The "narrative shift" of American AI data centers: from the excitement of "big deals" to the "blame game" of "constant delays"

The market sentiment in the U.S. artificial intelligence data center sector is experiencing a dramatic reversal. As unprecedented server cluster construction encounters real-world obstacles, the excitement previously generated by gigawatt (GW) scale transactions and record contracts is fading, replaced by frequent project delays and the ensuing blame-shifting.

In this competition for computing power, the most significant dynamic is that supply chain tensions have begun to substantially impact corporate performance. AI cloud service provider CoreWeave has warned investors that its revenue this quarter may face a write-down of $100 million to $200 million due to delays caused by third-party developers. This signal indicates that the complexity of data center construction is dragging billions of dollars in projects into a quagmire of delays, leading stakeholders to start blaming each other in an attempt to define responsibility as timelines slip.

This friction is not limited to individual companies but exposes the systemic bottlenecks facing the entire industry. Currently, as GPU delivery speeds have outpaced facility construction speeds, some companies are even forced to leave expensive hardware idle in warehouses awaiting deployment. Meanwhile, physical limitations from power acquisition to equipment delivery are colliding with customers' aggressive demands.

As major buyers like the "Fab Five" (OpenAI, Google, Meta, Anthropic, and xAI) intensify their urgency for computing power delivery, delays at the infrastructure level are reshaping market expectations. Investors and industry participants are gradually realizing that AI infrastructure construction is transitioning from an early capital frenzy into a turbulent period filled with execution risks and "blame games."

Revenue Warning Triggers "Responsibility Attribution" Dispute

As a supplier to giants like Microsoft and OpenAI, AI cloud service provider CoreWeave has become a typical case in this "blame game." CoreWeave CEO Mike Intrator admitted earlier this month that due to "delays caused by third-party data center developers," the company's revenue for the current quarter will be severely impacted. Although Intrator did not name names directly, there is widespread speculation in the industry that the developer in question is its major partner, Core Scientific.

This speculation is not unfounded. According to insiders, Microsoft reduced some contracts with CoreWeave eight months ago due to delays at a data center in Denton, Texas, where the power supply is managed by Core Scientific.

In February of this year, Core Scientific stated during an earnings call that the completion timeline for a certain data center project would be pushed back from 2025 to early 2026, likely referring to the aforementioned facility. Although OpenAI subsequently intervened and signed a $12 billion contract to lease servers at that facility from CoreWeave, the tension brought about by the delays remains palpable.

In response to the accusations of delays, Core Scientific CEO Adam Sullivan did not directly address specific projects but sharply criticized the current state of the industry. He pointed out that the timelines for many AI data centers are "unrealistic," unless developers have already secured long-lead equipment such as generators and skilled contractors in advance Sullivan stated to the media that when a listed company discloses delays in advance while the other party waits until the last moment to announce, it creates confusion and erodes market confidence. Additionally, Core Scientific shareholders previously voted against CoreWeave's $9 billion acquisition proposal, which may also be one of the factors contributing to the strained relationship between the two parties.

High-Stakes Game Under Slim Profits

Although project delays are common in the construction industry, the stakes are entirely different in the current AI computing power race. To meet the delivery pressures from clients like OpenAI, Oracle executives expressed their dissatisfaction loudly to contractors at a construction site in Abilene, Texas earlier this year.

This anxiety stems from stringent financial terms. It is understood that contracts between cloud service providers and clients often include punitive clauses: if the provider misses deadlines or if server downtime due to failures occurs, clients can reduce payments. For the GPU cloud leasing business, which already has slim profit margins, these operational issues can have a substantial negative impact on financial results.

This explains why all parties are so eager to find someone to blame when billion-dollar projects miss deadlines by even a few weeks or months. For companies that have committed to rapid delivery timelines, the race to launch NVIDIA GPU clusters remains a severe challenge.

Hardware Backlog and Strategic Adjustments

As power supply becomes increasingly scarce, delays in construction progress have led to a mismatch between hardware and facilities supply and demand. Several developers have revealed that the shipping speed of GPUs currently far exceeds the construction speed of data centers, forcing some companies to store rows of idle GPUs in warehouses, waiting for instructions on where to send them.

In light of this reality, large tech companies are adjusting their strategies to build buffers. Meta's Chief Financial Officer Susan Li acknowledged this tension during the earnings call in late October, stating that the company is "phasing the construction of data center sites," meaning they are preparing all facilities except GPU racks in advance so that capacity can be quickly ramped up when needed in the future.

The fact that even large developers like Meta must establish buffer mechanisms highlights the current industry's predicament: the physical limits of labor, equipment, utilities, and contractor bandwidth are colliding head-on with clients' endless demands. As acquiring power becomes more challenging, more clients may choose to collaborate with multiple data center providers to hedge risks in the future