2025 年，美国将关税提高到近一个世纪来的最高水平，中国随即反制，全球政策不确定性上升，但全球增长仍预计为 3.2%，与一年前的预测一致。这种表面稳定掩盖了结构性损害。类似英国脱欧的经验表明，冲击往往滞后显现，英国十年后相对于原先路径损失约 6% 至 8% 的 GDP，说明关税与政策混乱的真实成本尚未完全体现。

关税影响暂未显现，部分原因在于实际关税约为宣布水平的一半，但即便如此仍达 14%。其拖累被两股力量抵消：一是 AI 投资与股市上涨支撑了美国增长，并带动台湾、韩国等 AI 出口经济体；二是财政政策更为扩张，尤其在德国和中国。同时，关税成本中约 95% 由美国企业承担，已使通胀上升 0.7 个百分点，并令美国家庭平均损失约 600 美元；若无关税，通胀本可接近 2% 的美联储目标。

脆弱性正在累积。AI 行业估值与回报的落差引发修正风险，订阅价格难以覆盖推理成本，存在类似互联网泡沫的回调可能。到 2026 年，随着提前进口效应消退、成本进一步向消费者转嫁，关税损害将更明显。中国持续依赖出口导向增长、欧洲改革不足以及美欧关系疏远，都加剧了碎片化。2026 年的关键在于是否纠偏，否则生活水平将普遍下滑。

In 2025, the United States raised tariffs to their highest level in almost a century, China retaliated, and global policy uncertainty intensified, yet global growth is still projected at 3.2%, exactly as forecast a year earlier. This apparent stability masks structural damage. Brexit showed that shocks surface with delay, with the UK estimated to have lost about 6% to 8% of GDP a decade later, indicating the true costs of tariffs and policy chaos have not fully appeared.

Tariff impacts have been muted partly because effective rates are about half of what was announced, but still around 14%. Two forces offset the drag: AI investment and an AI-driven stock surge supported US growth and exporters like Taiwan and South Korea, while fiscal policy turned more expansionary, especially in Germany and China. Meanwhile, about 95% of tariff costs have been absorbed by US firms, adding 0.7 percentage points to inflation and making the average US household roughly $600 poorer; without tariffs, inflation could have been near the Federal Reserve’s 2% target.

Fragilities are accumulating. A widening gap between AI valuations and returns raises the risk of a dotcom-style correction as subscription prices struggle to cover inference costs. In 2026, as frontloaded imports fade and firms pass through more costs, tariff damage will become clearer. China’s export-led model, Europe’s reform gaps, and US-EU estrangement add to fragmentation. The central question is whether 2026 brings course correction; otherwise, living standards will decline broadly.