Celestica [TSX: CLS]
Celestica: Engineering the Computing Infrastructure Behind AI at Scale

For hyperscalers, the infrastructure challenge is no longer confined to putting more computing power into a data center. The architecture must connect accelerated computing, networking, storage, power and cooling while remaining adaptable as systems evolve.

Celestica approaches that challenge with its expertise in designing engineering and manufacturing complex electronics, supply chain and hardware platform solutions. It builds critical infrastructure for data centers, artificial intelligence (AI) systems and cloud networks.

The electronics manufacturing services company’s Connectivity and Cloud Solutions (CCS) segment designs products primarily for hyperscale data centers and applications, including AI workloads and cloud computing. On the other hand, its hardware platform solutions segment extends from hardware and systems-level design into software, program management, manufacturing and selected supply chain activities. These capabilities enable greater visibility across the layers that determine how infrastructure is developed and deployed.

“Strong operational execution across both segments, combined with unprecedented demand from our Connectivity & Cloud Solutions (CCS) customers, underscores the significant scale of our AI infrastructure buildout,” says Rob Mionis, Chair & CEO.

Connecting Infrastructure from Chip to Rack

Celestica’s approach begins with the architecture rather than an isolated component. Its AI infrastructure work spans networking, compute, storage, racks, power and cooling, allowing individual elements to be considered as parts of a larger system. The company describes this design-led approach as a way to align infrastructure more closely with the requirements of specific customer environments.

Networking illustrates that philosophy. The DS6000 and DS6001 switching platforms provide up to 102.4 Tbps of non-blocking switching capacity, with 1.6TbE connectivity and configurations supporting different rack and cooling environments. The platforms are designed for AI back-end networks and support scale-up and scale-out architectures.

The architecture also leaves room for customer choice. Celestica supports open networking through SONiC, while its networking portfolio allows customers to pair its hardware with their preferred network operating system. That separation between hardware and software gives hyperscalers greater control over how networking platforms fit existing environments and future development plans.

Making High-Speed Networking Practical

The DS6000 family shows how Celestica translates system requirements into a physical platform. Its 64-port 1.6TbE configuration is designed to address bandwidth requirements in large AI clusters, while support for copper and optical interconnects provides flexibility around how systems are connected inside the data center.

The company’s work with Broadcom further showcases its capabilities. The DS6000 series uses Broadcom Tomahawk 6 silicon and combines that switching technology with Celestica’s system engineering. The result is a platform built around high-density networking rather than treating the switch as a standalone piece of hardware.

Celestica’s AI networking portfolio also extends into emerging rack-scale architectures. It recently collaborated with AMD to bring the Helios rack-scale AI platform to market, undertaking research and development, design and manufacturing for scale-up networking switches based on the Open Compute Project Open-Rack-Wide form factor.

Turning Modular Design into Deployment

A distinct system perspective appears in Celestica’s modular data center work. Instead of treating deployment as a collection of separate construction and technology activities, its model uses prefabricated, factory-integrated modules that can arrive prepared for installation. The customer defines the IT equipment and site while Celestica handles integration across key infrastructure elements.

Strong operational execution across both segments, combined with unprecedented demand from our Connectivity & Cloud Solutions (CCS) customers, underscores the significant scale of the AI infrastructure buildout.

That process brings together rack integration, power distribution, cooling systems, system-level validation and deployment. The approach is intended to create standardized infrastructure blocks that can be expanded incrementally, rather than requiring an entire facility to be built around a single capacity assumption.

Cooling becomes particularly important in this model because high-density AI environments place different demands on physical infrastructure. Celestica’s AI infrastructure portfolio includes liquid cooling alongside networking, storage, compute and rack integration. This arrangement allows thermal management to be addressed within the system itself, and not after the computing architecture has been established.

Carrying the System Through Its Lifecycle

Celestica’s role does not end when a platform leaves the design stage. Its capabilities extend from product engineering and manufacturing through rack integration, logistics and fulfillment, deployment support, spare parts strategy and asset management. This creates a workflow in which development and physical deployment can be connected rather than managed as unrelated activities.

Its global manufacturing footprint supports that model. Celestica maintains a network of sites and design centers across North America, Asia and Europe. Across the continents, its technology lifecycle includes design, new product introduction, engineering, supply chain management, manufacturing, systems integration and testing.

That reach becomes important when a platform must move beyond an individual deployment. Celestica’s modular data center model emphasizes repeatable infrastructure across regions, while its broader services include logistics, fulfillment, maintenance, asset management and IT asset disposition. The objective is to keep infrastructure support connected to the systems being deployed.

Scaling Infrastructure with Open Collaboration

Celestica’s partnerships further demonstrate how its model works in practice. Its technology partner ecosystem includes AMD, Broadcom and Intel, with collaborations spanning accelerated computing, open networking and storage. These relationships place Celestica within development programs where hardware, software and system architecture must align closely.

The AMD-Helios collaboration is especially illustrative because Celestica is responsible for research and development, design and manufacturing of scale-up networking switches within the rack-scale architecture. The switches are intended to connect next-generation AMD Instinct MI450 Series GPUs using the UALoE architecture, reflecting the company’s role beyond conventional manufacturing.

Its recent investment in AI infrastructure capabilities also reflects the scale of that commitment. Recently, Celestica made a $3 billion equity offering intended to support investments in high-performance AI data center infrastructure and Ethernet networking due to a surge in global demand.

Celestica’s distinction lies in how these pieces connect. A hyperscaler can engage its engineering organization around a platform, use its networking and compute capabilities to shape the architecture, draw on manufacturing and supply chain resources to produce it, and extend into deployment and lifecycle support. That continuity gives the company a practical role across the infrastructure journey.

The result is an approach grounded in system behavior rather than a single product category. The DS6000 family addresses high-density networking, modular data centers address deployment architecture and lifecycle capabilities extend support beyond installation. Each capability contributes to the larger requirement of delivering infrastructure that can scale while maintaining performance, reliability and architectural flexibility.

That combination provides the context for Celestica’s recognition as Top Hyperscale Data Center Solutions in Canada 2026. The award reflects the characteristics that distinguish its contribution to the hyperscale environment, which include deep engineering, open platform design, integrated manufacturing, global execution and a portfolio that connects networking, compute, cooling, deployment and lifecycle services. Its value ultimately rests on how those capabilities translate into infrastructure that customers can deploy, expand and support at scale.