Sometime in the next planning cycle, someone will ask you a question with no comfortable answer. Could you justify a double-digit increase in AI infrastructure spending to your board, and would your sovereignty audit back you up?
For a growing number of enterprises, the question is already live. IDC FutureScape 2027, IDC’s annual set of top technology predictions, places three decisions in the closing window, where the big decisions are nearly made and an opportunity is about to become a disadvantage. One of them: IDC predicts that by 2027, 45% of G2000 organizations will justify exactly that kind of increase.
Missed the webinar? Watch all ten worldwide predictions in the webinar replay.
This post extends our launch blog, IDC FutureScape 2027: The Great Rewrite Comes With Deadlines, which introduced the closing, open, and emerging windows. IDC’s Rick Villars, who presented the webinar, said the main task in the closing window is rethinking resilience, at the business level as well as the technology level. Each decision gets a section below, with clips from the live webinar. The open and emerging windows get their own posts next.
The bill: how is techflation changing AI infrastructure budgets?
Villars told webinar viewers that IDC research found over 62% of organizations saw double-digit price increases on their latest purchases in the prior three months. Software and services are starting to follow networking, storage, and processors. IDC calls this techflation: the sustained rise in technology costs driven by demand for processors, memory, and datacenter power and cooling that supply cannot match.
Could you justify a double-digit CAPEX increase to your board this year, and would your sovereignty audit support it?
Organizations embedding AI into core processes will deploy more infrastructure in their own facilities, Villars said, and they are going to need to “get back into the CAPEX business.” IDC’s advice to CFOs: fund AI factories like warehouses, not incremental budgets, with multiyear committed spending plans before December 2027.
Start with an audit of security and data residency needs, within six months, so the commitment is sized to your requirements, and put legal, compliance, IT, and finance behind one sovereignty case. The multiyear hardware agreement follows, with price protection and with review and exit terms written in before you sign. CAPEX procurement cycles run longer than cloud activation, so start early, then recalibrate as costs and regulation move.
CAPEX is one answer to techflation, and sourcing is another. IDC predicts that by 2028, 50% of G2000 enterprises will adopt GPU-as-a-service over self-built infrastructure. Waiting carries a cost as well: IDC’s analysis of the macro drivers finds that delayed investment gets more expensive as costs rise.
Owners: the CIO and CFO lead, with the CTO sizing the architecture. A useful measure is IT spend as a share of revenue against your peers.
The exit: why are open models part of the cost answer?
Hardware is only half of the cost story; the other half is the model you run on it. By 2028, IDC expects enterprises to shift 60% of AI workloads from proprietary LLMs toward open-source models, using model routing (sending each workload to the model that best fits its needs) to reduce costs and hedge against vendor lock-in. This 60% counts workloads, while the AI-first figure above counts enterprises. IDC’s guidance projects AI operating costs falling 30% to 50% by 2028. Treat that as a projection that depends on routing working well, and net it against the skills and support costs the same guidance names.
Our read: open models are a negotiating position before they are a cost line.
A workload you can redeploy elsewhere changes your next renewal conversation. The harder part is the operating model. The guidance calls for routing that matches workloads on latency, accuracy, and cost, containerized multi-model capacity, and governance extended to open-source vulnerabilities and data residency. Procurement has to learn to evaluate open communities and support tiers, and IDC’s sovereignty research says model-serving and orchestration frameworks such as vLLM, Ray, and Kubernetes-based tooling need to be budgeted and staffed as a named 2027 item.
Owners: the CTO leads the routing architecture, with the CIO owning vendor terms and a named owner for agent knowledge standards. A useful measure is cost-to-value per workload, before and after a workload moves.
The factory: what does it take to treat the software factory as critical infrastructure?
The third decision starts with an image from IDC’s Katie Norton, who leads research on DevSecOps and software supply chain security. Picture a production line. “Every application a company ships inherits the trust of that factory,” she told webinar viewers, so a contaminated line makes everything it produced suspect and can force something like a recall.
In this prediction, the software factory means the pipelines, tools, plugins, and agents that create, build, and deliver code. IDC predicts that by 2028, 50% of large enterprises will treat those systems as critical infrastructure. AI changes the picture in two ways. It produces more code than human reviewers can absorb, and agents now take part in the factory themselves, opening pull requests, calling tools, and holding identities with access to build systems. Norton describes two ways a software factory can fail: it stops, which blocks fixes from shipping, or it is compromised, which means its output cannot be trusted. She sees the most urgency in connected products such as vehicles and medical devices, where an attacker could “ship a malicious update signed with the manufacturer’s own keys and every device will accept that as authentic.”
Could you still ship a fix tomorrow if your build system were compromised or offline?
Norton’s starting advice for 2027 budgets has three steps, and the first sounds simple: “You can’t secure what you can’t see.” Inventory every pipeline, tool, plugin, and agent with access. Most organizations she talks to cannot produce that map today, so name an owner and a deadline. Add the factory to business continuity planning using the question above, and put third-party delivery tools under supplier risk management. On the budget side, she suggests consolidating delivery environments onto fewer, centrally managed platforms. Our read: consolidation concentrates risk in fewer, larger platforms, which is why the inventory comes first.
Owners: the CISO and CTO lead, with the CIO sponsoring the budget. A useful measure is the time it takes to ship a fix with the primary build system offline.
Before the window shuts: what should tech leaders do first?
Each decision has a first move that fits inside one planning cycle, and a gate that tells you when to take the next step.
- AI infrastructure budget: audit security and data residency requirements and map sovereignty rules, then size a multiyear commitment from the result, with price protection plus review and exit terms. Recalibrate as costs and regulation move.
- Open models: pilot routing on one workload class picked from an audit of migration candidates, and measure latency, accuracy, and cost before you fund the platform and skills at scale. Name an owner for agent knowledge standards now.
- Software factory: inventory everything with build access and answer the ship-a-fix question before you decide on consolidation.
Villars closed the webinar with three priorities for the 2027 planning cycle: governance for agents and data, flexibility that makes cost more predictable, and leadership that can trust AI. The closing window decisions sit under the second and touch the first.
Want more FutureScape? Visit our dedicated website for FutureScape predictions, past and present.
What signals should you watch as the window closes?
Forecasts move as conditions move, so keep a short watch list. Start with memory pricing: IDC’s dedicated private cloud research expects memory prices to start easing in the second half of 2028. If supply loosens sooner or quotes soften at your next two purchases, revisit the terms of a price-protected contract, which is why review and exit terms belong in it. For open models, watch whether support tiers and communities mature, since routing savings grow with them. For the software factory, watch your inventory: consolidate delivery platforms once everything with build access is mapped.
Your IDC contacts are the fastest way to stay ahead of these signals. Ask your IDC account team to connect you with the analysts behind these predictions, and check in each quarter on what is shifting in pricing, supply, and adoption, so you can tell a blip from a change that should move your plan.
IDC Quanta lets you see the research behind a prediction alongside the follow-up questions it raises.
