The white paper diagnoses accurately but prescribes weakly. MEXT data acknowledges Japan's share of global AI top-conference papers fell from 6.2% in 2018 to 3.1% in 2024, with no domestic large model (except Sakana AI) ranking in LMSYS top 50. Yet the proposed solution—¥2 trillion over 5 years for 'national AI computing infrastructure'—replicates the EU's Gaia-X model while ignoring Japan's deeper talent fracture: 43% of University of Tokyo AI PhDs flow overseas or to finance rather than domestic tech firms. Without institutional transformation in equity incentive culture and failure tolerance, infrastructure investment becomes merely a dumping market for American chips.
The white paper's strategic value is underestimated. Its 'AI social implementation' framework explicitly positions generative AI as infrastructure solving super-aging society labor shortages rather than competing with America for AGI dominance. This pragmatic framing breeds unique vertical advantages: NEC and RIKEN's medical image generation model received pharmaceutical approval under JGA standards—the world's first clinical-grade generative AI. The ¥2 trillion compute investment mandates 30% allocation to 'Society 5.0' scenarios like manufacturing and quality control, avoiding resource misallocation from blindly chasing foundation models. On AI governance, the white paper's 'human-centered AI' principle offers greater technical flexibility than the EU AI Act, reducing compliance friction for Japanese firms' global expansion.
The white paper's genuine flaw is evading international supply chain dependency. Of the ¥2 trillion investment, 85% of GPU procurement still targets NVIDIA with only 15% allocated to domestic chip designers like Preferred Networks—a stark contrast to Japan's 1980s strategy of cultivating NEC and Toshiba's autonomous capabilities. More contradictory: the white paper emphasizes 'AI sovereignty' while adopting Sakana AI's 'model emergence' methodology as core innovation path, which fundamentally depends on continued access to foreign foundation models like Llama or Qwen. As geotechnological decoupling accelerates, this 'dependent autonomy' strategy accumulates risk. If America extends chip controls to model weight exports, Japan's 'social implementation' advantages collapse instantly.