KubeCon Japan 2026 in Tokyo is behind me, and like every edition of this conference it left me with more questions than answers — in the best way possible. Between the talks, the hallway track, and a few quiet moments in front of a temple, I recorded five short takeaways. Here they are, in one place.
From CPUs to GPUs: the workload has changed
The first thing that hit me at KubeCon Japan is how different the workload has become. A few years ago we were keeping microservices happily running on CPUs. Today the focus is on AI models — and not small ones. We have gone from models of a few billion parameters to ones pushing hundreds of billions, and soon into the trillions.
That scale changes everything for the platform team. A model can be so large it no longer fits on a single GPU, so we have to interconnect multiple GPUs and multiple servers. The networking and storage pressure is real: a modest model today sits at 20–30 gigabytes, and that puts stress on every layer underneath it.
The type of workload we are handling, especially with AI, is a different beast.
Accelerators beyond the obvious
When we say “AI hardware” we immediately think of GPUs and one vendor. But the accelerator landscape is far more diverse. Google’s TPUs are remarkably energy-efficient. NPUs and other programmable accelerators are shipping too. I am genuinely curious to see where this goes: if we can shape faster, better-quality, domain-specific, even micro open-source models, that benefits everyone.
A strong foundation still matters
Here is the part I keep repeating to anyone who will listen: don’t get scared. AI gives us velocity, but a good foundation is more important than ever. You still need senior people who understand deeply what a machine is doing — the intent, the specification, the why — to get real value out of these wonderful tools.
If you are learning programming today: yes, it is more important than ever. Build your hobby project. Create your crazy idea. Connect with these tools. They give us coding capability and speed, but translating a business requirement into value is still a human job.
Is it real, or is it AI?
I hear “is it real or AI?” from everyone now — technical and non-technical alike. My answer is the same every time: we still need a strong foundation. We still need prompt engineers. We still need people who run things locally and who understand what a transformer actually is, and how models are shaped to create more innovation. Keep studying. Stay hungry. There is still a job for you.
The developer career in 2026
Is it still worth pursuing a programming profession in 2026? I think so. Even the biggest AI companies are hiring developers by the thousands. The statistics keep showing a big, unmet demand for high-skill people. So don’t get scared — embrace your career, and help make this world a better place.
Watch the five takeaways
Each of these reflections is a short selfie video recorded during the trip. They are published across my channels this week:
- KubeCon Japan: from CPUs to GPUs — the AI workload as a different beast
- AI accelerators beyond Nvidia — TPUs, NPUs, and the diverse landscape
- Learn programming: a strong foundation matters — why seniors still win
- Developer career in 2026 — yes, it is still worth it
- Is it real or AI? — keep studying, stay hungry
Follow me for more, and I will see you at the next event. The cloud native community in Tokyo was, as always, brilliant — and the conversations are exactly what we need to shape the future of tomorrow.
#KubeCon #CNCF #CloudNative #Japan #AI
