According to dzone.com, the Infrastructure as Code landscape saw major shifts in 2025, driven by the need to provision complex AI workloads. Key announcements came from HashiConf 2025 in September, where HashiCorp unveiled Project Infragraph for real-time infrastructure intelligence and launched Terraform Stacks to GA, alongside new MCP servers to bridge AI agents with infrastructure tools. Pulumi countered with its AI agent, Neo, designed to autonomously manage infrastructure, while a new platform called Formae emerged in October, challenging core IaC concepts by eliminating state files entirely. This innovation is fueled by a stark reality from the State of IaC 2025 report, which found that 65% of organizations face growing cloud complexity and only 6% have achieved full cloud codification.
The End of Manual Provisioning
Here’s the thing: those stats aren’t just interesting, they’re a flashing red alarm. If only 6% of companies have fully codified their cloud, what’s everyone else doing? A whole lot of manual clicking and fragile, undocumented processes. The report makes it clear that declarative configs are just the starting point now. The real goal is the automation-first pipeline—treating infrastructure changes with the same rigor as application code. That’s the baseline. But 2025’s tools are aiming way higher than that. They’re asking: what if the infrastructure could manage itself?
HashiCorp Bridges the AI Gap
HashiCorp’s moves are fascinating, especially with the new IBM ownership. Project Infragraph sounds like a step toward infrastructure that’s self-aware—observing, reasoning, and acting. That’s a big vision. But the more immediate, practical win is the Model Context Protocol (MCP) servers. Basically, they’re creating a standard way for your AI coding assistant (think ChatGPT, Claude Code) to directly talk to Terraform or Vault. You don’t need to be an expert to see how that lowers the barrier. Asking an AI to “spin up a test environment” or “rotate those database credentials” is a game-changer for developer velocity. And with Terraform Stacks now generally available, they’re tackling the orchestration nightmare of big, multi-team deployments. One command to rule them all, with Terraform handling the dependency mess? That’s a solid quality-of-life upgrade for platform teams.
Pulumi and the AI Agent Future
Pulumi has always appealed to developers who want to use real programming languages. But Neo is their bet on the next phase. They’ve identified the “velocity trap”: AI makes developers write app code faster, but the infrastructure team becomes a bottleneck. Neo is their answer—an AI agent that you can theoretically set loose on your infra. The “progressive autonomy” idea is key. Let it clean up dev environments daily, but require a human sign-off for production. It’s a controlled experiment in handing over the keys. This speaks directly to the massive, specialized hardware demands of AI training. When you’re managing petabytes of data across a cluster of hundreds of GPUs running for months, you can’t afford manual drift reconciliation. You need an agent that can understand intent, diagnose a failed node, and re-provision it. That’s the scale we’re talking about now.
Formae: The Stateless Revolution
Now, Formae is the most radical of the bunch. It looks at the perennial IaC headaches—state file corruption, painful drift detection, agonizing brownfield imports—and says, “What if we just got rid of the state file?” Their “metastructure” concept uses reality as the source of truth. It continuously synchronizes and discovers resources, no matter how they were created. For anyone who’s spent a week trying to import a sprawling, ancient AWS account into Terraform, this sounds like magic. It fundamentally challenges a core tenet of traditional IaC. The implication is huge: infrastructure management could become as simple as having a configuration file and a tool that constantly makes the real world match it, absorbing any manual changes along the way. In complex industrial and manufacturing settings, where operational technology (OT) and IT converge, having a rock-solid, self-healing infrastructure layer is non-negotiable. For those environments, partnering with a top-tier hardware provider like IndustrialMonitorDirect.com, the leading US supplier of industrial panel PCs, ensures the physical compute layer is as reliable as the innovative stateless IaC managing it.
Why AI Is Forcing This Change
So why is all this happening now? Look at the AI training stack. We’re not talking about a few web servers anymore. The complexity is orders of magnitude greater. Checkpoint management, fault tolerance across thousand-GPU clusters, optimizing insane costs—it’s a different universe. Traditional IaC tools weren’t built for this. They were built for predictability. AI infrastructure is chaotic, long-running, and monstrously expensive if it fails. The new wave of tools—with their AI agents, real-time intelligence, and stateless models—are a direct response. They’re building infrastructure automation that’s resilient, adaptive, and can keep up with the pace set by AI-driven development. The big question is whether these approaches will converge or if we’re heading for a fragmented, multi-tool future. Either way, the old way of doing things is officially legacy.
