Amazon’s Message to Nvidia and Anthropic; Andreessen Horowitz’s Latest AI Agent Bet
Amazon CEO Andy Jassy and Amazon Web Services chief Matt Garman each went out of their way to cast shade at Anthropic and Nvidia, respectively, while they touted Amazon’s new AI models and chips that aim to compete with products from those companies.
“I figured that almost everybody would end up using Anthropic's Claude models...but they're also using Llama models, and they're also using Mistral models, and they're also using some of our own models... This kind of surprised us,” Jassy said during a keynote speech.
Did it actually surprise him, though? Amazon executives for the past two years have consistently borrowed language from Lord of the Rings in stating there would be “no one AI model to rule them all.”
In any case, Jassy suggested Amazon’s most advanced new model (some features of which we previewed last week) is basically as good as Anthropic’s (and OpenAI’s) latest models, and, notably, is priced more than three times less than those models.
The dig at Anthropic doesn’t change the startup’s importance to Amazon, however. Amazon is putting $8 billion into Anthropic, and Anthropic is spending much of that capital to rent Amazon’s cloud servers. And Amazon seems to be returning the favor by saying Tuesday it is building a large server cluster for Anthropic so the startup can keep up with OpenAI and Google.
Then there was Amazon’s valiant effort to convince cloud customers to use Amazon AI server chips instead of Nvidia’s as they develop and run AI applications. Nvidia’s dominance has been hard to shake, in part because AI programmers are so used to writing apps using Nvidia’s proprietary programming language, Cuda, and because its chips are the most powerful on the market. So far that’s been OK for cloud providers like Amazon, as they have made bank from renting out Nvidia-powered servers.
But Amazon, like Google and Microsoft, has soldiered on with its own chips so it isn’t entirely dependent on Nvidia. Why shouldn’t it? After all, the chip giant has its own cloud and software ambitions that has brought it into conflict with AWS in the past.
Garman said Amazon’s new AI chips, Trainium 2, were a much better value for AI developers than Nvidia’s flagship H100 chip. And he said Anthropic would be using a big cluster of Trainium chips to develop future models.
Amazon also turned heads by bringing an executive from Apple to help make the case that Amazon’s technologies—both the new AI models and its server chips—are just as good as anyone else’s. The Apple executive, Benoit Dupin, said Apple uses AWS’ Graviton server chips and is in the “early stages” of evaluating the new Trainium chips for AI. More here.
To be sure, Apple is already a huge customer of Amazon’s servers, and it wasn’t immediately clear how much of Amazon’s AI software it was using. Apple has many vendors—it is arguably Google Cloud’s most important customer and uses that company’s AI chips too.
The surprise Apple appearance seemed to have the intended effect: a slew of headlines in the business press.
Amazon’s shading of Anthropic and Nvidia stood in stark contrast to its conference a year ago, when the CEOs of both of those companies appeared on stage with Amazon executives.
It shows just how quickly things are changing in AI. Exclusive alliances are breaking apart (e.g. this and this) and everyone is angling to get as much of the enterprise IT budget as they can—especially given that budgets aren’t growing very much, despite the AI boom.