OpenAI is holding early talks with Amazon for a deal that could bring in at least $10 billion and push its valuation past $500 billion, and the whole thing is tied to Amazon’s push to get its Trainium chips into the big leagues.
OpenAI allegedly started speaking with Amazon around October, and that timing lines up with the company’s long corporate overhaul that ended with Microsoft owning 27% of the business.
That long reset gave OpenAI room to look outward again. Now Amazon is trying to move fast, because everyone in the chip race wants to chip away at Nvidia’s control.
Amazon pushes Trainium to rival Nvidia
If the deal lands, it will be a milestone for Amazon’s semiconductor group. Nvidia still leads the market and everyone knows it, but developers like Meta are already testing alternatives from Google and others.
Amazon wants in. According to Bloomberg, company says Trainium gives cheaper and more efficient compute than Nvidia’s GPUs for the heavy math behind large models, and it’s trying to win over teams who want to trim their bills.
Amazon Web Services is still the biggest provider of rented compute and storage.But AWS has not reached that same level with AI developers, and Microsoft’s backing of OpenAI has made the gap feel even wider.
Amazon wants a way to move the needle. A major funding tie-up with OpenAI, plus OpenAI using Trainium, would give Amazon something it has been trying to get for years: real weight in model training.
OpenAI’s valuation already hit $500 billion in an employee share sale, a number that briefly put it ahead of SpaceX. That kind of jump has made Wall Street nervous. Analysts warn about a bubble because some companies invest in customers only to push those customers to buy more of their own products. It becomes a loop. But for now, the money keeps coming in.
Amazon expands chips and models to get more AI customers
OpenAI and Amazon also signed another deal last month.AWS will provide $38 billion of cloud power to OpenAI over seven years.That agreement depends on huge batches of Nvidia chips.
Amazon is not dropping Nvidia anytime soon, but it wants Trainium in the mix. The new Trainium3 accelerator is now in a few data centers and will open to customers on Tuesday. Dave Brown from AWS said, “As we get into early next year we’ll start to scale out very, very quickly.” He said the pace will match Nvidia’s yearly chip cycle.
But Trainium has a weakness. Nvidia has massive software libraries that help teams get started fast. Amazon does not have that yet. Bedrock Robotics uses AWS for compute, but when it trains models to guide construction machines, it switches to Nvidia. CTO Kevin Peterson said, “We need it to be performant and easy to use. That’s Nvidia.”
Amazon is also trying to improve its AI models. Its earlier Nova models did not score high in public benchmarks. Rohit Prasad, who oversees model work at Amazon, said, “The real benchmark is the real world,” and he expects newer models to compete.
Amazon also built Nova Forge, a tool that lets advanced users customize Nova models while they are still training. Reddit is using it to check rule-breaking posts. CTO Chris Slowe said, “The fact that we can make it an expert in our specific area is where the value comes from.”
Want your project in front of crypto’s top minds? Feature it in our next industry report, where data meets impact.
