CiscoProfessional
Implementing Cisco NX-OS Switches and Fabrics in the Data Center
DCNX · On request
View course →
CiscoProfessionalOn requestUpdated: July 2026
DCAI Training for Your Team — Implementing Cisco Data Center AI Infrastructure
In 2026 the fastest-growing line item in most infrastructure budgets is GPU compute — and the fastest-growing source of incidents is the network underneath it. Distributed training jobs generate traffic patterns classic data center designs were never sized for: enormous synchronized bursts where a single congested link idles an entire GPU cluster at eye-watering hourly cost. DCAI (Implementing Cisco Data Center AI Infrastructure) trains the engineers who must make that not happen.
The course covers AI infrastructure as a system: GPU server platforms, lossless Ethernet fabrics with real congestion management, the separation of back-end GPU traffic from front-end services, storage and data pipelines that keep accelerators fed, and the telemetry to operate it all. It is the newest course in the Cisco data center portfolio for good reason — demand for engineers who can build AI-ready fabrics currently outruns supply by a wide margin.
We organize delivery through authorized partners with instructors working on production AI deployments — live online in your time zone, on site, or as a private group scoped to the cluster you are about to build.
Who this course is for
- Data center engineers whose organization is standing up its first GPU cluster
- Network architects designing fabrics for AI training or inference workloads
- Infrastructure teams translating AI-team requirements into buildable designs
- Integrator engineers delivering AI infrastructure projects for customers
What your team will learn
Topic areas below are grouped in our own words. The full official agenda is available on request — it comes from the authorized partner as part of your quote.
- What AI workloads actually demand from infrastructure — and why yesterday's fabric design chokes on them
- GPU compute platforms in the data center: servers, interconnects and the power and cooling small print
- Building lossless Ethernet fabrics for AI: congestion control that keeps training jobs from stalling
- Back-end versus front-end networks — separating GPU-to-GPU traffic from everything else
- Feeding the beast: storage and data pipeline considerations for training and inference
- Operating AI infrastructure: telemetry, job-level visibility and capacity that runs out faster than budgets
Key facts
| Duration | On request |
|---|---|
| Level | Professional |
| Prerequisites | Solid data center networking and compute knowledge; familiarity with AI/ML workload concepts helps but is not required |
| Formats | Live online (VILT), On-site at your location, Private group, Seats in scheduled groups |
| Languages | English + local-language instructors on request |
Formats and how we organize it
- Live online (VILT)
- On-site at your location
- Private group
- Seats in scheduled groups
One request → three options
Fill in the form below — it takes a couple of minutes. Within 5 business days you compare 2–3 concrete offers — provider, instructor, dates, price below the public list. We contract and administer the delivery end to end. Details:how we work and vendor training.
DCAI vs alternatives: which to choose
Factual criteria only — level, duration, audience and certification. The right choice depends on your team's starting point and stack.
Cisco DCNX (NX-OS Switches and Fabrics)
The general-purpose fabric course: Nexus platforms, vPC and VXLAN EVPN for ordinary data center traffic. Choose DCNX for the baseline fabric skills; DCAI builds on that ground and specializes in the lossless, GPU-centric designs AI workloads impose.
Cisco DCNAUTO (Data Center Automation)
The automation track: managing data center networks with code, pipelines and controllers. AI clusters are rebuilt and rescaled constantly, so the two courses pair naturally — DCAI for what to build, DCNAUTO for operating it without an army.
A custom intensive
AI infrastructure projects rarely match a stock syllabus: a custom program can combine fabric design for your GPU count, your storage stack and your MLOps tooling into one workshop for the exact team that must deliver it.
Need only part of DCAI?
If your team needs a subset of this program — or a mix of topics from several courses — we build a custom 2–4 day intensive on our own materials, shaped by your infrastructure. One custom program typically replaces 2–3 catalog courses.
Explore custom coursesDCAI: frequently asked questions
Why do AI workloads need special network training at all?
Distributed GPU training behaves unlike ordinary traffic: massive synchronized flows where one congested link stalls an entire multi-million-dollar job. Designing for that means lossless Ethernet, careful congestion control and separated back-end fabrics — patterns most data center engineers have never had to build before.
Is DCAI about networking only, or the whole infrastructure stack?
The whole stack, network-first: GPU compute platforms, the fabrics connecting them, storage and data-pipeline considerations, plus operations and telemetry. The course targets the infrastructure engineer who must make an AI cluster work, not the data scientist using it.
Do attendees need machine learning knowledge?
No. The course explains AI workload behavior — training versus inference, how distributed jobs communicate — strictly at the depth infrastructure decisions require. What is assumed is solid data center networking and compute background, roughly DCCOR-level familiarity.
We are still at the planning stage for a GPU cluster — is the timing right?
Planning stage is the ideal timing. The expensive AI infrastructure mistakes — undersized fabrics, wrong oversubscription, storage that starves the GPUs — are design mistakes. Teams who train before procurement ask their vendors noticeably harder questions.
How do we organize DCAI training?
Fill in the request form on this page — team size, format, timing. Within 5 business days you receive 2–3 options from authorized partners with vetted instructors, dates and prices below public lists.
Request 2–3 DCAI training options — quotes within 5 business days
No commitment — describe the task and we come back with concrete options, dates and prices.
Cisco, DCAI and course names are trademarks of their respective owners. Uptime Skills is an independent training operator and is not affiliated with or endorsed byCisco. Official courses are delivered by authorized training partners.