Last month, Las Vegas, USA, was filled with the enthusiasm of IT engineers and business leaders from around the world.
It is none other than HPE's biggest festival HPE Discover 2026 That is the reason.
The main stage that captured the attention of the world at this event was Antonio Neri, Chairman and CEOThe main keynote of and the subsequent day Rami Rahim, Executive Vice President of NetworkingIt was a session.
While many companies only pay lip service to AI, HPE Juniper Networks acquisition In the following months, they relentlessly demonstrated to the market the reality of the 'Self-Driving Network,' the complete integration of hardware and software.
“We have summarized the key highlights of Discover 2026, which declared that ”AI design starts in networks,” into three themes.
Theme 1: Will We Build AI on a Weak Foundation?
At the beginning of the presentation, Vice President Rami Rahim used the analogy of San Francisco's luxury skyscraper, the Millennium Tower, which slowly sank and tilted due to weak foundations. CEO Antonio Neri also issued a warning, stating, "No matter how flashy the AI agents and infrastructure you build, if the underlying data and network foundation cannot hold up, the entire architecture will collapse.".
Currently, numerous companies are pouring billions of won into purchasing GPUs and training massive models.
However, if the network acting as those blood vessels causes bottlenecks and induces latency, those expensive GPUs will simply sit there idly waiting. Networking is no longer just a simple connection channel, The most lethal strategic weapon defining the AI erano see.
To withstand the massive data movement and ultra-low latency requirements of large-scale AI workloads
Network as the central strategy of infrastructureIt must be this
Antonio delivered a hard-hitting fact, stating that if delays of microseconds accumulate, model training that would normally take 90 days could be dragged out by as many as 30 days.
Theme 2: Roadmaps of Words Are Over: 4 Key Integrated Points to Shake the Market
The technological convergence between the two companies unveiled at this Discover was demonstrated on stage as a 'current solution' that is immediately operational, rather than a 'future roadmap.'.
① Software Vertical Integration (Marvis's Advance)
- Marvis's Aruba Central Expansion: Juniper Mist's unrivaled conversational AI assistant Marvis has finally been installed on the Aruba Central platform.It works. The Global NOC View analyzes and automatically recovers wired and wireless-wide issues, such as complex causes of failure, in natural language (Plain English).
- Integration with GreenLake Intelligence: Marvis's networking data and GreenLake's server and storage data Perform mutual correlation analysis (Reasoning) at the cloud levelIt automatically completes tasks such as virtual machine (VM) migration, infrastructure optimization, and compatibility verification, which used to take weeks, in just a few minutes.
② Hardware Orchestration Integration
- Mist Integrated Management for Aruba CX Switches: Proven in the enterprise market Now centrally manage the HPE Aruba CX switch series from Day 0 to Day 2 on the Mist platformIt is possible. Onboard with a single QR code scan and measure wired Service Level Expectations (SLE) in real time. This is a measure that completely breaks down the barriers between vendors.
③ The Evolution of Data Center Automation:
- Apstra ➡️ Renamed to 'Data Center Director' and equipped with Marvis AIOps: Apstra, Juniper's 자랑하는 intent-based data center automation solution
HPE Networking Data Center DirectorThe official name has been changed to.
As an official full-stack verification solution in the HPE AI Factory portfolio, it has gone beyond simply monitoring equipment up/down through the Marvis AIOps engine for data centers. Including voltage, current, temperature, and CRC errors Predict physical failures such as optical cables in advance by analyzing over 30 key metrics in real-time using a machine learning model.do.
- Realization of Agentic Root Cause Analysis (Agentic RCA): Troubleshooting time that used to take hours to days when a failure occurred ‘A few minutes’ It dramatically shortens the timeframe to units. Marvis's agentic AI system performs multidimensional reasoning on switch telemetry data, application flows, and vast historical TAC technical support cases.
Especially internal ‘Marvis Minis (Digital Twin)’ Agents perform real-time virtual tests identical to actual user traffic to identify accurate root causes using human-engineer-level reasoning and logic, and even provide automated action guides.
④ Convergence of Network and Security (Secure AI-Native)
- Integrated SASE Orchestrator: Edge Connect SD-WAN and SSE are combined into a single console and a single data path (Single-Pass) to provide latency-free zero-trust guardrails.
- AI-Aware Firewall (Inline AI Control): We perform real-time NLP (Natural Language Processing) inference at the firewall level to ensure employees can safely use generative AI such as ChatGPT/Gemini/Claude. Corporate confidential keywords (
secret,restrictedIt showcased overwhelming governance that detects and blocks prompts/file uploads containing ) or source code in milliseconds (ms).
Theme 3: Large-scale AI Data Centers and a Monster Lineup for Inference
The biggest enemies that disrupt the AI lifecycle (from training to edge inference) are network bottlenecks and power consumption.
Antonio Neri emphasized sustainability and hardware efficiency, stating, “The essence of the AI factory is converting electricity (Electrons) into tokens,” and Juniper’s formidable hardware specifications supported this.
| division | Product Name | Key Technologies and Differentiation Points |
| Scale-Up | QFX 5252 | – Industry's first Ethernet switch dedicated to AMD Helios systems – Connect 72 GPUs in a single rack, 260Tbps Bandwidth provision |
| Scale-Out | QFX 5250 | – World's highest performance 100% Direct Liquid Cooled switch – Based on the Tomahawk 6 chipset 1.6T Connectivity Maximize power availability with support |
| Scale-Across | PTX 12000 | – Ultra-high density AI fabric router. Ultra-high-speed data lake connection between data centers (DCI). – Quantum-resistant firewall SRX 4700Combined with to form a powerful data protection shield |
| Edge Inference | QFX 5140 | – Dedicated switch for distributed AI inference at the edge – 16Tbps Equipped with capacity and AI-optimized load balancing |
“Architect Deliberately, Start with the Network”
CEO Antonio Neri concluded the keynote by making a powerful proposal.
“Design the architecture carefully, but be sure to start with the network.
(Architect deliberately, Start with the network).”
The days of old-fashioned networking, where human engineers were locked in the server room all weekend sifting through logs to fix a Friday afternoon outage, are over. Now, we have systems that predict threats, identify confidential information, detect failures, and self-heal. autonomous driving network.
This is the reality created by the collaboration of HPE and Juniper, and the true foundation that will lead the era of the agentic enterprise.
“Just because the network is up does not mean the user is having a good experience.”As Rami Rahim said, we have now fully entered the era of 'Experience-First' autonomous driving.
Are you ready to welcome the era of new networking architectures?
