Red Hat used its Summit Connect event in Lima, held on October 13, 2026, to argue that the ability to move applications and models between clouds, private data centers, and providers will be decisive for adopting artificial intelligence without losing control over data. The message came from Gilson Magalhaes, Red Hat’s vice president and general manager for Latin America, during a day that also covered the company’s AI portfolio, its relationship with IBM, and its response to open-source vulnerabilities, according to CTOPerú’s coverage.
The keys to Red Hat Summit Connect Lima 2026 in 20 seconds
- Red Hat argued in Lima, on October 13, that portability between clouds guarantees sovereignty.
- Its AI platform combines Linux, OpenShift, inference with vLLM and llm-d, and agents with human approval.
- IBM presented its managed inference and virtualization services with Red Hat on IBM Cloud.
- Lightwell was Red Hat’s answer to vulnerabilities found with AI.
For Magalhaes, the question is no longer whether organizations will use AI, but how to do it without compromising security, costs, or data sovereignty. Technology departments, he explained, have shifted from a support role to a central one in strategic decisions, and that shift forces companies to review where their data lives, what exposure their contracted services create, and whether applications could be moved if regulatory, geopolitical, or security conditions change. Since no organization knows for certain what its technology model will look like a few years from now, keeping the ability to change course is, in his view, an advantage. That’s Red Hat’s commercial pitch, which presents OpenShift AI as the way to run AI projects on any infrastructure, from mainframes to multicloud environments.
From Pilots to Production
José Luis Bugarín, a solutions architect at Red Hat Peru, focused his talk on the leap few companies manage to make. “Everyone runs pilots, everyone tests agents, installs their machines, but no one reaches production, no one knows how to operate it at scale,” he said. He identified four priorities for technology leaders: AI itself, cybersecurity — at a time when attackers also use these tools — virtualization, and the need to get more value out of existing infrastructure.
Bugarín described Red Hat’s strategy in three stages: standardizing Linux with Red Hat Enterprise Linux, standardizing Kubernetes and containers with OpenShift, and now an open platform for AI. That layer brings together support for accelerators from different manufacturers, inference services like vLLM and llm-d, model management and governance, and an environment for agent-based applications. The goal, according to the speaker, is for companies to stop being mere consumers of tokens and start producing them on their own infrastructure. He also cited Automation Orchestrator, a capability tied to Ansible that turns agent recommendations into automated workflows requiring human approval, and the collaboration with NVIDIA on Red Hat AI Factory, with early support for the Blackwell and Vera Rubin architectures.
As a reference case, Bugarín mentioned BNP Paribas, which uses OpenShift AI. According to his presentation, the bank processes around 900 million input tokens a day and has generated close to $600 million in business value across hundreds of use cases. The first figure matches what Red Hat published after its 2026 Summit, where the bank presented its sovereign LLM-as-a-service architecture; the second doesn’t appear in that material and comes only from the Lima talk.
IBM Brings the Managed Cloud
The relationship with IBM, Red Hat’s parent company, took up another part of the day. Narciso Lema, head of Data and Artificial Intelligence at IBM for northern South America, pointed to cost as the main obstacle: “One of the main reasons behind deploying projects into production is related to the costs of moving a pilot from a controlled environment to a production environment,” he said. He illustrated it with an analogy: assigning a full GPU to each workload is like renting an entire floor of a building that ends up half-empty, while vLLM allocates resources according to each request, like a shared workspace. According to Lema, that approach can avoid between 60% and 80% of idle infrastructure, an estimate he did not back up with data from any customer.
IBM’s offering is Red Hat AI Inference on IBM Cloud, a managed service the company announced on May 12, 2026 and that has been available since May 22, with APIs compatible with OpenAI’s and an initial catalog of open models such as Granite, Llama, Mistral, or gpt-oss. Lema highlighted three advantages: choosing a model for each use case, controlling who accesses models and data, and being able to move workloads to other environments, including the edge. For virtualization, IBM offers OpenShift Virtualization as a managed service on bare-metal servers, where it takes on patching and platform administration. According to Lema, it can be used without a full OpenShift license, a detail that IBM’s official announcement doesn’t mention. Víctor Cornejo, Red Hat’s chief architect for Latin America, rounded out the presentation.
Lightwell Against the Flaws AI Finds
On security, the event addressed Claude Mythos, Anthropic’s model capable of finding vulnerabilities in open-source software. Cornejo presented Lightwell as the response: the service from IBM and Red Hat that applies specific fixes to the version of each library a customer already uses, without forcing an upgrade to a major new release. According to him, delivery is backed by a service-level agreement (SLA), the artifacts are signed, and every patch is contributed back to the original project. Cornejo presented it as a capability no other provider offers, a claim that is his own.
Lightwell launched commercially on July 8, 2026 with a catalog of more than 6,500 fixed and signed Java and Python dependencies, according to IBM and Red Hat, and Cloud News reported in October that the project had fixed more than 400 previously unknown vulnerabilities in Java libraries. Cornejo also stressed keeping firmware and operating systems up to date, since attacks can come through any connected device.
The event included customer case studies from the region. Antonio Pineda, executive technology manager at Banco General de Panamá, described how the bank modernized its critical infrastructure amid explosive transaction growth, and Vanessa Vigo, services manager at SONDA, explained the migration of its regional cloud to OpenShift Virtualization, which combines virtual machines and containers, with projects in Chile, Brazil, and Colombia. Magalhaes noted that Red Hat has been operating in Peru for six years. For context on the relationship between the two companies, Cloud News looked back at six years of IBM and Red Hat.
Frequently Asked Questions
When was Red Hat Summit Connect Lima 2026 held?
On October 13, 2026 at the Westin hotel in Lima, according to the event’s website.
What is Red Hat AI Inference on IBM Cloud?
It’s a managed IBM service, available since May 22, 2026, that runs AI models in production using Red Hat’s vLLM-based inference engine, without the customer managing the GPUs.
What is Lightwell?
It’s a service from IBM and Red Hat, launched in July 2026, that fixes vulnerabilities in open-source libraries on the version a customer already uses and delivers signed artifacts.
What did Red Hat say about AI sovereignty?
That being able to move applications and models between clouds, private data centers, and providers will be key to reducing risk and dependency.
Sources:
