Anthropic's big AWS bet
- Anthropic committed more than $100 billion to Amazon Web Services to train and run its Claude models. - The deal spans ten years and supports large-scale training and inference on AWS infrastructure. - The commitment comes as Anthropic investigates unauthorized access to its restricted Mythos model, highlighting both scale and security concerns. (apnews.com; reuters.com)
Anthropic has agreed to spend more than $100 billion on Amazon Web Services over the next decade to train and run its Claude artificial intelligence models. (apnews.com) Amazon said Monday it will also invest up to $25 billion more in Anthropic under the expanded partnership. Anthropic said the cloud commitment secures up to 5 gigawatts of new computing capacity on Amazon’s systems. (reuters.com; anthropic.com) Anthropic said the agreement covers Amazon’s Graviton processors and Trainium chips, including Trainium2 through Trainium4, plus future generations of Amazon’s custom silicon. The company said it already uses more than 1 million Trainium2 chips to train and serve Claude. (anthropic.com) Cloud contracts have become as important to artificial intelligence labs as cash investments because training a frontier model requires vast amounts of computing power and electricity. Anthropic said more than 100,000 customers already use Claude through Amazon Bedrock, Amazon’s managed artificial intelligence service. (anthropic.com; aboutamazon.com) Amazon first invested $4 billion in Anthropic in 2023, making Amazon Web Services its primary cloud provider and giving Amazon a minority stake. The new agreement extends that relationship as Amazon pushes its own custom chips against Nvidia’s dominant artificial intelligence hardware. (aboutamazon.com; apnews.com) Anthropic announced the cloud deal while investigating unauthorized access to Mythos, a restricted cybersecurity-focused model it recently introduced for a small group of customers. Reuters, citing Bloomberg, reported that a small group of unauthorized users accessed the model. (reuters.com; bloomberg.com) Anthropic told Bloomberg the access came through a third-party vendor environment and said it was investigating the report. The company has described Mythos as powerful enough to enable dangerous cyberattacks if misused. (bloomberg.com; engadget.com) The partnership also expands Claude inference — the step where a trained model answers user requests — in Asia and Europe, according to Anthropic. That means the deal is not only about building bigger models, but also about paying for the day-to-day computing needed to serve customers at global scale. (anthropic.com) For Amazon, the deal locks in one of the largest long-term cloud commitments in the artificial intelligence race. For Anthropic, it ties the company’s growth to Amazon’s infrastructure just as demand for compute — and the risks around controlling access to advanced models — keeps rising. (apnews.com; anthropic.com; reuters.com)
