America's AI is advancing by leaps and bounds, yet its bubble risks have already sparked government concerns

The U.S. AI industry in 2025 is witnessing a feverish feast where capital and technology intertwine: from OpenAI’s valuation surging to $300 billion, to tech giants scrambling to pour hundreds of billions of dollars into expanding computing infrastructure, and the government rolling out national-level support initiatives such as the Genesis Mission, the entire sector is displaying a roaring momentum of expansion. Nevertheless, beneath the prosperous facade, hidden worries including inflated valuations, ambiguous profitability, and sky-high leverage are accumulating steadily. Warnings about an AI bubble are growing louder on Wall Street, leaving the industry standing at a crossroads between opportunities and risks.

Since the start of this year, the investment moves of industry giants have repeatedly set new records: OpenAI, in collaboration with SoftBank and Oracle, launched the $500 billion Stargate AI infrastructure project, aiming to build a computing cluster with a total capacity of nearly 7,000 megawatts; Google announced a $25 billion investment over two years to deploy data centers across 13 states, while Meta also finalized a multi-hundred-billion-dollar plan to construct AI-dedicated data centers. Fueled by capital, the valuations of core enterprises have soared. Through a new round of $8.3 billion financing, OpenAI is advancing its $40 billion fundraising target, and NVIDIA’s market capitalization once exceeded $5 trillion—surpassing the entire annual GDP of Germany. On the policy front, the Trump administration has intensively issued artificial intelligence action plans, signed multiple executive orders to accelerate data center construction, and even repealed several regulatory rules to "unshackle" AI development, wagering the nation’s resources on the AI track.

Behind this frenzied expansion, concerns about bubble formation have emerged as an inescapable shadow. The most prominent issue is the severe disconnect between valuations and actual performance. NVIDIA’s stock price has surged approximately 11-fold since November 2022, far outpacing the S&P 500 index. However, profitability struggles among downstream enterprises have gradually come to light: Oracle reported a $100 million quarterly loss due to NVIDIA chip leases, with its AI cloud services posting a mere 14% gross profit margin. Even more alarming is the widespread "circular financing" model within the industry. NVIDIA invests in OpenAI, OpenAI purchases computing power from Oracle, and Oracle in turn buys chips from NVIDIA. This self-reinforcing "foot-on-foot" operation continuously inflates book value without genuine cash flow support.

Unclear profit paths and overinvestment have further exacerbated risks. According to Goldman Sachs data, U.S. tech giants are projected to spend $1.4 trillion on AI infrastructure capital expenditures between 2025 and 2027, yet returns remain nowhere in sight. A McKinsey survey indicates that nearly 80% of enterprises deploying AI have failed to boost net profits, and 95% of generative AI pilot projects yield no direct financial returns. While OpenAI recorded $4.3 billion in revenue in the first half of the year, its net loss skyrocketed to $13.5 billion, with the third-quarter loss alone exceeding $11.5 billion. Massive investments have driven tech companies to shift from asset-light to asset-heavy business models. By the end of September, U.S. tech firms’ bond financing reached $157 billion, a year-on-year increase of 70%, and high-leverage operations have planted significant credit risks.

Market opinions are sharply divided on whether an AI bubble exists. Proponents such as Jensen Huang emphasize that current demand for AI computing power is genuinely robust, with the six major cloud computing giants’ capital expenditures set to rise to $632 billion by 2027—a stark contrast to the idle "dark fiber" during the internet era. However, actions such as short-selling by "big shorts" on AI star stocks and reductions in NVIDIA holdings by institutions like Bridgewater Associates highlight growing concerns about an overheated market. Figures including Bill Gates and IMF Managing Director Kristalina Georgieva have warned that the current AI investment boom mirrors the late-1990s internet bubble, cautioning that a large portion of investments may ultimately turn into bad debts.

Industry insiders widely agree that a dialectical perspective is needed to weigh AI’s long-term value against short-term risks. There is no doubt that AI will revolutionarily enhance productivity, but unlocking industrial value requires an extended period. Historical experience shows that a bubble burst is not entirely negative: the low-cost infrastructure left in the wake of the internet bubble laid the groundwork for the subsequent digital economy boom. The core challenge facing the current U.S. AI industry lies in whether capital speculation will crowd out resources for technological iteration and application implementation. Striking a balance between capital enthusiasm and industrial rationality, and enabling AI technology to truly empower the real economy, has become the key to the sustainable development of the U.S. AI sector.