Two agentic-development vendors led with where software runs, not how well, across three announcements in sixteen days. That convergence is the story.
Both figures in IBM’s release are vendor-selected. The 44 and 46 per cent shares are a Futurum Research PROJECTION to 2030 quoted by IBM; the 68 per cent is a survey by IBM’s own Institute for Business Value. Neither measures current adoption.Nothing below adjudicates the competing on-premises claims made by Factory and IBM, and nothing below benchmarks any model. Nothing here is investment advice.
On 1 October 2026, IBM announced that its agentic software development platform, IBM Bob, can now be self-hosted. The options cover on-premises, private-cloud, sovereign-cloud, and air-gapped environments.
IBM’s release contrasts Bob explicitly with tools “that require organizations to move data to external services.” The alternative it offers is deployment “where their code, application context, and data already reside.” Enterprises can run supported models on-premises under licences they hold. They can also connect Bob to external model services in hybrid configurations. IBM is not imposing model lock-in.
That announcement was the third in sixteen days to make deployment location the headline. Factory did it on 15 September, then again on 23 September when it described its Router inside Factory Private as “the first and only on-premises model router built for software development agents.” None of the three led with a benchmark. All three led with where the software runs and who controls the data.
The key insight: When two vendors in the same category land on the same pitch across three announcements in sixteen days, it is unlikely to be imitation alone. They are responding to the same buyer signal. Deployment control has become the purchase criterion that model quality used to be.
A projection quoted by a vendor inside a product launch is not evidence of current adoption. It is the number that supports the thing being sold.
The Structural Read
Neel Sundaresan, GM of AI and Automation at IBM, states the principle plainly: “The future of enterprise AI will depend on security, governance and sovereignty.”
That is a procurement argument, not a capability one. It only works as a pitch if the buyer already believes that the available models are close enough in quality that residency and governance decide the purchase. That is what the marketing implies. This publication benchmarked nothing, and neither vendor published a comparison to support the implication.
The shift fits the Permission Layer framework. When regulators and procurement teams set the terms — data residency rules, sovereign-cloud mandates, air-gap requirements for defence and financial clients — the layer of governance above the model becomes the competitive surface. Vendors that can satisfy it win the deal before a capability comparison even starts.
Neel Sundaresan · GM, AI and Automation, IBM
“The future of enterprise AI will depend on security, governance and sovereignty.”
The Numbers, Labelled
IBM’s release carries two figures. Both are worth using. Neither measures current adoption.
The first is a Futurum Research projection to 2030: hybrid and edge deployments capturing forty-four per cent of the AI infrastructure market; public cloud declining to forty-six per cent over the same period. Those two figures add to ninety. Those two shares sum to 90 per cent, and the release does not account for the remaining ten points.
The second figure is IBM’s own. Its Institute for Business Value, in a June 2026 report titled The Calculus of AI Sovereignty, found that sixty-eight per cent of surveyed executives consider meeting data residency and sovereignty requirements across geographies challenging. One number is a third-party forecast quoted by the vendor inside a launch. The other is the vendor’s own research institute surveying executives about a difficulty its new product addresses.
That does not make the trend false. It means the public evidence for it is chosen by the people selling into it.
What Is Concrete
Self-hosted IBM Bob covers on-premises, private cloud, sovereign cloud, and air-gapped environments. IBM Bob is positioned beyond code generation, at software delivery and modernisation. Models run on-premises under customer licences or externally through hybrid configurations. IBM is not asking for model lock-in.
On the competitive overlap between Factory and IBM: both claims simply exist, and this publication does not adjudicate them. Factory described its Router as “the first and only on-premises model router built for software development agents” on 23 September. IBM has now shipped self-hosted agentic development with model choice on 1 October.
They are not the same product. A model router that selects between providers per task is a different object from an agentic development platform that can run entirely inside a customer environment. Readers should hold both statements as made rather than treat one as refuting the other.
What is absent from IBM’s release: pricing, a general availability date, a list of supported models, any named customer, any adoption figure, and any benchmark result.
Three Implications
IMPLICATION 1 · The Pitch Has Moved Upstream
When vendors lead on residency and governance rather than benchmarks, they are competing at the procurement layer — security reviews, compliance checklists, sovereignty mandates — before a capability comparison begins. Winning that argument blocks rivals from the conversation entirely.
IMPLICATION 2 · IBM’s Structural Advantage Is Distribution
IBM already has long-cycle enterprise relationships in financial services, government, and healthcare — precisely the verticals where air-gap and data-residency requirements are strictest. Bob’s self-hosted option is not just a product feature; it is a reason for an existing IBM account to expand rather than evaluate a startup.
IMPLICATION 3 · The Evidence Gap Is a Risk for the Whole Category
Two vendors are converging on the same demand signal with vendor-selected data behind the pitch. If enterprise adoption of self-hosted agentic development turns out to be slower than the Futurum projection implies, the category narrative could overshoot actual procurement behaviour. The sixty-eight per cent figure measures a pain point, not a purchasing decision.
Business Engineer Framework
The Permission Layer
The Permission Layer maps how regulation, procurement policy, and governance requirements sit above the model layer and determine which vendors are allowed to compete. When that layer tightens — data residency laws, sovereign-cloud mandates, air-gap requirements — it becomes the primary competitive surface. The Map of AI shows where each of the nine stack layers is thickening. Deployment control is the current frontier.
IBM Bob’s self-hosted launch is a product announcement. The fact that two agentic-development vendors made the same strategic choice — lead on deployment control, not model quality — across three announcements in sixteen days is a market signal. The differentiator in enterprise AI software is moving from what the model can do to where the model is allowed to run. Whoever owns the governance argument owns the deal.
Sources: IBM Newsroom, 1 October 2026 · Factory.ai, 15 September 2026 · Factory.ai, 23 September 2026 · IBM Institute for Business Value, The Calculus of AI Sovereignty, June 2026 · Futurum Research, cited in IBM press release. Nothing in this article is investment advice.
91,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
The IBM details above come from IBM’s own newsroom release of 1 October 2026, read directly. The Factory statements come from Factory’s own posts of 15 and 23 September 2026, read directly for earlier coverage. Nothing has been independently verified and nothing here has been benchmarked.Both figures in IBM’s release need their provenance carried with them. The 44 per cent hybrid-and-edge and 46 per cent public-cloud shares are a Futurum Research projection to 2030, quoted by IBM inside a product announcement.
The 68 per cent executive figure is from a report by IBM’s own Institute for Business Value. Neither measures how many organisations deploy this way today, and the two projected shares sum to 90 per cent with the remaining ten points unexplained in the release.Nothing above adjudicates between Factory’s description of its Router as the first and only on-premises model router for software development agents and IBM’s launch of self-hosted agentic development.
They describe different products and both claims are reported as made.The suggestion that the competitive argument has shifted from model quality to deployment control is a reading of how these companies market themselves. It is not a finding about modelperformance; no model was tested for this piece.Not established and therefore absent from the IBM release: pricing, an availability date, which models are supported, any named customer, any adoption figure and any benchmark result. IBM is a listed company. Nothing above predicts anything, and nothing here is investment advice.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
