围绕 AI 的增长预期出现巨大偏差。部分预测声称 AI 可使全球产出每年提升 30%，但主流经济学家仅预计约 1% 的年均增幅，而且麦肯锡调查显示约 80% 使用生成式 AI 的企业表示其对盈利“无实质影响”。与此同时，资本正加速集中于少数公司，例如 Nvidia。AI 投资带来安全与社会风险：模型错误率高于人类，可协助恐怖主义、伤害弱势群体、监视组织，且俄朝中黑客已利用 AI 强化勒索攻击，使其更难阻断。

为分摊成本，OpenAI、Microsoft、Nvidia 等公司正推动政府加入“国家级 AI 基建”投资。10 月下半月，Nvidia 的黄仁勋访问韩国，与政府与三星、SK 海力士讨论 AI 基础设施；Anthropic CEO Amodei 访问东京，与日本首相会面、与 AI Safety Institute 签署合作并开设东京办公室；OpenAI 则发布《AI in Japan》报告，提出可为日本经济创造超 100 万亿日元（6,380 亿美元）的价值，并以三大支柱（社会基础、数据中心投资、教育）概括所谓“国家战略”。但这些方案基于理论推演而非现实成果，也与日本传统的审慎增量式创新不符。

AI 扩张面临现实约束，尤其是能源。数据中心耗能巨大，日本若提升容量可能需重启核电；人口老龄化也使数字化难以快速推进。Altman 向美国政府要求建设大规模能源产能，声称首个 1 万亿美元 AI 基建可在三年内额外带来 5% GDP 增长。另有提案建议美国每年新增 100 吉瓦能源以避免在 AI 时代落后中国。更深层问题是：若 AI 取代劳动、数字企业不承担社会责任，税基与消费将缩减、出生率与认知健康将恶化。真正的 AI 成功应衡量其减少琐事、提升效率与释放人类创造力，而非依靠夸大的增长承诺。

Forecasts for AI’s economic impact vary wildly. Some claim AI could boost global output by 30% annually, yet most economists expect roughly 1% yearly growth at best, and a McKinsey survey found around 80% of companies using generative AI saw no material effect on profitability. Nevertheless, capital continues flooding into a few firms, notably Nvidia. AI systems remain error-prone, can aid terrorism, harm vulnerable users, and enable corporate or personal surveillance, while Russian, North Korean, and Chinese hackers have incorporated AI into ransomware to make attacks harder to defeat.

To offset soaring costs, OpenAI, Microsoft, Nvidia and others are urging governments to co-fund national AI infrastructure. In late October, Nvidia CEO Jensen Huang coordinated with the South Korean government and major firms; Anthropic’s CEO met Japan’s prime minister, signed cooperation agreements, and opened a Tokyo office; and OpenAI published “AI in Japan,” projecting over ¥100 trillion ($638 billion) in added economic value built on social infrastructure, data centers, and education. Yet these proposed “national strategies” rely on theoretical gains rather than proven outcomes and clash with Japan’s incremental, needs-driven innovation model.

AI expansion faces physical constraints, especially energy demand. Data centers consume vast electricity, implying Japan may need nuclear restarts to scale capacity; rapid digitalization also risks backlash in an aging society. Sam Altman urged the US to build massive new energy capacity, arguing the first $1 trillion in AI infrastructure could add 5% GDP growth over three years, while some proposals call for 100 gigawatts of new US capacity annually to keep pace with China. Deeper issues remain: if AI displaces labor and digital firms evade social obligations, tax bases, consumption, fertility, and cognitive health will erode. True success should be measured by reducing drudgery, eliminating unproductive work, and expanding human creativity—not inflated growth promises.