科技巨头正对生成式 AI 下下巨额资本赌注，把他们的支出规模比作赌徒把全部身家押在单一轮盘数字上。4 月底提交的文件显示，全球最大的 4 家科技公司表示，他们计划今年合计投入 7250 亿美元作为资本支出，而 OpenAI 执行长 Sam Altman 则把公司到 2030 年的资料中心预算定在 6000 亿美元。文章认为，这并不只是纯粹的投机：训练前沿模型并让它们持续运行需要庞大的运算能力，而从免费的消费者聊天机器人转向更商业化的 AI 代理与程式设计工具，也重新强化了这种投资最终可能会回本的感觉。

AI 热潮的证据正在市场与企业预算中显现。Menlo Ventures 表示，企业在 2025 年于 AI 上花了 370 亿美元，而投资人也因此受惠，因为科技股约占 S&P 500 的 1/3；过去一年 Nasdaq 上涨约 40%，主要由 AI 相关股票带动。但这波涨势看起来很脆弱：Meta Platforms 在披露沉重的 AI 支出后股价下跌，而且有些融资愈来愈像循环操作，例如 Nvidia 投资 OpenAI，而 OpenAI 再大量购买 Nvidia 的产品。如果 OpenAI 和 Anthropic 早在今年秋天就上市，市场对这些公司的曝险还可能进一步加深，两者合计估值或许接近 1.9 兆美元。

最大的问题是新创公司的营收能否足够快地成长，以证明这些支出合理。OpenAI 表示其年营收约为 250 亿美元，目标是在 2030 年达到 2800 亿美元，但它必须把企业使用扩展到销售、金融、医疗、HR、物流、制造与法律工作；《华尔街日报》的一篇报导称，它未达成 2025 年的营收与使用者目标，尽管 OpenAI 财务长 Sarah Friar 表示，公司整体上正高于计划进度。Anthropic 已在 6 个月内把年化营收从 90 亿美元提高到 300 亿美元，但许多受监管产业仍保持谨慎，保险公司也可能排除与 AI 相关的损失，而像 CoreWeave 与 Nscale 这类 neocloud 公司若需求降温也可能变得脆弱。文章最后把当下比作 1999 年的网路泡沫：繁荣期间建起的基础设施最终证明有用，但许多早期投资者却血本无归；而 AI 只有在尽管面对今天的警讯，仍能在医疗、教育、能源与就业上带来广泛收益时，才可算是真正的成功。

Tech giants are making enormous capital bets on generative AI, comparing the scale of their spending to gamblers staking everything on a single roulette number. In late April filings, 4 of the world’s largest tech companies said they plan to spend a combined $725 billion on capital expenditures this year, while OpenAI CEO Sam Altman has put his company’s data center budget through 2030 at $600 billion. The article argues that this is not mere speculation: training frontier models and keeping them running requires vast computing power, and the shift from free consumer chatbots to more commercial AI agents and coding tools has renewed the sense that the investment may eventually pay off.

Evidence of the AI boom is showing up in markets and corporate budgets. Menlo Ventures says businesses spent $37 billion on AI in 2025, and investors have benefited because tech stocks make up about 1/3 of the S&P 500; the Nasdaq has risen about 40% over the past year, largely on AI-related shares. But the rally looks fragile: Meta Platforms’ stock fell after it disclosed heavy AI spending, and some financing is increasingly circular, such as Nvidia investing in OpenAI and OpenAI then spending heavily on Nvidia products. The market could deepen its exposure further if OpenAI and Anthropic go public as early as this fall, with combined valuations potentially approaching $1.9 trillion.

The biggest question is whether startup revenues can grow fast enough to justify the spending. OpenAI says it generates about $25 billion in annual revenue and aims for $280 billion by 2030, but it must expand enterprise use into sales, finance, healthcare, HR, logistics, manufacturing and legal work; a Wall Street Journal report said it missed 2025 revenue and user targets, though OpenAI CFO Sarah Friar said the company is beating its plan at a high level. Anthropic has raised its annualized revenue to $30 billion from $9 billion in 6 months, but many regulated industries remain cautious, insurers may exclude AI-related losses, and neocloud firms such as CoreWeave and Nscale could be vulnerable if demand cools. The article closes by likening the moment to the 1999 dot-com bubble: infrastructure built during the boom eventually proved useful, but many early investors were wiped out, and AI may only count as a true success if it delivers broad gains in healthcare, education, energy, and jobs despite today’s warning signs.