Amazon Invests $2.75B in AI Startup Anthropic

What is it that that Amazon and AWS saw in Anthropic that that pushed you to commit? Well, thanks Ed. Appreciate you having me here. It’s great to be here. You know, Anthropic are a provider of what we call foundation models and they are at the very centre of the generative AI revolution. This ability for the first time to be able to reason and integrate across very large amounts of information, to be able to build assistance and chat bots, but also completely reimagine every aspect of the customer experience. Anthropic have some of the best foundation models available anywhere today and it was been a delight to get to know that team more closely. And our investment really reflects the way that we’re partnering very expansively with Anthropic to bring these remarkable foundation models that anthropic pools clawed to customers across every single industry and every single use case. It’s really been remarkable over the past year to watch so many customers from financial services like ADP and intuit manufacturing with Siemens, transportation with Delta Airlines, Pfizer and life sciences AT and telecom. Just every industry that you can imagine starting to use these foundation models in order to be able to completely reimagine the experience. Now Anthropic makes available Lord 3, which is currently the best performing set of foundation models available anywhere. And so we’ve been able to partner with them to bring those models to customers through AWS through a service we have called Bedrock. And we’re also partnering to allow Anthropic, who have selected AWS as their primary cloud provider for their mission critical training workloads going forwards to optimize their workloads. For our own custom silicon that we use for machine learning which we call Trainium, which is a specific chip designed to accelerate machine learning and artificial intelligence model training and Inferentia, which is a specific chip designed to accelerate the prediction or the use cases map of foundation models. Let me let me jump in here when when this news broke yesterday I I posted on social media for example the the story and some people replied saying this looks very much like a quid pro quo where Amazon or AWS invests in anthropic but anthropic is required to use AWS compute and Trainium and Inferentia. Can you can you kind of give me some some numbers behind Anthropic’s use of of both of the the proprietary chips from AWS? Have they started using this? How is the kind of onboarding of Anthropic operationally into AWS going? Yeah, as I say the anthropic of selected AWS. We’re collaborating both on model training, but also the work that we’re doing with Anthropic is helpful in how we develop the next generation of these pretty remarkable chips which provide better efficiency and better cost profiles for training these very large foundation models. The investment in Anthropic is an equity investment. It’s a convertible note. We don’t have a seat on the board. We have no say in how the money gets spent and so they can use that investment as they see fit. What’s really interesting and Anthropic have told us this on the program very recently is that they are getting a lot of success through Bedrock in getting their clawed models out to enterprise commercial customers. What is Amazon and AWS getting from Anthropic and their activity from a kind of business volume perspective, you know what is the the, the, the benefit or the sales boost to to AWS, Yeah for sure. So, so Bedrock is our capability for making these foundation models available and we make available models from Anthropic and we also make available models from other providers including Amazon and other providers like Stability for image generation. We have publicly available models from Meta and Mistral and AI 21 Labs and Cohere and so on and so forth. And So what we found is that customers want to be able to access the best performing, most remarkable models in an environment which is private and secure, where their data is not reviewed by humans, does not go into training the underlying models, does not travel over the public Internet and have all the additional capabilities which make you successful with generative AI such as grounding these models in your own business data so they can answer questions on information that they haven’t been trained on yet. And so that’s what bedrock enables customers to do. Let’s talk about chips because you know that that I love to. I think I’m right in saying you were at GTC or of course you’re aware of what was announced at GTC, which is Blackwell and Tropics using Trainium and Inferentia. But just give me your and you are an expert. Give me your expert opinion and reaction to Blackwell and what you think you’re up against there. Yeah, I mean Blackwell is a remarkable artificial intelligence platform that combines GP US and CP US. It’s likely to be the absolute best, you know, beast at providing machine learning training. And so what’s really important and our approach here maybe differs from others is that optionality of compute and models really matters with generative AI. And so we make available NVIDIA GPUs. We were the 1st and have got a, we were the first to bring those GPUs to customers through AWS. We’ve got a very long deep partnership with with NVIDIA that goes back nearly 13 years. I’ve been on stage with Jenson. Jenson joined our conference. We’ve worked closely together for for over a decade now. And NVIDIA also run their own AI supercomputer on AWS that we’re partnering to build together, which will have 10s of thousands of Blackwell platforms connected together with very fast interconnects, which is going to deliver over 400 exaflops of compute power so that NVIDIA can build their own foundation models, simulate the earth and so on and so forth. And so optionality really matters and we’re going to continue to build our own chips. We’re going to continue to bring chips from other providers and we’re going to continue to invest and partner with NVIDIA.