DGX Spark tops Kakaku.com ranking as local LLM boom hits

The local LLM boom rests on three converging foundations: model quality, hardware, and software. Chinese developers like DeepSeek, which shocked the market in January 2025 with R1, and Alibaba's Qwen family have pushed open-weight models to rival closed US frontier models. Qwen3.8-27B, released in August 2026, is said to match Anthropic's Claude Opus 4.6, while remaining small enough to run on consumer GPUs or even smartphones.

This maturity comes amid growing distrust of cloud dependencies. The June export restriction that cut off Claude Fable 5 worldwide, and the risk of Google account bans affecting Gmail, showed users that cloud AI can be revoked abruptly. Local models offer permanence—once downloaded, they cannot be taken away.

The ranking of a ¥1.13 million DGX Spark on Kakaku.com signals that consumers are willing to invest in dedicated AI hardware. What remains to be seen is whether local models can sustain their performance gains, especially as US companies like Google, NVIDIA, and Meta also release smaller models, potentially reshaping the competitive landscape.