Amazon is betting big that OpenAI stays relevant

The following post is a collaboration with my long-time friend, colleague and writing partner Alex Wilhelm. Subscribe to Alex's newsletter ‘Cautious Optimism‘ here. You'll be glad you did!

In 2023, following an uninspiring AWS re:Invent, Amazon's enormous annual customer event in Las Vegas, the company found itself in an unusual position when it came to generative AI. Instead of leading the way, as it often did with new tech, it was playing catch-up, as rival Microsoft and its partnership with OpenAI held the industry's full attention. 

But back in 2023, Amazon had one big factor in its favor: generative AI was so nascent that it was always possible that Amazon could close the gap within a year or two. 

The company has certainly made strides since 2023, launching its own models, investing in Anthropic, and bolstering its chip business, but a deal with OpenAI remained elusive until last month when it agreed to an investment that could be worth up to $50 billion. The catch was OpenAI commanded an $840 billion valuation, making it a very expensive investment indeed.

It's worth noting Amazon is paying just $15 billion of the investment up front; the rest will only be paid if certain conditions are met. Reports suggest those conditions are steep, involving either an IPO or achieving AGI (artificial general intelligence). 

It seems much more likely OpenAI goes public before it can bring about AGI. The term remains loosely defined and is often equated with human-level intelligence, a milestone some believe could be decades away. The company could go public anytime, of course, but for now, call the conditions a long shot until we hear otherwise.

The devil is in the details

In addition (exchange?), OpenAI is agreeing to buy $100 billion worth of AWS infrastructure services over the next eight years, kicking up its prior $38 billion commitment. That involves purchasing "approximately 2 gigawatts of Trainium capacity through AWS infrastructure," according to OpenAI. 

OpenAI has signed a bunch of these circular deals, which involve it taking money from a strategic investor and agreeing to buy a certain amount of services from them. The AI lab signed a similar arrangement with Nvidia last September, too, as an example. 

Right now, OpenAI has committed $600 billion towards infrastructure between now and 2030. When you consider that the company pulled in revenue of $13.5 billion last year, per CNBC, it's going to have to generate a lot more dough to make this pay off.

But despite this being a circular deal and the conditions on the $50 billion investment, this was finally a chance for Amazon to get OpenAI on its Bedrock model management platform, where the coveted GPT models had been conspicuously missing. 

The deal also differs from OpenAI’s with Microsoft because it involves stateful workloads — apps or agents that remember what happened before, then use those memories to process each new request.

The concept sounds tailor-made for agentic AI, but it probably won't be easy to pull off since it involves a complex set of interactions, making it feel like another longshot piece of the deal.

Amazon's cost of doing business

We could call it a $15 billion investment for now until we hear otherwise, but even that amount of money came at a high cost. As Om Malik pointed out, the investment came in today's dollars at a value of $840 billion post-money. Sure, Amazon finally got a piece of the AI lab, but it wasn't cheap.

It would have been better to invest a little earlier, all things held equal. But before we condemn Andy Jassy and company, recall that for some time, hyperscalers chose an AI lab to work with, and didn’t invest in those that had selected other hyperscalers. 

Today, however, AI labs run compute on every hyperscaled cloud infrastructure. Microsoft has backed Anthropic (a traditionally Amazon-aligned company), just as Amazon has now backed OpenAI (a traditionally Microsoft-aligned company). Google has invested in Anthropic while providing compute for both labs, despite being in direct competition with all of them.

Not that Amazon is shy about putting capital to work. The company’s $200 billion capital expenditure target for 2026 was large enough to worry investors weeks before it announced the OpenAI deal. Clearly, those concerns failed to slow Amazon’s AI push. (Sorry, decels!)

Was the price tag worth it?

Did Amazon overpay for OpenAI shares? And is it merely giving a vendor cash to pay it back for cloud services? No. Amazon is hedging its bet on Anthropic and its own in-house AI efforts by securing a specialized revenue stream with a leading AI lab and betting heavily on future compute demand.

This bet-hedging element requires little explanation. Anthropic has been showing up OpenAI thus far in 2026 with its rate of revenue growth, but no leading AI lab makes for a fully secure bet given the pace of model improvement. The risk is real: with any proximate release, OpenAI or Anthropic could lose their footing (recall that GPT-5 launch?).

The logic driving the specialized revenue stream is also clear. How much market demand there will be for the OpenAI-Amazon agentic memory setup is not clear, but it’s good for something. It also helps Amazon’s critical goal of keeping AWS revenue growth high, so the juice here is worth the collective squeeze. (Amazon’s cloud business is growing far faster than its ecommerce ventures.)

But what about future compute demand? Here we see Amazon using its wealth to both companies’ advantage:

• Amazon generated $139.5 billion worth of operating cash flow in 2025, so it can afford to invest into OpenAI, as it has both ready capital and the ability to raise debt at low rates. 
• OpenAI is growing quickly, but is incredibly cash-hungry. Its latest round ensures it is well-capitalized through IPO, but it didn’t raise $110 billion for fun.
• The AI lab has massive compute needs that will be better realized with more capital in hand: the more OpenAI can invest today, the better its AI products will become (in theory), and thus the more compute it will need thanks to growing demand.
• Amazon needs to ensure that demand for its cloud infra not only continues to grow, but accelerates.

Adding it all up: Amazon gets to invest between $15 and $50 billion into OpenAI today, helping the latter  continue sprinting, while locking in points of growth for AWS, which in turn will help the cloud business keep growth hot. 

Even if we try to be maximally cynical, Amazon is spending $50 billion to secure $100 billion in revenue. And it’s also earning equity upside in OpenAI while carving out the stateless runtime setup.

Is the agreement cheap? Heck no. But it isn’t the self-gratifying, merry-go-round of capital that many think it is. Instead, the OpenAI-Amazon deal is a marriage of relative strengths and needs, across time, that ensures both companies get what they need, when they need it.

And why is AWS so critical for Amazon? In the fourth quarter of 2025, AWS accounted for just under 17% of all Amazon revenue, but it generated half ($12.5 billion) of total operating income ($25 billion). Cloud compute is a delightfully margin-positive business for Amazon, and core to its long-term cash flow.

• The most important number for Amazon today is AWS’ growth rate. Amazon has managed to reaccelerate the figure in recent quarters, and every point of incremental growth is likely valued in the tens, if not hundreds, of billions of dollars.

Which is to say it's long-term value. The technology titan is spending a little more today to help ensure that its gravy train stays welded to the rails. How can we hate that?