Insights · 5 November 2026
Viewpoint

Microsoft didn't underspend on AI last quarter because it wanted to.

Microsoft didn't underspend on AI last quarter because it wanted to. It underspent because the GPUs didn't show up on time.

That one detail from this week's hyperscaler earnings says a lot. Alphabet and Amazon both expect capex to rise again in 2027, yet Microsoft's spend came in light because third-party GPU capacity slipped. The bottleneck isn't money anymore. It's delivery.

The same day, NVIDIA shipped CUDA 13.5 with general-availability support for Vera Rubin, which means the next platform is now ready for production. Meanwhile, Moonshot released open weights for Kimi K3.1 Pro: 2.8 trillion parameters and a 1M-token context window.

Put those together and the picture is clear. Models are getting bigger and more open, and silicon is getting faster. The hard part now is the stretch in between: getting servers specced, racked, cooled, validated and running real workloads on schedule.

At NexCore.TW Corporation, this is the work we focus on every day. In this cycle, the winners won't just be the companies that order the most GPUs. They'll be the ones that can get them into production fastest.

Is your AI roadmap limited by budget, or by deployment speed?

Originally shared on LinkedIn