To The Moon

Global AI infrastructure is running into the hard walls of grid congestion and community resistance. BlackRock CEO Larry Fink has issued a contrarian warning, stressing that slowing down computing power construction will directly push inflation higher.

The CEO of BlackRock, the world's largest asset manager, warned on Wednesday that a slowdown in artificial intelligence data center construction could further drive up computing costs, intensify inflationary pressure, and threaten economic growth. According to Bloomberg, Fink said during the Global Dialogue conference in Berlin, Germany, that computing costs remain high, and failing to continuously build new data centers could instead create greater inflationary pressure. Fink believes that accelerating data center construction can increase market computing supply and thereby lower computing costs more quickly. He further warned that slowing the expansion of related technological infrastructure could threaten economic growth prospects. These remarks come as global AI investment faces multiple challenges, including rising financing costs, tight electricity supply, and opposition from local communities. Some data center projects have already been delayed, and investors have become more cautious about the ability of large technology companies to sustain capital expenditure expansion.

Power Supply Becomes the Main Bottleneck

Fink had previously downplayed concerns about a bubble in the artificial intelligence industry, arguing that the market needs to focus more on the shortage of computing supply. He noted that AI demand growth has exceeded previous expectations, and slow infrastructure expansion could limit access to related technology services for enterprises and users. The latest industry research shows that the pace of power infrastructure construction has become an important factor constraining data center expansion. According to a Goldman Sachs research report disclosed by Axios on Tuesday, global data center electricity demand is expected to grow by 170% between 2025 and 2030, with more than 60% of the increase coming from the United States. Jason Tofsky, Goldman Sachs' global head of digital infrastructure investment banking, believes that electricity may become the main long-term bottleneck for AI infrastructure. The bank noted that about 2,600 gigawatts of proposed power generation and energy storage projects in the United States are currently waiting to connect to the grid, more than twice the existing installed capacity, but many of these projects may ultimately fail to materialize. From applying for grid connection to achieving commercial operation, the median waiting time for related projects is about five years. The gap between the construction cycle of power supply and the pace of AI technology development exposes data center operators to risks of construction delays and extended investment payback periods.

Project Delays Have Affected Corporate Performance and Investment Arrangements

Infrastructure bottlenecks are translating into real operational pressure for companies. Australian data center operator Nextdc said on Wednesday that power supply for its Melbourne M3 data center project has been delayed due to local residents' opposition to the construction of related power facilities. According to Bloomberg, Nextdc therefore cut its EBITDA guidance for the fiscal year ending June 2027 by more than one-third to AUD 245 million to AUD 265 million, and announced the suspension of all new development projects and land acquisitions. The company is also assessing whether it can invoke force majeure clauses in relevant contracts. After the news was announced, Nextdc shares plunged by about 45% at one point during Sydney trading hours. Similar resistance has also appeared in the United States. Oracle's data center project in New Mexico faces legal challenges over power facilities, air quality, and water resource permits, affecting some construction arrangements. Bloomberg, citing Data Center Watch data, reported that in the first half of this year, at least 106 data center projects in the United States were blocked or delayed due to local opposition, involving planned investment of about USD 198 billion. U.S. President Trump also publicly addressed community resistance to data center construction on Wednesday. According to CNBC, Trump called on data center operators to reach economic compensation arrangements with local communities, including paying community or resident dividends, to win support needed for project construction. He also said that large data centers should build their own power generation facilities and sell or transmit surplus electricity to local communities and the grid to reduce the impact of projects on residents' electricity costs.

Financial Institutions Have Mixed Views on Financing Risks

Although Fink emphasized the need to accelerate infrastructure construction, financial markets remain divided on the investment returns and financing risks of AI projects. Goldman Sachs estimates that from 2026 to 2031, global capital expenditure related to computing power, data centers, and power infrastructure could total USD 7.6 trillion. A large amount of construction funding needs to be raised through debt markets. Goldman Sachs estimates that global hyperscale cloud service providers' bond issuance could reach USD 420 billion in 2027, an increase of more than 60% from the forecast level for 2026, with debt financing expected to account for about 35% of capital expenditure. The rapid expansion of financing scale has also made investors more focused on project construction progress, future revenue, and debt repayment capacity. HSBC Holdings CEO Georges Elhedery said in a CNBC interview on Tuesday that artificial intelligence has transformative potential, but investors must carefully select data center projects, especially evaluating the quality of tenant agreements and contract pricing. He pointed out that not all data center projects are equally attractive, and the commercial terms of some projects may make financing and investment more challenging. Citadel founder Ken Griffin said that investors are almost certainly in an AI bubble, partly because many companies are still trying to figure out the reasonable scale of AI investment and its actual economic value. Fink's remarks this time focused attention on the speed of computing supply expansion. In his view, delays in data center construction could leave the market facing high computing costs for a long time and limit the economic benefits brought by artificial intelligence technology. As grid connection, project permits, and financing conditions gradually become real constraints on data center construction, whether related companies can complete infrastructure investment as planned is drawing joint attention from investors and policymakers.