Author | Wan Chen
Editor | Jing Yu
On the afternoon of September 25, Amazon announced on the company's official website that it would invest up to US$4 billion in the large modeling company Anthropic, which is famous for its chatbot Claude that rivals ChatGPT.
Cloud computing companies have invested heavily in large model companies. Microsoft has made a good start in February this year, acquiring OpenAI, a major customer of Microsoft's cloud Azure, and the priority use of ChatGPT for US$10 billion. Now it seems that this deal is a sure profit for Microsoft. Since this year, Microsoft's market value has increased by 40%.
But now, half a year later, the relationship between upstream and downstream in the field of large models is becoming complicated. Amazon’s investment in Anthropic is not just about “locking in” customers for AWS, it is not even necessarily for large models.
Official cooperation details show: Anthropic will use AWS Trainium and Inferentia chips to build, train and deploy its future basic models. In addition, the two companies will also collaborate on the development of future Trainium and Inferentia technologies. It is worth noting that AWS Trainium is a customized ML (machine learning) training chip launched by AWS at the end of 2020, and Inferentia chip is a high-performance machine learning inference chip launched by AWS in 2019.
By investing in Anthropic to deepen cooperation, Amazon intends to accelerate the development of self-developed AI chips.
A few days ago, The Information exclusively broke the news that NVIDIA wants to "charge a handling fee" from the three major cloud vendors - NVIDIA wants to lease NVIDIA servers from cloud vendors, so that it can provide AI applications to AI users through the cloud instead of servers or chips. Directly providing services, these applications also include the three major cloud vendors.
However, only Amazon rejected the offer.
Now it seems that Amazon’s reason for rejecting Nvidia is to find ways to speed up the upgrade of its self-developed AI chips. In the era of large models, Amazon’s core competitiveness in stabilizing its number one position in the cloud computing market lies in AI chips.
01
The first cloud platform launches
OpenAI’s biggest enemy
For this cooperation, Amazon said it will initially invest US$1.25 billion to purchase a minority stake in Anthropic. As part of the deal, Amazon said it had the option to increase its investment in Anthropic to $4 billion.
As part of the expanded collaboration:
Anthropic will use AWS Trainium and Inferentia chips to build, train and deploy its future foundational models, benefiting from the price, performance, scale and security of AWS. The two companies will also collaborate on future Trainium and Inferentia technologies.
AWS will become Anthropic's primary cloud provider for mission-critical workloads, including security research and future foundational model development. Anthropic plans to run the majority of its workloads on AWS, further providing Anthropic with advanced technology from the world's leading cloud provider.
Anthropic has made a long-term commitment to provide AWS customers around the world with access to its next-generation foundation models through Amazon Bedrock, a fully managed service from AWS that provides secure access to the industry's top foundation models. In addition, Anthropic will provide AWS customers with the unique capability of early access to model customization and fine-tuning capabilities.
Amazon developers and engineers will be able to build with Anthropic models through Amazon Bedrock, so they can incorporate generative AI capabilities into their work, enhance existing applications, and create new customer experiences across Amazon's businesses.
The companies did not disclose the valuation of Amazon's investment in Anthropic.
Dario and Daniela Amodei｜Image source: Times
Anthropic was founded by former OpenAI employees Daniela Amodei and Dario Amodei, who are the core developers of OpenAI GPT2 and GPT3. Currently, Anthropic has one of the most popular chatbots, Claude.
A challenger to ChatGPT maker OpenAI, Anthropic received a $400 million investment from Google earlier this year and was valued at more than $4 billion after a round of funding from Spark Capital in March.
02
What does Amazon want to do?
Investing up to $4 billion in Anthropic. What exactly does Amazon want to do?
An obvious reason is to "grab" customers. As the largest customers of cloud computing (in the future), large model manufacturers and AI application companies have become a battleground for major cloud manufacturers.
Since this year, Google, Microsoft, AWS, Oracle and Nvidia have all tacitly and unanimously done one thing - "locking in" customers through strategic investments (paying money to buy customers), although this operation is financially controversial.
But in fact, Anthropic has been a customer of AWS since 2021. AWS has deepened its cooperation with Anthropic through US$4 billion, aiming for deeper cooperation rights: large models, and most importantly-self-developed AI chips.
In other words, by strategically investing this tuition, Amazon learns how to build large models. It also developed AI chips that could even subvert NVIDIA GPUs by discussing with opponents that compete with OpenAI. After all, GPUs are not designed for training neural networks. Relying on CUDA and various technologies to "magically modify" scene by scene is an option, but it is not the optimal solution.
Amazon CEO Andy Jassy's speech confirmed this, saying: "We believe we can help improve many short-term and long-term customer experiences through deeper cooperation."
Short-term and long-term customer experience correspond to Amazon's large models and self-developed AI chips.
He further added, “Customers are very excited about Amazon Bedrock, AWS’ new managed service that enables companies to build generative AI applications using a variety of underlying models, and AWS Trainium, AWS’ AI training chip. AWS’s AI training chip and our cooperation with Anthropic should help customers get more value from these two functions.”
In fact, in the first half of the year, Amazon launched its own large model Titan and announced the typical customer of this large model. However, a few days later, this typical customer jumped on it and revealed to the media that Amazon's large model was not easy to use.
It can be seen that Amazon is not fully prepared for self-developed large models. Therefore, Amazon turned to the market to vigorously promote the large model platform Amazon Bedrock, where customers can call the services of other mainstream large model manufacturers, including Anthropic.
On the other hand, Amazon needs to stabilize its fundamentals in cloud computing.
In the era of large models, cloud computing faces different workloads and needs to explore new technologies to achieve faster reasoning capabilities. In this regard, Amazon is actually a pioneer. Self-developed data center chips and servers were once regarded as "one of the most important things" AWS has done to "differentiate from Microsoft and Google." Its servers have higher speeds and save power.
In terms of AI proprietary chips and servers, Amazon is also the earliest among the three major cloud vendors (Microsoft Cloud, Google Cloud). However, the progress and performance of its AI chips have not been released separately. Instead, they are bundled in servers and provided to customers through cloud services. Therefore, customers directly perceive only the performance of cloud computing, not the performance of the chips in it.
Now, Amazon needs to understand which workloads are best suited for which processors, and working with Anthropic is one way to do that.
Currently, of the 69 companies in The Information's generative AI database, 32 use Amazon, 26 use Google, and 13 use Microsoft as a cloud provider. Of course, some companies use multiple cloud providers.
In the era of large models, cooperation and competition in cloud computing, large models, and AI applications are becoming more complex. The cloud computing industry, which has not experienced major changes for a long time, has finally ushered in an opportunity for change.
Babbitt Park is open for cooperation!
Chinese Twitter: https://twitter.com/8BTC_OFFICIAL
English Twitter: https://twitter.com/btcinchina
Discord community: https://discord.gg/defidao
Telegram channel: https://t.me/Mute_8btc
Telegram community: https://t.me/news_8btc
