U.S. AI infrastructure company Crusoe announced a pause on its 1.8 GW large-scale data center project in Wyoming, prompting the market to reassess the pace of AI infrastructure deal execution.

Article author and source: Wall Street Journal

News of a data center project being paused is prompting the market to reassess the pace of execution for U.S. AI infrastructure deals.

On June 10 local time, AI infrastructure company Crusoe announced the suspension of its 1.8 GW large-scale data center project in Cheyenne, Wyoming (codenamed "Project Jade") at the request of an undisclosed client.

Upon the announcement, Bloom Energy, which has supply chain ties to the project, saw its stock plunge more than 9% on the day, closing at $235.92, a daily decline of $23.69. The Nasdaq Composite and S&P 500 also declined that day, with the technology and AI sectors facing broad pressure.

Goldman Sachs partner Rich Privorotsky immediately highlighted this incident in internal communications, stating that the news “may have drawn attention” and directly pointing to a core vulnerability in the market: “In a market where everything is tied to AI capital expenditures, even minor delays, postponements, or shifts in priorities can be enough to force investors to reassess their assumptions about future demand.”

How large is this project? What does the halt mean?

What does 1.8 GW mean? One GW equals 1,000 megawatts—a scale typically associated with large power plants, not a single data center. Crusoe’s “Emerald Project” was originally planned for Cheyenne, Wyoming, and included 900 megawatts of “behind-the-meter” energy, specifically designed to bypass grid bottlenecks—demonstrating that the project involved specialized power arrangements, not a conventional undertaking.

According to Data Center Dynamics, Crusoe halted construction only after the client requested it. The identity of the client and the reason for the halt have not been disclosed.

Rich Privorotsky also acknowledged, “The project is only just beginning development, and customer identities have not been disclosed, so it would be incorrect to draw broad conclusions at this stage.”

But the issue is: the market doesn’t wait for conclusions. Stock prices have already begun to react.

Why is Bloom Energy the first to be affected?

Bloom Energy's steep decline is due to its stock price being built on a narrative heavily reliant on the realization of demand for AI data centers.

According to Investing.com, the Emerald Project is linked to a conditional power purchase agreement with AEP Energy, which in turn is tied to Bloom’s supply of solid oxide fuel cells. In its January SEC filing, AEP disclosed that its non-regulated subsidiary agreed to purchase the majority of solid oxide fuel cell options for approximately $2.65 billion and entered into a 20-year power purchase agreement with a highly investment-grade customer to supply power from a planned fuel cell power plant near Cheyenne.

Any loosening in this supply chain would immediately call Bloom’s revenue projections into question.

Bloom’s fundamentals are far from unsupported. In April this year, Bloom announced that Oracle plans to purchase up to 2.8 GW of fuel cell systems, with 1.2 GW already contracted and under deployment. Bloom’s revenue for the most recent quarter reached $751.1 million, a 130.4% year-over-year increase, with product revenue surging 208.4%. The company has raised its full-year 2026 revenue guidance to a range of $3.4 billion to $3.8 billion, representing approximately an 80% increase over the prior year at the midpoint.

But on Wednesday, these figures temporarily lost their appeal. The funds that had been hyping the expectation of inclusion in the S&P 500 the day before quickly withdrew, and market focus shifted back to a more realistic question: When will orders truly become revenue?

This script has already been performed once last year.

This is not the first time AI infrastructure stocks have plummeted due to "construction halts."

A similar market reaction occurred last year with CoreWeave. According to The Wall Street Journal, CoreWeave’s market capitalization dropped by $33 billion over six weeks, with its stock price falling 46%. One of the triggering factors was a delay in the construction of a large AI data center in Denton, Texas. Located north of Dallas, the construction site experienced approximately 60 days of delays due to heavy summer rains and strong winds, preventing contractors from pouring concrete.

The project is approximately 260 MW, and CoreWeave plans to lease it to OpenAI. Completion has been delayed by several months, with additional delays caused by adjustments to parts of the data center design.

Communication issues intensified pressure on the stock price. During the earnings call on November 10, CoreWeave CEO Michael Intrator initially stated that the issues were limited to “one data center”; later, the CFO corrected this, clarifying that the delays were concentrated with “one data center supplier.” The following day, Intrator repeated the “one data center” claim during an interview with CNBC, only correcting it after being prompted by the host. CoreWeave’s stock dropped 16.3% that day and continued to decline throughout December.

CoreWeave’s business model also makes investors more sensitive. It uses high-interest debt to purchase large quantities of NVIDIA’s advanced AI chips, deploys these chips in data centers, and rents out computing power to customers. Its clientele is concentrated among a few major clients, such as OpenAI, Microsoft, and Meta.

CoreWeave's revenue increased from $583 million in the same period last year to nearly $1.4 billion last quarter, but the company remains unprofitable, reporting a quarterly loss of $110 million. Gil Luria of DA Davidson described CoreWeave’s balance sheet as “the worst in tech” and noted its operating margin is approximately 4%, less than half of its debt interest costs.

CoreWeave’s example illustrates that AI infrastructure deals are not just about orders and client lists—they also depend on construction, power supply, financing, and delivery. Any delay in these areas can impact valuation.

Goldman Sachs warning: Cyclical risks are accumulating

Rich Privorotsky’s statement is not just explaining Bloom’s decline—he is pointing to a more systemic issue.

Current market momentum returns are at the 90th percentile over the past five years, with total exposure at the 99th percentile. As demand for leverage rises, funding spreads have widened, and retail participation through leveraged ETFs remains substantial.

Everything is increasingly tied to AI spending. It’s part of a hardware bull market, driving a portion of GDP and much of the market’s performance. The cyclical nature is becoming harder to ignore.

His assessment is: If anything disrupts the AI spending cycle, the vulnerability will become obvious. This risk has existed for some time, but the market is accelerating toward it.

This is also why the news of Project Jade’s suspension was amplified. The project has only just begun, and no clients have been disclosed; yet, market positions and narratives are already betting on AI capital expenditures continuing to expand. Any mention of “delay,” “pause,” or “priority shift” prompts investors to ask: Will future demand continue to accelerate, or will the pace of realization fall below valuation assumptions?

Privorotsky also broke down the uncertainty in the AI investment cycle into two interpretations.

On the bullish side, the entire pie continues to grow, requiring more edge computing, more data centers, more memory, more power, and more network capacity.

On the bearish side, many economically valuable tasks may already be runnable on existing hardware, and demand may materialize later than implied by current valuations.

He wrote that the debate is increasingly less about model quality and more about where reasoning ultimately takes place: on expensive, centralized cloud servers or on cheaper, more open, localized models.

This is the core of this round of adjustment. It’s not about any single project determining the AI cycle, but rather the market beginning to reassess whether AI capital expenditures, data center construction, power infrastructure, and corporate revenues can all continue to align at their previous pace.

Crusoe, meanwhile, is working to stabilize market sentiment. On Tuesday, the company announced that it has signed agreements totaling 4.9 GW for AI infrastructure, covering its data center projects and Crusoe Cloud business, with an overall development pipeline exceeding 40 GW. CEO Chase Lochmiller said, “Demand from leading global tech companies for AI infrastructure—rapid and large-scale—has never been stronger.”

However, against the backdrop of the "Jade Project" being suspended, the market has temporarily chosen to remain skeptical.