• AWS expected to get $5 billion of Nvidia GB300 racks, followed by $3 billion more each month.
• Some of these GPU systems are being stored elsewhere due to lack of data center power.
• Amazon CEO Andy Jassy sees AI as a major opportunity, but getting capacity online is a challenge.

Amazon is racing to secure AI hardware earlier than before, even as data center power shortages leave some GPU systems waiting in warehouses.

Under a new Early Rack Receipt process, AWS expected to receive roughly $5 billion in Nvidia's GB300 server racks at US facilities before final deployment, according to August internal emails reviewed by Business Insider. The company projected roughly $3 billion more in GB300 racks each month "for the foreseeable future," one of the emails said.

Some AWS sites had room for more racks but not enough available power to turn them on, people familiar with the effort said, leaving finished equipment to be staged at intermediary warehouses until the facilities were ready.

At least one AWS team did not have "a clear forecasted path" for getting the incoming racks into data centers at the time, according to one of the emails.

Dark GPUs

A lack of new electricity capacity is one of several bottlenecks slowing the buildout of some AI data centers. This has led to a phenomenon known as dark GPUs, where AI chips and servers are purchased but can sit idle due to power supply issues or other challenges. (Dark GPUs can also refer to system design flaws that mean this gear isn't fully used after being plugged in).

This is partly why financial pressure is mounting on big data center projects. Money must be spent upfront to secure supplies of Nvidia GPUs and other valuable AI components. But if some data centers aren't ready to run, those early investments can tie up huge amounts of cash. Debt financing can fill in the gaps, but the market has become less willing to lend against some of these projects lately.

Trade-off

Amazon's early-rack program is one piece of the cloud leader's broader AI buildout, which is putting new pressure on its cash flow, as Business Insider exclusively reported this week. Amazon expects roughly $220 billion in capital expenditures this year — more than any Big Tech peer. Its free-cash-flow turned negative last quarter for the first time since 2023, and could stay negative for a while.

For AWS, the trade-off is securing scarce GPU racks before competitors do, even if some data centers are not yet ready to turn them on. But taking ownership earlier can tie up billions in equipment before it starts generating cloud revenue.

Previously, AWS took ownership of a finished rack only upon delivery to a data center, according to internal documents. Under the new approach, it can pay for and record certain racks earlier in the supply chain.

An Amazon spokesperson said the company routinely stages finished hardware near deployment sites so it can install it once capacity becomes available, calling the practice "best business practice." The spokesperson added that Amazon has long-established systems for tracking infrastructure assets and disputed that the broader company lacked a path for getting racks into data centers.

Amazon CEO Andy Jassy previously said the company has a "clear line of sight" to strong returns from its AI investments, calling the technology a "once in a lifetime opportunity."

"We've been through this cycle with the first big AWS growth wave, and liked the results," Jassy wrote in Amazon's annual shareholder letter. "We expect to feel similarly about this next wave."

Have a tip? Contact this reporter via email at ekim@businessinsider.com or Signal, Telegram, or WhatsApp at 650-942-3061. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.

• Microsoft AI recently announced its latest model MAI-Base-2.
• AI CEO Mustafa Suleyman compared the model to a similar offering from OpenAI.
• Microsoft trained its model more efficiently than OpenAI, he said in an internal meeting.

Microsoft AI CEO Mustafa Suleyman told employees an in-house model was better than a similar offering from OpenAI, one of the software giant's biggest customers and partners.

"We now have a stronger pre-trained base model than the model that currently powers GPT 5.6 Sol," Suleyman said during a town hall meeting with employees on Thursday. "And in fact, we used one-quarter of the compute that OpenAI used to train their model."

Pre-training is the first stage of building an AI model, when it learns patterns from huge amounts of text, images, and other information. This takes massive amounts of expensive computing power, so Microsoft could save money if it's worked out a way to do this more efficiently.

Microsoft announced the model, MAI-Base-2, earlier on Thursday.

Suleyman said Microsoft privately tested this model against OpenAI models, including one nicknamed "Spud" that powers GPT-5.6. He added that Microsoft's model performed better on tests measuring how well AI can understand and predict patterns in information.

The comparisons are based on technical information Microsoft has about OpenAI's models, Suleyman said.

"Privately and confidentially, we've also compared it to the OpenAI models," Suleyman told employees, while suggesting that Microsoft has special access to information on how OpenAI models perform.

It's notable that Suleyman would choose to highlight limitations of OpenAI models, even in private discussions. Microsoft has invested billions of dollars in OpenAI, and owns a stake in the startup that is likely worth more than $100 billion. Any criticism of OpenAI could potentially dent the value of this huge investment.

The two companies also have a major partnership, with Microsoft offering OpenAI models to its Azure cloud customers. Microsoft has used OpenAI models for its own products such as Copilot, too.

But the relationship between the two companies has frayed in the past year or so. Increasingly, Microsoft has tried to branch out from OpenAI, including by working on its own models.

Suleyman also said Microsoft's new AI model used "12 times less computation than was used for DeepSeek, for example, on compute efficiency."

Suleyman didn't specify which DeepSeek model, but a public blog post described the pre-training of MAI-Base-2 being a lot more efficient than a similar offering from the Chinese lab, called DeepSeek-V4-Pro-Base.

Microsoft in March created a new combined team for its commercial and consumer Copilot products and reassigned Suleyman to focus on building AI models in the company's superintelligence unit.

Microsoft declined to comment and OpenAI did not respond to a request for comment on Thursday.

Have a tip? Contact this reporter via email at astewart@businessinsider.com or Signal at +1-425-344-8242. Use a personal email address and a nonwork device; here's our guide to sharing information securely.