Deciding the Future - Building New Networks for the AI Era
Author’s Note
Vishal Shukla- CEO, Aviz Networks
When you talk to traditional vendors, the conversation quickly turns into a product catalog. You don’t get a clear picture of the trade‑offs between different approaches, or the historical patterns behind them. You get slides, not a mental model.
That gap is why I decided to write this book.
I happen to be the CEO of Aviz, and yes, Aviz appears in these pages. But this is not meant to be a sales brochure. The true purpose is simpler and, I hope, more useful:
- To give executives and technology leaders a clean way of thinking about networking in the AI era
- To break a messy topic into modular pieces that can be discussed in the boardroom as well as in the lab
- To provide objective frameworks for evaluation, not just opinions or marketing claims
Throughout the book I lean heavily on history: how previous technology waves evolved, how standardization shifted value, how lock‑in played out, and what that means for AI networks. If you understand the pattern, the current noise becomes much easier to navigate.
My hope is that when you finish this book, you won’t just have answers. You’ll have a reusable way to ask better questions.
Key Highlights from the Book
Why AI Is Forcing a Network Rethink
AI workloads are changing traffic patterns, bandwidth assumptions, and infrastructure economics — faster than most organizations realize.
The Two Decisions Every Leader Must Make
- How should we build networks for AI workloads?
- How should we use AI to run and secure networks?
Most companies blur the two. This book separates them clearly.
Scale-Up vs. Scale-Out — In Executive Language
Understand the strategic trade-offs behind growth architecture choices — without drowning in protocol jargon.
ROI & TCO in the AI Era
AI changes both revenue opportunity and cost structure.
The book helps you rethink infrastructure decisions through a business lens.
Standardization vs. Lock-In
History shows where value ultimately moves.
Leaders who recognize the pattern early win.
Pragmatic Starting Points
You don’t need a greenfield AI factory to begin.
There are structured ways to modernize without disruption.
Insights from the Leaders Defining the AI-Driven Networking
“Most books about AI talk about models and applications. Networks in the AI Era is the first I’ve seen that explains, in plain language, what the network has to do to make any of that real. It gives executives a way to think about fabrics, observability, and AI‑driven operations without getting lost in product charts.” - Partner, Global Networking Company
“This book captures what we hear from our customers every day: they want freedom to choose hardware and still have a coherent operating model. The frameworks in these chapters make it much easier for teams to align on the path forward for AI data centers.” - Partner, Systems & Manufacturing Company
“When we advise customers on AI infrastructure, we keep coming back to the two questions this book is built around: networks for AI, and AI for networks. Having that structure – plus concrete evaluation worksheets – has already changed the quality of our design conversations.” - Cloud / Integration Partner
Resources
History & Evolution of Networks
- ARPANET
- ARPANET – Wikipedia
- First computer-to-computer message – LiveScience
- Data Center Multi-Tier Model Design – Cisco
- Three-Layer Hierarchical Model – GeeksforGeeks
- Three-Tier vs. Leaf-Spine – Intelligent Visibility
- Spine-Leaf vs. Traditional DC Architectures – STORDIS
AI Data Centers as “AI Factories”
- Turning Data Centers into “AI Factories” – NVIDIA Blog
- The Data Center is the New Unit of Computing (video)
- Datacenters as AI Factories – OfficeChai
- Nvidia’s $100bn bet on “gigantic AI factories” – Financial Times
AI Traffic Patterns & Fabrics
- East-West vs. North-South for AI – Mulcas
- DPUs/SmartNICs for AI Fabrics – Techwrix
- InfiniBand vs. Ethernet for AI Clusters – ArcCompute
Scale-Up Interconnects & GPU Systems
- NVIDIA NVLink & NVSwitch
- NVIDIA NVLink High-Speed GPU Interconnect
- Fifth-Generation NVIDIA NVLink – AMAX
- Scaling AI Inference with NVLink & NVLink Fusion – NVIDIA
- Ultra-fast GPU Communication with NVLink – Uvation
Ultra Ethernet & Open AI Fabrics
- Ultra Ethernet Consortium (UEC)
- UEC Specification 1.0 – Linux Foundation
- Ultra Ethernet for Scalable AI – Cisco
- Future of AI Networking with UEC – Nokia
- Understanding the UEC – DriveNets
Commoditization, Hyperscalers & Linux Analogy
- The Commoditization of Server Hardware – Data Center Knowledge
- Has the Server Been Commoditized? – Moor Insights & Strategy
Philosophy of Change
- Heraclitus – Wikiquote
- Heraclitus: “There is nothing permanent except change.” – The Socratic Method
Technology, Operations & AIOps
- Bill Gates technology quote – BrainyQuote
- Definition of AIOps – Gartner
- What Is AIOps? – IBM
- What Is AIOps? – Splunk
- Gartner on AIOps – Aisera
Streaming Telemetry & Monitoring
- Streaming Telemetry vs SNMP – IBM Community
- Benefits and Drawbacks of SNMP & Telemetry – Kentik
- SNMP vs. Telemetry – ITcare
- Why Use gNMI Over SNMP in 2024 – kmcd.dev
- gNMI & Streaming Telemetry – MapYourTech
Tool Sprawl & Operational Friction
- How to Tackle Tool Sprawl – The New Stack
- Tool Sprawl Kills IT Productivity – ScriptRunner
- Context Switching Is Killing Productivity – Lokalise
- Hidden Cost of SaaS Sprawl – Litcom
AI Strategy, Roadmaps & Governance
- AI Roadmap – Gartner
- AI Maturity Model & Roadmap Toolkit – Gartner
- Gartner AI Governance Guide – Atlan
- 40% of Agentic AI Projects Scrapped by 2027 – Reuters / Gartner
Autonomy & AI System Analogies
AI Factories & GPU-Centric Data Centers
AI Infrastructure Spend (>$100B–$200B)
- AI Infrastructure Spending to Surpass $100B – IDC / BusinessWire
- AI Infrastructure Spending to Exceed $200B – FutureIOT / IDC
AI Workloads, East–West Traffic & Fabrics
Bandwidth & 100G–1.6T Evolution
- AI Data Center Networking – Ethernet for the AI Era – Supermicro
- AI Networking in Data Centers – Cisco
SONiC as Open, Hardware-Agnostic NOS
Mainstream Vendor Alignment Around SONiC
- Craft Your AI DC with Cisco 8000 and SONiC
- Empowering AI & Telco Networks with SONiC – Dell
- Arista Switches Powered by SONiC
- Juniper, HPE Join the SONiC Foundation
SONiC + AI Networking Startup Wave
- Nexthop AI Launches with $110M
- Aviz Raises $17M – Futuriom
- Nexthop AI – Lightspeed
- Aviz Networks Raises $17M – SiliconANGLE
LLM-Based AIOps / NetOps
- AIOpsLab: Evaluating AI Agents for Cloud Ops – Microsoft Research
- A Survey of AIOps in the Era of LLMs – Huang et al.
Deep Observability & Software-First Visibility
- Deep Observability Market to $2B by 2027 – 650 Group
- Gigamon Deep Observability Market Share – 650 Group
Overall AI & AI-Infra Spend
AI DC Networking / Ethernet for AI
Deep Observability TAM
AI for Networks / AIOps / NetOps TAM
Agentic AI & IT Budgets