As we wrap up the DIY HomeLab series, I've been thinking back over how we got here. It started with one Raspberry Pi 4 running a simple proof of concept, and it's now a distributed setup that powers real client projects. I've learned a lot along the way.
This is a checkpoint more than an ending. Self-hosted infrastructure, edge computing and AI integration are all changing quickly, and our HomeLab keeps changing with them.
In this post I'll cover what's on our roadmap, the advanced topics we're exploring, and where I think HomeLab infrastructure is heading over the next few years.
A quick recap
Over the past five posts we built the HomeLab from the ground up:
- Hardware: a distributed mix of Raspberry Pis, Orange Pis and dedicated servers
- Networking: ZeroTier and Cloudflare for secure connectivity without the usual hassle
- Container management: CasaOS, which made Docker deployment something I actually enjoy
- Applications: Node-RED, databases, monitoring and productivity tools
- Real-world use: systems that run actual business operations
The whole thing cost around $2,100 in hardware and $120 a year in cloud services. What we got back in learning, capability and business impact is hard to put a number on, and we're still adding to it.
Advanced topics
AI integration beyond simple inference
Our current AI setup runs local LLM inference on the dedicated server. We're expanding that in two directions.
The first is distributed AI processing:
- Edge AI on Raspberry Pi with Google Coral TPU accelerators
- Computer vision pipelines for security and automation
- Real-time decision making for IoT deployments. Our Alpha Block educational robotics platform already runs swarm coordination on a CM4 with AI-assisted firmware (100% written by Claude Code + GLM)
- Federated learning across multiple client sites
The second is using AI to run the infrastructure itself: automated system optimization based on usage patterns, predictive maintenance for hardware and applications, resource allocation across the cluster, and natural language interfaces for system management.
We're already testing AI agents that analyze our infrastructure metrics, spot optimization opportunities and sometimes apply the changes themselves. It works like a systems administrator who is on shift around the clock.
Edge computing
We want more of the processing to happen at the edge, close to where the data comes from.
For client sites, that means ruggedized Pi deployments for industrial environments, satellite connectivity for remote locations, and systems that run on their own and sync with our central systems periodically, processing locally and using the cloud for backup and analytics.
We're also looking at mobile edge units: portable HomeLab setups for temporary deployments, vehicle-mounted systems, kits for emergency response and disaster recovery, and rigs for field research and data collection.
Networking beyond ZeroTier
ZeroTier has been great, but we're exploring two more advanced areas.
Mesh networking would give us self-healing network topologies, automatic failover and load balancing, bandwidth optimization across multiple connections, and integration with satellite and cellular networks.
Network Function Virtualization (NFV) means software-defined networking on ARM hardware: virtual firewalls and load balancers, traffic shaping and QoS management, and network analytics.
Storage beyond local disks
Our storage started with simple USB SSDs. The next step is distributed storage, with GlusterFS or Ceph clusters across multiple Pis, automatic replication, tuning for ARM, and cloud storage for a hybrid setup.
On top of that we want smarter data management: automated tiering based on access patterns, compression and deduplication, predictive caching and prefetching, and AI-driven backup and archival.
Emerging technologies we're watching
RISC-V
ARM has served us well, but RISC-V is where I expect open hardware to go. The instruction set architecture is completely open, with no licensing fees or restrictions, it can be customized for specific workloads, and the range of development boards keeps growing.
We're already testing early RISC-V boards and planning how we'd migrate once the ecosystem matures.
WebAssembly (WASM)
WebAssembly is no longer only a browser technology. It gives you a universal runtime for any architecture, near-native performance with sandboxing, and freedom to write in whatever language you like, which suits edge computing well.
We're experimenting with WASM-based microservices that run identically on x86, ARM and, eventually, RISC-V hardware.
Quantum-safe cryptography
As quantum computing advances, our security has to keep up. We're reading up on post-quantum cryptographic algorithms, quantum key distribution for highly sensitive communications, and hybrid classical-quantum security systems, so the infrastructure we build now doesn't need ripping out later.
What $2,100 in hardware enabled
The financial case is simple. We avoid cloud hosting costs for development, staging, IoT monitoring and internal tools, and that alone covered the hardware within the first few months.
The less obvious returns matter more to me. When a client asks "can you build X?", we can have a working proof of concept running on lab hardware in hours instead of days, and several of our biggest projects started as quick HomeLab demos. When we recommend an architecture, we can show clients the same stack running in our own environment. And every debugging session, failed upgrade and 3 AM hardware swap has made us better at the work we do for clients.
The $2,100 paid for itself quickly. The bigger return is the infrastructure knowledge our team now has, which you can't buy.
Lessons for your own HomeLab
If you're thinking of starting your own HomeLab, this is what I'd pass on from our experience.
- Start with a purpose, not hardware. Don't buy hardware because it's cool. Buy it because you have a specific problem to solve or skill to develop.
- Treat failure as learning. Every crashed container, failed update and network outage teaches you something. Write down what went wrong and how you fixed it.
- Lean on the community. The open-source communities around HomeLab tools are excellent. Join in, contribute, and learn from other people's setups.
- Think long term. Build infrastructure that can change. Today's experimental setup might become tomorrow's production system.
- Build in security from day one. Don't leave security for later. Make secure practices part of your workflow from the start.
What we're excited about next
This isn't a corporate roadmap, just the things we're curious about right now:
- RISC-V boards are getting good enough to run real workloads. We have a couple on the bench and we're testing how our stack translates.
- WebAssembly for edge computing, so the same microservice runs on ARM, x86 and eventually RISC-V without recompiling. We're experimenting with this now.
- AI-powered operations, using local models to analyse infrastructure metrics and suggest optimisations. We already do this by hand; automating it is the next step.
- Distributed storage across our Pi cluster using GlusterFS. SD cards die, and we need better redundancy.
A personal reflection
Three years ago, when I set up that first Raspberry Pi to solve a client's IoT problem, I had no idea it would lead here. It started as a way to save money. It became a learning platform, then something the business depends on, and honestly a lot of fun.
I get real satisfaction from building systems that work, solve real problems and that I understand completely. With so much hidden behind cloud abstractions and managed services, having hands-on control of your own infrastructure feels refreshing.
Whether you're setting up your first Raspberry Pi or already running a distributed setup, start small, break things and keep learning.
Happy building.
We share our HomeLab configs and lessons on the Alpha Bits blog and GitHub. If you're on a similar journey, we'd love to hear what you're building.