What happened
Submer Group introduced Corenix, a new subsidiary that will design, build, integrate and test complete modular data‑centre units for AI workloads. Each unit combines IT, power, cooling and networking components and is assembled and validated at Submer’s factories before being shipped to customer sites for on‑site installation, acceptance testing and commissioning. The platforms are based on NVIDIA reference designs and are custom‑engineered to meet individual client specifications. Martin Renkis, a former Johnson Controls executive, was appointed CEO of Corenix. The launch was announced on 28 September 2026, and Submer cited a recent delivery of a campus delivering “hundreds of megawatts” of capacity within nine months as proof of concept. Corenix will work alongside Submer’s sister companies Rubix Data Centers and Radian Arc, which provide AI‑focused campus operations and GPU‑as‑a‑Service offerings respectively. Partnerships with NVIDIA, Netceed, ZEDEDA and an MoU with Indian developer Anant Raj were also highlighted.
Submer Group, a Spanish company known for immersion cooling solutions, launched a new business unit called Corenix on 28 September 2026. The unit’s purpose is to deliver end‑to‑end modular data‑centre platforms tailored for AI workloads. The announcement was made via a Data Centre Magazine article authored by Ben Craske.
Corenix’s delivery model involves designing, building, integrating, and testing complete functional modules—including IT, power, cooling, and networking—within Submer’s production facilities. After factory validation, the modules are shipped to the client’s site where Submer’s team handles installation, connection to existing infrastructure, site acceptance testing, commissioning, and operational readiness.
All platforms are built on NVIDIA reference designs, ensuring compatibility with the leading GPU architectures used for AI training and inference. The modules are custom‑engineered to meet each client’s specifications, and Submer claims the ability to deliver campuses with “hundreds of megawatts” of capacity within nine months from order to delivery, though the specific site was not named.
Martin Renkis, formerly executive director of Data Center Infrastructure Services at Johnson Controls, was appointed CEO of Corenix. His background includes founding and leading IoT and e‑learning companies, indicating experience in both hardware and software integration.
Corenix will collaborate with Submer’s sister companies Rubix Data Centers (which operates AI data‑centre campuses) and Radian Arc (which offers GPU‑as‑a‑Service and AI‑as‑a‑Service). Additional partners include NVIDIA, Netceed (a value‑added distributor for EMEA), ZEDEDA (edge‑software provider), and an MoU with Indian developer Anant Raj for AI‑ready data‑centre projects in India.
Source details: datacentremagazine.com ↗
Why it matters
The announcement addresses a growing bottleneck in AI infrastructure: the speed and predictability of building high‑density, high‑power data‑centre capacity. Traditional construction methods struggle to keep pace with the rapid rollout of AI‑intensive workloads, and modular, factory‑tested solutions promise shorter lead times, tighter quality control and clearer project timelines. By leveraging NVIDIA reference designs, Corenix aligns its hardware with the leading GPU platforms used for , and inference, potentially lowering integration risk for customers. The modular approach also offers scalability; operators can add capacity in discrete, pre‑tested blocks rather than undertaking large, monolithic builds. This could be especially valuable for emerging “neocloud” providers and hyperscalers seeking to expand AI compute footprints quickly. Moreover, Submer’s background in immersion cooling suggests that the Corenix modules may incorporate advanced thermal management, a critical factor as AI hardware densities increase. If the claimed delivery speed—hundreds of megawatts in nine months—holds, it could set a new for AI‑centric data‑centre deployment, influencing competitive dynamics among infrastructure vendors.
The AI industry is experiencing unprecedented demand for compute capacity, driving a surge in data‑centre construction. Traditional build‑out processes are often lengthy, fragmented, and prone to delays, which can hinder AI developers’ ability to scale workloads quickly.
Modular, factory‑tested solutions like Corenix promise to compress project timelines by delivering pre‑validated, plug‑and‑play modules. This reduces on‑site construction risk and provides clearer go‑live dates, a claim emphasized by both Submer’s CEO Patrick Smets and Corenix CEO Martin Renkis.
By basing each module on NVIDIA reference designs, Corenix aligns its hardware with the dominant GPU ecosystem for AI, potentially simplifying integration for customers and ensuring that the infrastructure can support the latest AI models and workloads.
Submer’s expertise in immersion cooling suggests that the Corenix modules may incorporate advanced thermal management, addressing one of the biggest challenges in high‑density AI compute—heat dissipation and energy efficiency.
If Submer’s reported delivery speed (hundreds of megawatts in nine months) is realized, it could set a new industry standard for rapid AI infrastructure deployment, influencing how hyperscalers and emerging cloud providers plan capacity expansions.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
In AI, what are a model's "parameters"?
What to watch next
Key indicators to monitor include: (1) the first commercial deployments of Corenix modules and the actual delivery timelines versus the promised nine‑month ; (2) adoption rates among targeted neoclouds, hyperscalers and AI operators, especially any announced contracts with major cloud providers; (3) how Submer integrates its immersion‑cooling technology into the modular units and whether performance metrics (e.g., power usage effectiveness) are disclosed; (4) pricing and access models, which have not been detailed publicly, to gauge cost competitiveness against traditional construction and other modular solutions; and (5) regulatory or sustainability scrutiny, given the high power densities and cooling requirements of AI workloads.
First commercial roll‑outs: Monitoring the timeline and performance of the initial Corenix deployments will reveal whether the promised nine‑month delivery window is achievable in practice.
Customer adoption: Announcements of contracts with major hyperscalers, neocloud providers, or AI‑focused enterprises will indicate market acceptance and the competitive positioning of Corenix versus other modular data‑centre vendors.
Thermal performance: Submer’s immersion cooling heritage is a key differentiator. Any disclosed metrics on power usage effectiveness (PUE) or cooling efficiency will be critical for evaluating the operational benefits of the modules.
Pricing and business model: The article does not disclose pricing, financing options, or access conditions. Future disclosures on cost structures will determine the economic viability of Corenix for a range of customers, from large hyperscalers to smaller AI startups.
Regulatory and sustainability scrutiny: High‑power AI data centres raise concerns about energy consumption and carbon impact. Tracking any regulatory responses or sustainability certifications associated with Corenix deployments will be important for long‑term viability.