Amazon may be preparing to spend billions on a company that could help it get more out of its AI chips. The tech giant is in talks to acquire startup Decart for around $7B, according to the Wall Street Journal.
The San Francisco-based company develops software that makes it easier to run AI workloads across different chips. That's an attractive proposition for cloud providers looking to squeeze more computing power out of their hardware without constantly buying more of it.
Negotiations are ongoing, but a deal could come together soon.
Third time's the charm?
Decart isn't exactly short of admirers. The startup raised $300M in May at a $4B valuation, with Nvidia and Radical Ventures among its backers. Amazon's reported offer would represent a substantial premium just months later.
But two other AI heavyweights have already tried their luck. Anthropic walked away in September following due diligence, while Nvidia also failed to reach a deal.
So what's attracting all this attention? Decart's software is designed to make AI models run faster and more efficiently. Its Decart Optimization Stack, or DOS, claims processing speeds of 1.6K tokens per second.
The company has also developed Lucy and Oasis, technologies focused on real-time video generation and physical AI simulation.
Amazon, meanwhile, isn't approaching Decart as a stranger. The company is already a strategic customer and has provided the startup with early access to its Trainium chips.
Amazon knows this playbook
There's a familiar strategy behind the potential acquisition. Back in 2015, Amazon bought Israeli chip designer Annapurna Labs, which went on to play a central role in developing its Trainium and Graviton processors.
That deal helped Amazon build its own chips. Buying Decart could help it make those chips work harder.
Industry estimates suggest Decart's technology can run AI models up to 8x faster, potentially improving the economics of running AI workloads. Amazon Web Services has already given the startup early access to Trainium3.
The appeal goes beyond faster processing. More efficient software could help AWS get better returns from its computing infrastructure while reducing its dependence on outside chip suppliers, including Nvidia.
There's also the question of scale. At $7B, Decart would rank among Amazon's largest acquisitions, alongside MGM and Globalstar. That's a substantial bet on software for a company already developing its own AI hardware.
The bigger question is what Amazon would do with those efficiency gains. Better performance could improve AWS's margins, but it could just as easily encourage customers to run even more demanding workloads.
Either way, Amazon appears interested in owning more of the technology behind AI computing, from the chips themselves to the software that makes them useful.
