Amazon Cloud Warehouses Billions of Dollars of Nvidia GB300 Racks Due to Power Shortage

According to internal Amazon documents and a recording of an employee meeting recently obtained by Business Insider, Amazon's cloud computing division AWS is temporarily storing Nvidia GB300 racks in an "intermediate warehouse" because the target data center "has room for more racks but not enough available power to turn them on."

Internal emails show that Amazon expected to receive about $5 billion worth of GB300 racks in August alone, and will continue to receive about $3 billion worth each month thereafter, maintaining that pace "for the foreseeable future."

At an internal employee meeting in September, AWS Chief Financial Officer John Felton joked to Prasad Kalyanaraman, the executive responsible for AI infrastructure construction: "You turned our company's free cash flow negative. Congratulations." Kalyanaraman did not dispute this, and said the company's free cash flow is expected to remain negative for "the next few years or so."

In July, when Amazon announced its second-quarter earnings, its trailing free cash flow had already turned clearly negative, recording -$7.6 billion, the first time in years.

The impact of this situation on the market is already apparent. Chips sitting idle in warehouses cannot produce any computing capacity, causing online computing power to remain scarce and GPU rental prices to stay high, forming a classic inflationary bottleneck—large amounts of capital, labor and energy are consumed in advance, while output is delayed by months or even years.

"Early Rack Receipt": the chip scramble forces payment upfront

Amazon's previous usual practice was to pay only after equipment was actually installed in a data center, with suppliers bearing the financing cost of equipment in transit. However, internal documents show that Amazon has introduced a new process called "Early Rack Receipt," taking ownership before the racks arrive at the data center.

Media interpretation suggests that the competition for Nvidia GPUs is so intense that Amazon must lock in hardware before it even has an installation location, to avoid competitors seizing it. This means Amazon now pays first and powers up later; any installation delay will postpone the point at which these assets begin generating revenue, while the cash has already flowed out.

The report also noted that at least one AWS supply chain team had no "clear forecast path" for "how to get racks into the data center." In response, an Amazon spokesperson said the company's usual practice is to temporarily store hardware near deployment sites while waiting for power to be ready, calling this "a best business practice."

Research firm Bernstein concluded that power permits, cooling systems, electrical equipment, construction and hardware each advance at different paces, making it "difficult to align the timing perfectly."

Power bottleneck: the core constraint on data center construction

The root cause of the chip backlog lies in the severe lag in power infrastructure. Kalyanaraman revealed at the internal meeting that in West Texas, due to a shortage of plumbers and electricians, Amazon had to fund the construction of worker housing so that construction workers could live on site for the two to three years before the project was completed.

According to Cleanview Research data, Amazon currently has 44 gigawatts of data center capacity under development across 16 U.S. states, more than three times its existing operating scale of 13.9 gigawatts. And that 13.9 gigawatts is already the largest in the industry, 69% more than Microsoft and about twice that of Google and Meta.

AWS even has a dedicated internal dashboard called "Secured Power at Risk," tracking transmission bottlenecks, curtailment requirements and "use it or lose it" clauses—if a project misses milestone deadlines, Amazon may lose grid allocations it has already obtained.

Kalyanaraman also revealed that in 2023 Amazon promised CEO Andy Jassy that AWS would emerge from capacity constraints by 2027. Now, Amazon expects demand in 2027 to still exceed available capacity.

Chips pile up in warehouses, yet GPU rents rise instead

By conventional logic, a large backlog of chips in warehouses should push down computing prices. But the opposite is happening.

The reason is simple: GPUs sitting in crates produce zero GPU hours. What is truly scarce is "powered computing capacity," and its price keeps climbing.

Rich Privorotsky, head of Goldman Sachs' Delta-One trading desk, noted in an internal report on October 16:

"Relative to the available supply of investable powered compute, there is too much capital looking for a home. Repricing is a process the market will continue to go through, sometimes gradual, sometimes violent, and it could evolve into a systemic risk. The interesting contradiction is that spot compute pricing remains extremely strong. B200, H200, H100, and even the older A100 all maintain considerable rental rates. Compute that theoretically should depreciate quickly is still holding its value. The scarcity of available compute is undeniable, and this is structurally positive for those who already hold compute."

Goldman Sachs-tracked Silicon Data B200 spot rental index (non-hyperscale cloud) stands at $6.01 per GPU hour, close to the March record of $6.11 and up about 40% from the January low of $4.29.

$8 billion chip sale: balance sheet relief or something else

The above background provides a new interpretive lens for a recent high-profile Amazon deal. According to a Financial Times report on October 2, Amazon is seeking to sell about $8 billion of Nvidia Grace Blackwell chips to outside investors through a special purpose vehicle (SPV), and then lease the chips back for its own use. The SPV is intended to be financed with debt and to sell up to 10% equity; Amazon will hold no stake in the entity, and the stated goal is a more "asset-light" balance sheet.

However, when this transaction is set against internal documents, the timeline is rather intriguing: in August, Amazon launched an "early receipt" process and took delivery of about $5 billion of GB300 racks that month, followed by about $3 billion per month thereafter; in September, AWS's chief financial officer confirmed at an internal meeting that the company's free cash flow had turned negative; and on October 2, Amazon went out to seek buyers for $8 billion of chips. The sum of the expected delivery amounts for August and September happens to equal the size of the SPV.

Sources cited by the Financial Times said the chips involved in the SPV have already been deployed in more than a dozen data centers across five U.S. states, and the report did not specify whether they are GB300. Either way, investors buying into the SPV are in effect purchasing a steadily depreciating asset—Grace Blackwell is about to be superseded by Nvidia's next-generation Vera Rubin architecture, and every month a chip sits idle in a warehouse means one month of its five-year economic life is consumed without generating any revenue.

Who is paying interest on this pile of gathering-dust chips

Microsoft CEO Satya Nadella came close to publicly acknowledging this problem last November, saying: "The biggest problem we have now is not compute surplus, it's power," and admitting that Microsoft "may actually have a bunch of chips sitting in inventory that I can't plug in."

By April of this year, about half of the U.S. data centers originally planned to come online in 2026 faced cancellation or delay; of the 16 gigawatts planned, only about 5 gigawatts were actually under construction.

According to media accounts, Kalyanaraman's own principle is: "If we can spend capital as late as possible, that's better for us." The AI arms race has completely overturned this logic: Amazon is now rushing to spend money as early as possible, buying chips it cannot use, simply to keep competitors from taking them first—and if competitors get them, chances are they will mostly sit unused as well.

Extrapolating this logic to every hyperscaler and emerging cloud service provider competing for the same batch of Nvidia allocations yields a warehouse-idling problem measured in tens of billions of dollars. Most of this inventory is likely booked under "construction in progress," and before it is formally put into use it does not even enter depreciation calculations, leaving no trace on the financial statements...

From a broader perspective, this AI infrastructure buildout boom is increasingly reliant on debt financing. Every month a chip sits unpowered means a month of interest paid on an asset with zero return. At the same time, compute rental rates keep climbing, and the cost ultimately passes through to every AI lab, enterprise, and startup that rents GPUs.