Good morning everyone. It's great to be here at Q 42. 2025. My name is Xi and I'm part of Microsoft Cloud infrastructure team. Over the past several years, I've worked on scaling hyperscale data centers across multiple regions. What I'll share today is shown directly from real world experiences, the success, some mistakes, and the lessons learned. Our focus is simple, how we can accelerate data center build outs. Not by cutting corners, but by improving coordination, panelization and ness across every layer of the infrastructure. This is increasingly relevant today as organizations push for faster cloud expansion, lower time to market, and more efficient resource utilization. Let's dive in. Here's a quick overview of the flow today. We'll start with the foundational readiness, where we'll discuss why early planning and segregating critical workloads from the backbone of acceleration. Then we'll cover dependency management, essentially, how to identify and visualize interconnections between the tasks that are often invisible, but have massive downstream effects. Next network connectivity, one of the most time sensitive areas in any build out. I'll show how to manage the underlay and overlay networks in parallel. I'll then move to server and service deployment where the automation and pre staging really make the difference between a three month and a nine month ramp up. Finally, we'll tie it all together with integration and scaling and close with some key metrics and takeaways that you can apply to your own data center projects. Let's start by grounding ourselves in a core challenge. Modern data centers aren't single threaded project. The highly interdependent ecosystems involving land power, cooling, networking compute compliance are moving in parallel. The problem is a delay in any one of these domains, cascades into everything else. For example, a late fiber trench permit, delays and network installation, which delays testing, which then starts production rollout. The insight here is simple but powerful. We can't think linearly anymore. Acceleration requires breaking the chain of sequential dependency and moving towards parallelism. The rest of this talk will unpack how to do that exactly. Foundation readiness is where acceleration truly. Site selection, land acquisition, and permitting, those need to start at least 18 to 24 months before you target operational date. If you're waiting for design approval before starting zoning utility negotiations, you're already lost valuable. Physical construction likewise must be modular. We are seeing a shift towards designs that support phase expansion, smaller faster deployable modules rather than single large build and power and cooling shouldn't be afterthoughts. There should be parallel streams. Teams working on utility agreements, backup systems and ancy planning should operate independently, but synchronously. The main takeaway, segregate early plan parallel, and never let one workflow block another. That mindset alone can shave off months of the project timeline. Once your teams are operating in parallel, you need a way to keep visibility across the chaos. That's where the dependency metrics comes in. It's essentially you have blueprint for cooperation coordination between various teams. It maps every component, every milestone, and their interdependencies. For example, power readiness affects lag delivery, which in turn of X network turn up and validation. By visualizing this network of relations project managers can identify bottlenecks. Before they materialize, and this matrix is not static. It must be revisited and updated continuously as conditions evolve. Think of it as a single source of truth. It's what separates reactive teams from proactive ones. Our infrastructure is often the pacing item in data center construction. Utility coordination alone can take over 18 months from negotiating, creating the connections to securing substation permits. That's where those conversations must start. Almost as soon as society selected. Power sufficient and redundancy planning need to factor in future expansion and density. We are not talking about a 15 to 20 kilowatts per rack in many cases. And with the ai tracks this power demand will be going up in an exponential fashion. One of the biggest acceleration levers here is pre-ordering long lead time components like generators, transformers, and UPA systems, even before final electrical design is locked down. Yes, it's a gas letter risk, but the alternative is waiting months for delivery and losing your schedule buffer. Slide o provisioning cost is far cheaper than a six month delay. Network is another critical bottleneck. Fiber installation alone can take six to 12 months. Competing agreements of right of way permits can drag that even longer. So network design and external carrier engagement must begin at the same time as land and power planning, not after assess existing fiber infrastructure before final site selection. This can be the difference between a one year and a two year build out. Also, consider alternative models like dark fiber or IGOs can bring you flexibility and time. Remember, connectivity delays are preventable if addressed early. Network build out doesn't have to wait for construction completion while the building is going up. Design and vendor selection can happen in parallel. Network design can be based on different stages. They can be on different architectures, like a class fabric leaves, spine architectures but it should be validated at a modular system designed to scale. Modules can be built in panel, and this parallelism endures that network readiness aligns with physical completion, and we can compress the scheduled timelines. Traditionally teams waited for the data center to be complete before installing servers that's slow and inefficient. Instead, the pre-stage and validate servers offsite ship ready to deploy racks and US automation to configure the racks. This transformation basically brings in the deployment timeline from months to weeks, which is an enormous efficiency gain. Acceleration continues at the software layer with infrastructure as code. All configurations, network storage, compute are defined programmatically, versioned, and tested through the Ci CD pipelines. Containerization abstracts services from hardware dependencies. 11. Deployment and development in parallel tracks. Then comes automated testing thousands of checks across clusters to ensure everything behaves predictably at scale. Finally, progressive rollouts can utilizes and stage deployments, allow services to go live in small controlled phases. The net effect, your data center isn't waiting for the perfect moment. It's gradually becoming productive, even as the final construction finishes. Complaint is often viewed as a bottleneck, but starting early flips that dynamic. For instance, embedding security and compliance experts from the day of one ensures your design processes and automation all meets STAs for the start. Preliminary audits during construction, automated policy validation and real time compliance documentation all minimize surprises, filter early compliance equals faster, safer co lives. The traditional commissioning is sequential and slow. The fast track approach performs component level testing During installation, multiple teams can work in commissioning different availability zones in panel and independent motor sections of the data center can be hone handed off progressively. This method reduces post-construction testing from months to mere weeks. Accelerating time to service availability. Integration is where acceleration becomes real. Every system lacks power cooling, network compute need to interlock seamlessly. The most effective approach is cross-functional integration teams. Engineers from all domains collaborate regularly with shared dashboards and lack dependency tracking. Integration isn't just technical, it's also cultural. When every team shares ownership for readiness, you eliminate silos. A clearly defined handoff and testing process ensures smooth transitions into operations. Measuring acceleration is crucial for continuous deployment and improvement. Tracking metrics. Like overall project duration, time to production can showcase the time based acceleration improvements. ROE Acceleration shows how data center stack generating value, percentage of parallel execution shows the efficiencies of our acceleration strategy. These indicators reveal how well your acceleration strategies are performing and where to focus next. Continuously measuring this helps you refine your approach. Every project should get slightly fast, faster, and more efficient. To summarize, let's recap what you discussed so far. Early workflow segregation prevents dependency bottlenecks, dependency metrics, keep your team aligned and adaptive. Network and power planning must happen in paddle, not sequentially. Server pre-staging and automation drastically reduce time to like compliance and commissioning when done early accelerate instead of delay. So these are our key learnings. So for your next project, your workflows, identify your top bottlenecks and establish parallel execution paths. Hope the session is useful. Thank you for your attention. I'd be happy to discuss how these strategies can be applied to your environment. Thanks and have a great day.