Every era has an infrastructure gap. This one is yours to close.

AI is the biggest innovation to happen in technology since the internet. I have written about what that means for infrastructure in two earlier posts: The AI engineer era has an infrastructure problem and The infrastructure AI actually requires. I co-founded CentOS in 2004, built Singularity at Lawrence Berkeley National Laboratory in 2015, launched Rocky Linux in 2021, and founded CIQ to build the commercial product layer the enterprise market requires on top of it. Each time, a need appeared, the tools to meet it didn't exist and the organizations that recognized it early built on the right foundation while others waited. That is where we are now.

CIQ was built for this moment, and the ones after it

My purpose across 25 years hasn't changed: help other people do amazing things. The projects I have built have helped researchers cure diseases, find better energy sources, launch rockets, and drive satellites. The tools were available, people ran with them, and the foundation held. CIQ exists to keep that foundation solid through the AI era.

In 2004, I founded CentOS because the Linux ecosystem needed a stable, free, community-supported Enterprise Linux distribution. The project worked. When Red Hat ended it, the community needed an answer. I launched Rocky Linux in 2021 to provide one. My approach has been to identify where the community is exposed, build the tool that closes the gap, build it in the open, and build it to last.

CIQ supports Rocky Linux commercially and builds the product layer the enterprise market requires on top of it. The portfolio is the complete infrastructure stack for the AI era, built from the kernel up by the people who invented the tools the industry runs on.

Rocky Linux is already running at millions of actively deployed instances worldwide, across 90 percent of the Fortune 100. When three back-to-back critical kernel CVEs hit in May 2026, CIQ shipped patched kernels to customers nearly a week before other distributions responded, and immediately contributed those fixes to the Rocky Linux community. Protecting the community has always mattered more than commercial advantage. That operating principle does not change when the pressure is highest.

Measurable GPU waste and rising compute costs

The numbers this year are specific enough to put in a board presentation.

Enterprise GPU fleets run at roughly 5% utilization, according to Cast AI's 2026 State of Kubernetes Optimization Report, which measured this across 23,000 production clusters. Meanwhile, AWS raised reserved H200 capacity pricing by roughly 15 percent, breaking a 20-year pattern of falling compute costs. The organizations that will lead are not the ones buying more GPUs. They are the ones extracting more value from the ones they have.

A 2026 survey of 225 enterprise leaders by Kiteworks found that 100 percent of organizations have AI on their 2026 roadmap, yet 63% can't enforce purpose limitations on AI agents, and 60% can't quickly terminate one that misbehaves, or isolate AI systems from broader network access. Many also lack evidence-quality audit trails for AI operations entirely. Those are infrastructure exposures, and they determine whether an AI incident remains a contained event or becomes an enforcement action.

IDC forecasts that by 2027, 80% of organizations will modernize legacy infrastructure by shifting to platforms built for AI workloads. That rebuild is already underway in the procurement decisions and architecture reviews infrastructure leaders are running this quarter.

Each layer of the stack holds production-grade stability and AI speed

Organizations need to move fast on AI innovation and run rock-solid production at the same time. Until now, almost all AI work has been development work. My entire background is about running in production. That means stability, consistency, and abiding by the principle of least astonishment, while delivering the right features to innovate at speed. Production meets innovation. CIQ's portfolio delivers on both.

The foundational infrastructure in AI has three layers, and CIQ's portfolio directly addresses each one.

The first is the OS. Linux runs beneath roughly 88% of machine learning workloads, but not all Linux distributions are the same. Research-optimized distributions prioritize flexibility over throughput, and that tradeoff creates configuration debt the moment a workload moves to production. Enterprise distributions built for stability are slow to adopt the kernel and driver updates that AI performance depends on. RLC Pro is the baseline answer: stable enough for production compliance requirements, current enough for the inference era, and maintained by the team that co-founded the upstream project it is built from.

From RLC Pro, two purpose-built variants address the requirements that security-first and performance-first environments each need. RLC Pro Hardened adds runtime kernel protection, FIPS 140-3 validated cryptography, and CAVP-certified post-quantum algorithms on top of that stable base, a proactive security layer that reduces the kernel attack surface before a CVE is published, not after. RLC Pro AI adds the complete NVIDIA stack pre-validated, AMD Instinct drivers and ROCm, and AI-specific kernel tuning, so the OS contributes to performance instead of taxing it. In benchmarks on identical hardware, RLC Pro AI delivered up to 32% faster inference throughput on vision and segmentation workloads and up to 10% faster LLM inference, without changing the GPU, the model, or the application.

The second layer is provisioning. Warewulf Pro, the enterprise version of the cluster provisioning tool I created in 2001 and still maintain, ensures every node, every boot, is in a known state. That guarantee matters whether you are running a national laboratory supercomputer or a GPU cluster serving production AI workloads.

The third layer is orchestration. Fuzzball brings the orchestration logic that HPC developed over 30 years to production AI: multi-cloud, multi-site, and workload-portable, so teams can focus on outcomes instead of the infrastructure beneath them. Ascender Pro closes the automation gap with consistent, observable, reliable IT automation on an open platform, and a direct migration path for organizations carrying Chef EOL debt or managing Ansible complexity at scale.

Each product stands alone. Together they form a full stack built from the kernel up that is built for production AI, on infrastructure the customer owns, controls, and can audit.

The decision is already in motion

The destination is sovereignty and control: AI adopted in a way that manages your controls, meets your compliance requirements, and keeps your data and workloads inside the boundaries your organization defines. That is the condition under which AI adoption becomes sustainable at enterprise scale.

I have seen this pattern resolve twice. It resolves faster and cleaner for the organizations that recognize it early.

The AI era is an infrastructure era. CIQ builds what it requires. And when the next inflection point arrives, we will be ready for it.