcount will come from reskilling existing employees rather than hiring ^[7][8]. Mike Healy, managing partner at IBM Consulting, told TechCrunch the curriculum covers OpenAI Codex, the OpenAI APIs, cybersecurity, and certifications tied to consulting AI solutions ^[9]. IBM will also assemble a group of "Forward Deployed Experts", trained through OpenAI's partner network, to sit with customers during implementation ^[10].

On the product side, GPT-5.6, Codex and ChatGPT Work are to be integrated into the IBM Consulting Advantage platform, aimed at work automation, code generation, data analysis and digital security ^[11][12]. The two companies also intend to build sector-specific offerings for financial services, the public sector, telecoms and retail ^[3]. Security is the thread with the most prior history: the companies launched the OpenAI Daybreak Cyber Partner Program in June, and OpenAI models are now to be integrated with IBM Autonomous Security ^[14][15]. Financial and other commercial terms were not disclosed ^[13].

Read the channel, not the announcement. OpenAI gains another route to large accounts through IBM's global consulting arm ^[2], which is the same thing Anthropic gained a year ago ^[4]. IBM, meanwhile, keeps its own Granite family in the mix and continues to serve third-party models through watsonx ^[16], which means its consulting bench can now be pointed at at least three named model families depending on what a client will sign ^[1]. That is not neutrality as a virtue; it is neutrality as a hedge. The integrator that certifies against every frontier vendor is insulated from picking the wrong one, and it collects the same billable hours either way.

The number worth interrogating is "tens of thousands", because IBM says the bulk will be retrained staff ^[7][8]. Reskilling is cheap to announce and hard to audit. A certification badge on a consultant who spent last year on a different vendor's stack is not the same asset as an engineer who has shipped Codex into a bank's change-management process. The Forward Deployed Experts team is the tell that IBM knows this: a small cadre trained inside OpenAI's partner network is where the real delivery capacity sits, and it is deliberately not sized in the announcement ^[10].

What to watch. First, whether the Anthropic practice keeps its bench and its pipeline now that a second vendor practice is competing for the same certified bodies ^[4][5]. Second, whether any customer-facing pricing or revenue-share detail surfaces later, since none was disclosed ^[13]. Third, how the security integration is scoped in practice, given that Autonomous Security is the one place where model output touches production defence rather than a slide deck ^[15]. Fourth, whether "model-agnostic" survives contact with the first big deal where one vendor offers better commercial terms for exclusivity.