AI Hardware Buildout Hits Power Bottlenecks
Amazon Web Services (AWS) is accelerating its procurement of advanced artificial intelligence hardware, specifically Nvidia GPU racks. However, the company is reportedly encountering significant infrastructure challenges, including power shortages, which are causing substantial amounts of recently purchased equipment to sit idle.
Internal company communications reviewed by journalists indicated that AWS expected to receive approximately $5 billion worth of Nvidia’s GB300 server racks at U.S. facilities prior to final deployment. Furthermore, the company projected acquiring an additional $3 billion worth of GB300 racks every month for the foreseeable future.
The Challenge of “Dark GPUs” and Infrastructure Strain
Industry sources reported that while some AWS sites possess physical space for additional racks, they lack the necessary electrical capacity to activate the equipment. Consequently, completed hardware is being temporarily stored and staged at intermediary warehouses until the facilities can be brought online.
A key concern noted in internal emails was that, at the time of the planning, some AWS teams did not have a “clear forecasted path” for integrating the incoming racks into active data center operations.
This scarcity of available electricity capacity represents one of several major constraints slowing the development of AI data centers, leading to a phenomenon termed “dark GPUs.” This refers to expensive AI chips and servers that have been purchased but remain unusable due to insufficient power supply or other technical limitations. The requirement for massive upfront spending to secure components like Nvidia GPUs creates financial pressure, as early investments in equipment can tie up vast amounts of capital if the data centers are not yet operational.
Shifting Investment and Capital Expenditures
AWS’s aggressive expansion is driving significant financial commitments. Amazon is expected to spend roughly $220 billion in capital expenditures this year, a figure exceeding that of any other major technology competitor.
This intense spending has impacted cash flow, as the company’s free-cash-flow turned negative in the last quarter, marking the first time since 2023. To manage this scale of spending, AWS has altered its acquisition strategy. Previously, the company only assumed ownership of a rack after it was delivered to a data center. Now, under a new “Early Rack Receipt” process, AWS can pay for and formally record certain racks much earlier in the supply chain.
Amazon’s Position and CEO Outlook
When questioned about the stored equipment, an Amazon spokesperson stated that the company routinely stages finished hardware near deployment sites for installation once capacity becomes available, characterizing this practice as “best business practice.” The spokesperson also disputed the notion that the broader company lacked a plan to move racks into data centers, noting that Amazon maintains established systems for tracking infrastructure assets.
In contrast, Amazon CEO Andy Jassy has maintained a highly optimistic outlook regarding the company’s AI investments. Jassy previously stated that Amazon has a “clear line of sight” to strong returns from its AI investments, describing the technology as a “once in a lifetime opportunity.” In the annual shareholder letter, Jassy wrote, “We’ve been through this cycle with the first big AWS growth wave, and liked the results… We expect to feel similarly about this next wave.”
