Japan HP's Hiroshi Katsuya: don't chase 100% local AI

Japan HP ran the HP Partner Communication 2026 conference in Tokyo on July 16, 2026, where the company pushed partners to shift from selling AI products to embedding AI into customer operations. The backdrop Katsuya cited is labor shortage, with hiring alone unable to fill gaps. He split the AI workforce into three roles — assistants, specialists, and managers — and mapped adoption into three stages: chat use, internal data use, and automation with AI agents.

Katsuya's own practice shows the split is not binary. He uses cloud AI for public information, but keeps non-exportable business data fully local. He also draws a line between running an app on a PC and running inference there, noting that OpenAI's Codex app operates on the PC while inference happens in the cloud. His local experiments with the open-source agent OpenClaw ran into processing stalls, so he still chooses Codex in some situations.

Hardware follow this logic. For individual research he favors the small HP ZGX Nano G1n AI Station, while departments may need workstations. He said AI agents will make memory, SSD, and CPU as critical as GPUs, since any shortfall becomes the bottleneck. Cost and continuity are the other hinge: he said cloud can be shut off, while partners can go local for token savings, though the choice is a corporate judgment — and his own approval step shows humans can become the bottleneck.