Guest: Arthur Tyde (LinkedIn)
Company: CIQ
Show: 2026 Predictions
Topic: AI Infrastructure

The race to train AI models is creating a blind spot that could cost enterprises everything. As organizations rush to consolidate intellectual property for model training, many are overlooking a foundational risk: the security posture of the infrastructure running it all. Arthur Tyde, Senior Vice President of Global Business Development at CIQ, argues that 2026 will be the year the industry is forced to reckon with this gap.

CIQ traces its roots to high-performance computing, but Tyde is clear that the company’s focus has always been on where the industry is heading. “We believe that in the future, all workloads are going to become performance workloads,” he says. With AI now dominating enterprise infrastructure conversations, CIQ sees its HPC expertise as a direct competitive advantage — one it is applying through two core offerings: Rocky Linux and Fuzzball.

The security argument Tyde makes is visceral and practical. He describes walking into customer meetings where phones are taped over to prevent IP leakage from whiteboards — yet those same organizations plan to train their LLMs on Debian derivatives with little to no hardening. “They’re super diligent about protecting their IP, but they’re going to pull all of their intellectual property into a machine and train their models on it. I think that’s insane.”

His point lands hard: a breach during model training doesn’t just cost you data. It costs you your competitive edge. The model itself becomes the liability. CIQ has responded by segmenting its OS offerings — RLC-H, a hardened variant for secure model training, and RLC-AI, optimized for performance once a model is built. The message is deliberate: security and performance are not interchangeable, and they do not belong in the same phase of development.

Beyond security, Tyde identifies the convergence of traditional HPC and AI infrastructure as a defining shift for 2026. He points to a large automotive customer running ANSYS, Siemens Star-CCM, and LS-DYNA simulations for airflow and heat distribution analysis. When AI came into scope, IT leaders peered into that infrastructure and realized what HPC practitioners had known for years — it looks a lot like AI. Budgets tripled. Complexity multiplied. The gap between what organizations have and what they need became impossible to ignore.

“Now they have to not just build out this incredible infrastructure, but figure out new and better ways to run it,” Tyde says.

This is precisely where CIQ’s Fuzzball platform enters. Fuzzball containerizes workloads, inventories both on-premise and cloud resources, and intelligently allocates compute based on availability and workload requirements. It abstracts away the complexity of SSH access, command-line job submission, and resource hunting — letting users submit workflows the way they would interact with a consumer AI tool. “The end user doesn’t have to worry about SSH into machines or doing command-line stuff. They can just submit their jobs and workflows, like you would use Claude or ChatGPT or something like that.”

The compliance dimension compounds all of this. Tyde flags the “compliance velocity gap” as a major bottleneck for 2026: security standards are evolving faster than most organizations can adapt. Getting a single FIPS certification can take a full year. For enterprises operating in regulated sectors — defense, finance, automotive — this is not a theoretical problem. It is an operational crisis. CIQ’s pitch is that it absorbs this compliance burden on behalf of customers, providing FIPS, Common Criteria, CVE patching, and secure supply chain assurances as part of its Rocky Linux stack.

Tyde has watched the open source enterprise Linux space for three decades. He wrote early checks to found the OSDL, the Free Standards Group, and served as the first CTO of the Linux Foundation. When Red Hat ended free enterprise Linux, he saw the gap not just as a market opportunity but as a structural problem for the ecosystem. “Somebody needed to step in with an Enterprise Linux that filled not just the community angle, but also the compliance pieces, the security pieces, the secure supply chain.”

For 2026, his predictions converge on a single theme: the tools and assumptions that carried organizations through the HPC era are not sufficient for the AI era. Security must be designed in at the infrastructure layer. Workload orchestration must become intelligent and accessible. And the operating system, long treated as a commodity, must once again be recognized as a strategic asset.

“We see that in every one of our customers,” Tyde says. The realization is coming. The question is whether enterprises get ahead of it or get caught.