AI projects now start with hiring and fiber, not models

SSA’s newest AI move isn’t a model selection or a pilot announcement. It’s a buying signal. The Social Security Administration posted a request for information seeking input on an enterprise AI strategy that explicitly includes “agentic capabilities,” and responses are due Sept. 28, according to FedScoop. In procurement terms, an RFI is where requirements start to harden, and SSA’s framing is unusually operational: pipeline for use cases, talent acquisition and retention, enterprise architecture, training, and governance.

Around the same time, three separate announcements in state IT, enterprise infrastructure services, and fiber supply point to the same shift: AI programs are increasingly constrained by the capacity to deliver and operate them, not the ability to demo them. New York state is using an enterprise agreement to spread AI and hybrid cloud tooling across agencies at lower unit cost, Government Technology reported. Kyndryl and Broadcom are packaging VMware Cloud Foundation services as an AI-ready private cloud operating model, with a pledge to skill up several thousand certified staff, according to Network World. And Zayo is locking in fiber optic cable supply with Corning to keep materials from becoming the gating factor in its network expansion for AI demand, Telecompetitor reported.

SSA’s RFI makes agentic AI a governance and architecture requirement

FedScoop reported that SSA’s RFI asks for input in five areas, including a mechanism for sourcing and prioritizing use cases, ways to connect with external partners for ongoing coordination, and approaches to training and change management so staff actually adopt tools. Two parts stand out for CIOs and program leaders watching federal demand: SSA is asking respondents to describe architectural approaches and models, not vendor and product pitches, and it is explicitly calling for governance that balances speed with accountability, transparency, and public trust.

The agency already has an enterprise AI strategy, released in September 2025, but FedScoop noted the 13-page document did not mention agentic AI. SSA’s current AI inventory lists a single agentic use case, a non-high-impact IT tool in pre-deployment aimed at developer efficiency and time and cost savings, according to FedScoop. That matters because it sets the baseline. SSA is effectively acknowledging that “agentic” will likely move from a developer productivity wedge into broader workflows, and it is trying to build the controls and operating model before deployments spread across the agency.

New York’s IBM deal shows how agencies are buying AI as a shared service

Government Technology reported that the New York State Office of Information Technology Services (ITS) signed a three-year enterprise agreement with IBM, announced in August but signed in the second quarter of 2026. The state expects $6 million in savings over three years, and the agreement is intended to make IBM technology available to participating agencies for AI, cloud management, cybersecurity, and modernization. For operators, the procurement structure is the story. ITS is using a portfolio approach so agencies do not each negotiate their own tooling and terms. The portfolio spans AI-assisted productivity and workflow automation, application modernization, hybrid and multicloud management, data protection, and asset and facilities management, among other categories, according to Government Technology.

ITS said it has negotiated 14 enterprise-level agreements that it credits with $58 million in annual savings, Government Technology reported, using New York’s own announcement as the source. The practical takeaway is that AI adoption inside government is moving toward centralized contracting vehicles that can standardize pricing and controls across dozens of agencies. That model also creates a clearer target for vendors: it rewards platform breadth, licensing flexibility, and onboarding and enablement plans that work across a large, mixed workforce.

Kyndryl and Broadcom are selling delivery capacity around VMware, not a new platform

In enterprise infrastructure, Network World reported that Kyndryl and Broadcom rolled out new consulting services for VMware Cloud Foundation (VCF) positioned as an AI program, and said they will invest in skills development for several thousand certified Kyndryl consultants, architects, and delivery specialists. Network World’s reporting highlighted a consistent operator reality: many enterprises already run VMware. The harder work now is modernizing estates, tightening controls, and making environments reliable enough to host AI workloads and agentic workflows alongside existing systems.

Several analysts quoted by Network World framed the skills investment as the meaningful part. The article cited Moor Insights & Strategy’s Mike Leone describing the training push as a response to a loss of deep VMware talent at many companies. Greyhound Research’s Sanchit Vir Gogia, also quoted by Network World, argued the announcement is more “scaffolding” around an existing relationship, and that a defensible reading is selective workload placement rather than a broad shift from public cloud back to private cloud. Either way, the operational issue is the same: hybrid estates need people and processes to run AI workloads safely and consistently.

Zayo’s Corning agreement brings AI planning down to the bill of materials

Telecompetitor covered a Zayo press release about a long-term supply agreement with Corning. The deal locks in a substantial share of the fiber Zayo expects to require through the rest of the decade. Zayo framed the agreement as a way to manage execution risk. The company said AI-driven demand is reshaping where capacity must be added and the pace of those builds, and it wants build timelines set by plans, not by whether materials are available.

Telecompetitor also reported Zayo’s construction goal: the company aims to add 15,000 new route miles by 2030. The release pointed to over 8,000 miles of new long-haul fiber, built in collaboration with NVIDIA, as part of that effort. For enterprise operators, the key takeaway is not Zayo’s specific routes. It is that AI capacity planning is moving earlier in the process. Data center leases, interconnect design, and WAN refreshes can slip when physical supply is tight, and fiber is a foundational component that takes time to manufacture, ship, permit, and install.

The operator read: AI readiness is becoming an “operating model” decision

Taken together, these developments suggest a more mature buying pattern for AI in 2026. SSA is asking for governance and architecture that can support agentic tools at scale, not one-off pilots, according to FedScoop. New York is using enterprise agreements to spread tools and pricing across agencies, according to Government Technology. Kyndryl and Broadcom are pushing a skills-backed operating motion around VMware Cloud Foundation, Network World reported. Zayo is buying supply assurance for fiber as it expands routes for AI connectivity, Telecompetitor reported. The common thread is execution capacity: people who can implement and operate platforms, contract vehicles that can onboard large groups, and physical infrastructure supply that shows up on schedules.

What to put in next quarter’s AI and infrastructure planning meetings

• For AI program offices and IT governance teams: when agentic workflows are in scope, what is the approval path for tools that can take actions, and what logs and human-in-the-loop controls are mandatory before production? SSA’s RFI highlights the need to define this early (FedScoop).
• For procurement and sourcing: can AI tooling be purchased through an enterprise agreement that standardizes security, data protection, and onboarding across business units, as New York ITS did with IBM, and can savings be quantified on a three-year basis (Government Technology)?
• For infrastructure ops leaders: if VMware Cloud Foundation is part of the private cloud estate, is there a skills plan for modernization and day-2 operations, including AI governance and workload placement, and does it rely on internal staff or an external delivery bench like Kyndryl is building (Network World)?
• For network and data center planners: do WAN and long-haul connectivity plans assume fiber supply is “available,” or are lead times and supply assurances being treated as program risks, as Zayo’s Corning agreement implies they should be (Telecompetitor).