Amazon plans to take Nvidia racks before AWS sites can power them
Amazon plans to temporarily store Nvidia GB300 server racks because some data centres lack the electrical capacity to operate them. The plan would make Amazon pay for the hardware earlier, before it can use it.
AI-assisted synthesis · Based on 5 outlets' reporting · Article text read from 3 of 5 outletsHow it's made
View timeline · 7 entries
File photoWhat happened
Key facts
- Some AWS sites have space for additional server racks but lack the electrical capacity to operate them (per computerbase.de, Business Insider, newsquawk.com).
- Amazon has typically taken ownership of server racks when they were installed; the described approach would transfer ownership earlier, before use (per computerbase.de).
- One email said at least one AWS team lacked a clear forecasted path for installing the incoming racks (per Business Insider, computerbase.de, newsquawk.com).
The brief
Amazon plans to take delivery of Nvidia GB300 server racks at U.S. facilities before those racks are ready for final deployment, according to August internal emails reviewed by Business Insider.
The approach would move Amazon’s ownership and payment point earlier than its usual practice of taking ownership when racks are installed, leaving the company to pay for equipment before it can put it to use. The operational problem described in the reporting is a mismatch between incoming hardware and data-centre readiness.
Some AWS sites have space for more racks but lack the electrical capacity to run them. Business Insider reported that at least one AWS team did not have a clear forecast for how it would get the incoming racks into data centres. The financial scale is substantial.
August emails projected roughly $5 billion in GB300 racks for AWS facilities, followed by about $3 billion in additional racks each month for the foreseeable future, according to one email reviewed by Business Insider. Those figures are projections in internal correspondence, not confirmed records of deliveries completed.
The process is called Early Rack Receipt in the August emails. ComputerBase also described Amazon’s plan as storing some racks until sites are ready, and explained the change against the company’s usual practice of taking ownership at installation.
Newsquawk’s live-feed item independently reported the roughly $5 billion expected receipt at U.S. facilities before final deployment, while the headlines from Cryptopolitan and Yahoo Finance likewise characterized the hardware as arriving before power or deployment readiness.
The reports point to an infrastructure constraint rather than a lack of physical room at every site: some AWS locations can accommodate more equipment, but do not yet have enough electrical capacity to operate it.
The early-receipt approach shifts the timing of payment and ownership, while the installation issue remains with the teams responsible for getting the racks into data centres. Amazon expects about $220 billion in capital expenditures this year, more than any other Big Tech peer, Business Insider reported.
The projected rack purchases therefore sit within a much larger investment program, though the reporting does not assign a specific share of that annual spending to the early-receipt arrangement. The emails describe planned deliveries and an installation bottleneck, not the eventual number of racks stored or the length of time they may wait.
The central operational question is whether AWS sites gain the electrical capacity and installation path needed to put the incoming equipment to work.
Why it matters
- AWS bears the immediate operational cost: some sites have room for racks but lack the electrical capacity to run them, while teams face an unclear installation path.
- Amazon would pay for GB300 racks earlier than its usual installation point; the emails project roughly $5 billion in initial racks and about $3 billion more monthly.
- The planned purchases sit within Amazon’s projected $220 billion in capital expenditures this year, the highest among Big Tech peers according to Business Insider.
What to watch next
- Whether AWS sites obtain enough electrical capacity to operate the incoming GB300 racks.
- Whether AWS teams establish a clear path for installing the racks in data centres.
- Whether the projected roughly $5 billion in initial deliveries and about $3 billion monthly thereafter are reached.
How it's being covered
Coverage analysisShow details5 source articles · 5 unique outletsOutlet political mix: Too few rated· 2 Center · 3 unratedWhere the outlets are based: US/Canada 40% · Europe 20% · Other 40%
How outlets' accounts compare
2 framing differences · 1 uncorroborated claim · 3 information gaps- ComputerBase emphasizes storage at sites that lack electrical capacity and the earlier ownership and payment point; Business Insider centers the expected delivery values and installation uncertainty.
- Newsquawk reports the expected receipt at U.S. facilities before final deployment without the operational and capital-spending detail included by Business Insider.
- An August email said one AWS team did not have “a clear forecasted path” for installing the incoming racks.
- The sources do not specify which AWS sites lack electrical capacity or quantify their power shortfall.
- The sources do not state how long racks might remain in storage or what the early payment means for Amazon’s financing costs.
- The projected rack deliveries are forecasts; the reporting does not establish whether they were reached or how many racks are being stored.
Timeline
Developments from this story's sources, dated background we compiled and fact-checked, and our earlier coverage. Each row says which.- Jensen Huang, Chris Malachowsky and Curtis Priem founded Nvidia.Background · compiled and fact-checked by Updat3
- Amazon launched Amazon Web Services.Background · compiled and fact-checked by Updat3
- Nvidia introduced its CUDA platform for general-purpose computing on GPUs.Background · compiled and fact-checked by Updat3
- Amazon introduced its Elastic Compute Cloud service.Background · compiled and fact-checked by Updat3
- Nvidia announced its Blackwell GPU platform.Background · compiled and fact-checked by Updat3
- At AWS re:Invent on December 3, 2024, Amazon announced EC2 P6e-GB200 instances built around Nvidia’s Grace Blackwell chips, extending its cloud-computing business into increasingly power-intensive AI workloads.Background · from our historical context
- Nvidia announced the GB300 Blackwell Ultra platform on March 18, 2025, continuing the hardware generation behind that expansion.Background · from our historical context
How we got here
At AWS re:Invent on December 3, 2024, Amazon announced EC2 P6e-GB200 instances built around Nvidia’s Grace Blackwell chips, extending its cloud-computing business into increasingly power-intensive AI workloads. Nvidia announced the GB300 Blackwell Ultra platform on March 18, 2025, continuing the hardware generation behind that expansion.
The infrastructure model dates to Amazon’s launch of Amazon Web Services in 2006: its Elastic Compute Cloud (EC2), introduced in August, let customers rent computing capacity rather than operate their own servers. As demand grew, Amazon built and operated data centres to supply that capacity.
Original sources
1 of 5 articles · Unknown · Show allEvery story links to the reporting it was built from. How we choose sources
Spotted a factual error, a broken link or a misleading summary? Tell us.