Stefan Dyckerhoff

AI infrastructure is entering a new phase. While AI model training has traditionally been compute-bound, the next wave of AI, large-scale inference powering agentic AI, is increasingly memory-bound and latency sensitive. Today we introduced the first product in our new Penguin MemoryAI™ family of innovations, designed specifically for inference-focused AI factories. We’re excited to announce the industry’s first production-ready CXL-based KV cache server, delivering up to 11TB of CXL-based memory to dramatically expand the memory available to GPUs and accelerate performance large-scale AI inference. "CXL" or Compute Express Link, is a high-speed interconnect that enables scalable, shared memory across GPUs and CPUs, allowing access to large pools of external memory. “KV,” or key-value cache, stores inference context to accelerate large language model or LLM responses. Without caching, the model recomputes data for all previous tokens every time a new token is generated. That’s wasteful. KV cache stores store past attention data so the model doesn’t recompute it every step, making response text generation much faster. Think of writing a summary of a book: Without KV cache: reread the entire book every time you write a new sentence 😅 With KV cache: remember what you've already read and just build on it. A CXL-based KV cache is an extension of the standard KV cache idea, where the cache is stored not just in GPU memory, but in CXL-attached memory (Compute Express Link) to overcome memory limits. So what is a CXL-based KV cache? It means: 👉 Instead of storing all KV cache data in GPU memory, 👉 you offload (or tier) KV cache into CXL-attached memory If you’re attending NVIDIA GTC AI Conference this week, stop by Penguin Solutions Booth #1031 to explore this innovation. Proud of the incredible team at Penguin Solutions who brought this innovation to life based on the experience of designing, deploying and managing AI Factories around the world for our Enterprise and Sovereign Cloud customers.

51

3 Comments

Ericsson supported Sabinal Energy with enterprise 5G with centralizing SCADA and delivering cloud-managed wireless WAN. They also gained: ✔️ Bandwidth of up to 500 Mbps in remote offices ✔️ Sub-40 ms latency ✔️ Centralized control through NetCloud As Sabinal’s VP of Infrastructure put it: “Ericsson’s enterprise 5G solution was a game-changer for us.” This is the kind of impact that inspires me to keep building the future of enterprise networks.

3

Agentic AI promises more than traditional automation; it learns, reasons, and acts autonomously. But without orchestration, it can create chaos instead of value. In this blog, Shawn F R. Roberts, Automation Solutions Architect at Broadcom, explains why enterprise orchestration is the foundation for scaling agentic AI safely and strategically. Read the full post here: https://lnkd.in/dVAq7wXB

20

UC has evolved dramatically, but one thing hasn’t changed: latency, jitter, and packet loss still affect call and meeting quality, and no amount of bandwidth can eliminate them without a thoughtful QoS strategy. In this article, we take a fresh look at QoS for modern UC environments, including multi-media sessions, distributed networks, and cloud-centric architectures, and explain how to implement QoS on modern networks. #UnifiedCommunications #QoS #UCaaS #Networking #SDWAN #EnterpriseIT https://hubs.li/Q03ZRqvz0

3

Palantir Technologies & Lumen Technologies to Redefine Real-Time AI Infrastructure 🚀 Palantir has announced a major new AI partnership with Lumen Technologies, combining its AI decision-making systems with Lumen’s vast fibre network to deliver faster, real-time insights across telecoms, energy, and defence. The deal marks one of Palantir’s biggest steps yet toward embedding AI directly into the backbone of US digital infrastructure. With government and enterprise contracts fuelling a 311% total shareholder return over the past year, this collaboration signals a wider trend, the convergence of AI software and data infrastructure to enable faster, more secure decision-making at scale. At Answertree, we see the same principle driving change in bidding and proposal management. Just as Palantir connects intelligence with infrastructure, our platform connects AI-powered content retrieval, automation, and task coordination, transforming how teams respond to complex RFPs and manage high-stakes documentation. Read more: https://lnkd.in/ehmeKx2X #AI #Automation #DigitalTransformation #Answertree #BidManagement #Efficiency #FutureOfWork #Innovation #Palantir

9

Networks hum flawlessly 150ms latency, <1% packet loss, green dashboards everywhere. IT sector celebrates perfect QoS. Then users abandon blurry video calls mid-sentence. Technical perfection meets human frustration. QoS engineers prioritize packets and tune buffers religiously. Bandwidth flows smoothly. SLAs shine. Yet customer support reps hang up on frozen screens, costing $200K monthly in churn. Reality hits when perception enters the equation. The same 5% packet loss that barely dents CRM updates obliterates VoIP conversations. Video demands 4x the tolerance of email. Device screens, network congestion, user expectations—all reshape the same signal into delight or despair. Alignment begins with correlation. AI dashboards now map MOS scores against buffer pools dynamically. Video streams steal priority from bulk transfers. Sentiment analysis from call transcripts recalibrates thresholds weekly. CSAT climbs 25% as the network learns what users actually feel. Business transformation follows. Churn drops 18% when NOC teams track delight scores over uptime percentages. Product managers iterate on perceived friction points, not theoretical bottlenecks. Revenue per session compounds through retention math. #QoSvsQoE #CustomerExperience #NetworkTruth #BIJ

8

Fixed Wireless Access (FWA) is rising from a “last-mile fix” to a strategic growth engine in the 5G and 6G era. With advanced 5G/6G tech, AI-driven network planning, and new CPE capabilities, FWA now expands market reach, drives revenue, and enables new services. That's not just in rural, but urban environments, too. Key to success: integrate FWA with broader network strategies, leverage AI/automation for planning and assurance, and target diverse use cases (SMEs, IoT, temporary setups). As 6G introduces even more intelligence and edge capabilities, FWA will be foundational for future-ready telecoms. #FWA #5G #6G #Telecom #NetworkStrategy #AI #Broadband #DigitalTransformation #Connectivity #EdgeComputing #TelcoInnovation

8

1 Comment