插图:CHRISTIAN GRALINGEN
In the fall of 2023, when Amazon Web Services (AWS) announced a $1.25 billion investment in generative AI startup Anthropic, tech professionals and media outlets responded in unison. They agreed that AWS is leaving no stone unturned to stay relevant in the post-ChatGPT era, "trying to keep up" with Microsoft and Google [the New York Times] or "catching up" with them [CNBC] because it "doesn't look like a leader" [Business Insider].
While AWS is the world's dominant cloud service provider, far larger than Microsoft's Azure cloud service or Alphabet's Google Cloud, it doesn't matter. The conventional wisdom is that cloud services have suddenly entered a new phase — the cloud 2.0 era, in which giants must compete based on how well their software, storage, and other tools support AI — and AWS is not in a position to win under the new rules.
When asked if AWS is "catching up," CEO Adam Selipsky didn't explicitly deny it. He told Fortune: "We only ran about three steps in the 10-kilometer race. Which player is half a step ahead or half a step behind? It's not really a question that matters. The important questions are: who are the runners, what does the track look like, what do the spectators and judges look like, where do we think the race is going?"
In other words, what Selipski wants to tell his skeptics is this: stop being overly anxious and focus on the long term. That's exactly the message you'd expect from a company that seems half a step behind in artificial intelligence. But it's also the message that Amazon has been preaching internally since its inception (and with great success) — quoting its famous Leadership Principles to "accept the truth that we may have been misunderstood for a long time."
Still, customers, competitors, and investors have reason to wonder if AWS's investment in Anthropic portends a broader strategic question: Will AWS fall from the throne in the era of cloud computing 2.0, and who will be the world's leading provider of AI in the long run?
To solve these problems, we must start with Anthropic. AWS bought the start-up, took a minority stake, and committed to an eventual investment of up to $4 billion for two reasons. One obvious reason is that AWS will be the main cloud service provider for Anthropic, which will provide the base model Claude (similar to the vast digital network that powers ChatGPT), and the AI assistant of the same name to accompany it. (Anthropic launched the Claude in March 2023; the Claude 2 in July, and the faster, cheaper version of the Claude Instant 1.2 in August.) A future version of Claude will be available to AWS customers through a service called Bedrock. There are a number of base models that can be accessed through Bedrock.
AWS has built or gained access to several underlying models, which it says provides with a wider range of cybersecurity and services than OpenAI currently provides. However, AWS's model is not comparable to OpenAI's GPT-4 in terms of scope and efficacy, hence the "laggard" argument.
Another less obvious reason (at least for non-technical people) is that Anthropic will use AWS's proprietary AI chips for the training and deployment of future models. This is important because AWS customers can use this type of model at a much lower cost than if they were based on hardware from other companies, such as Nvidia's popular but expensive chips. (Largely thanks to the frenzy for artificial intelligence, Nvidia's stock price has more than tripled in 2023.) Brad Shimmin, an analyst at research firm Omdia, explains: "If AWS visits a company and says, 'You only need to spend a third of the cost of hosting GPT [a model built by OpenAI] to be able to run Anthropic yourself," explains Brad Shimmin, an analyst at research firm Omdia. Then they'll win customers. ”
Broadly speaking, cost is a key factor in the future development of AI. Building a base model is expensive. Sam Altman, OpenAI's chief executive, said the cost of training the latest model, GPT-4, was "more than $100 million," in part because OpenAI used Nvidia's top-of-the-line chips. Amazon's chief executive, Andy Jassy, once told investors that he was "optimistic" that many foundational models would eventually be built on AWS's chips, and that AWS's scale made that aspiration even more feasible.
To learn more about Amazon's strengths, look to the experience of LexisNexis, a longtime AWS customer that provides software and online services to the law firm. Amazon recently launched a service called Lexis+ AI, which includes drafting briefs and contracts, summarizing judicial decisions, analyzing the company's own legal documents, and other features.
Jeff Reihl, CTO of LexisNexis, said his company had been working with the startup before AWS invested in Anthropic, and that the company had access to Bedrock. As AWS and Anthropic continue to expand and work together, the combined strengths of both companies provide the best version of Lexis+ AI it needs. Rael adds: "We will select the best model that will solve a specific problem with the best performance, the highest accuracy and the best price/performance ratio. ”
AWS is Xi for offering a wealth of options and winning at cost. Selipski pointed out that proponents of the Cloud 2.0 era mistakenly believe that the importance of AI lies in its impact on the fundamental nature of the cloud computing industry. His take on conventional wisdom is: "AI is going to be a very, very, very important thing in cloud computing — and it really is — 'Oh, that's a different thing. But in fact, he continued, generative AI hasn't changed the fundamentals of the cloud computing industry. "The experience of the past 17 years has shown that businesses should not operate their own data centers and buy their own servers, nor should they worry about network issues," he said. The world will soon agree on the same approach to generative AI, which is that businesses should figure out how to use it, not how to do rack stacking. ”
Daniel Newman, CEO of Futurum, a technology research and consulting firm, further argues Selipski's point. He said the "huge cost" of AI meant that "only a very small number of companies can afford it." Actually, I think AWS, Microsoft, and Google are too big to fail. ”
Ben Recht, a computer scientist at the University of California at Berkeley, who focuses on emerging industries, said partnerships like Anthropic's with Amazon, OpenAI and Microsoft made Newman's predictions more likely to come true. "At this point, the people behind 'start-ups' and 'big tech' are the same," Recht said. The same rich people make each other richer. ”
One can't help but make a bold prediction: AWS is likely to remain the leader for a long time. For a decade, AWS's cloud market share has remained stable between 31% and 33%, while Microsoft and Google are still far behind (23% and 10%, respectively). But the market share of these two competitors has been steadily increasing. Millions of companies have contracted with Microsoft through the use of Office 365 and other productivity software, and Azure can help these customers integrate AI into Excel, Word, and other Microsoft products. At the same time, Google's computer scientists made a breakthrough in artificial intelligence years ago. For AWS, the competition in the age of AI is only going to intensify.
Artificial intelligence has also dramatically accelerated the pace of change. "The progress made over the past 18 months has been incredible," Rael said. "The year ahead is likely to bring more surprises.
Anthropic has become a partner to big tech companies. This is an unexpected chapter for one of the most unusual start-ups in the field of artificial intelligence. (At the same time that AWS announced its investment in Anthropic, Google also announced its investment in Anthropic.) Anthropic was founded in 2021 by former OpenAI employees who left the startup over disagreements over OpenAI's partnership with Microsoft, including concerns that OpenAI had abandoned its mission to develop safer and more ethical AI and become too focused on business issues. Anthropic's CEO, Dario Amodei, was previously OpenAI's vice president of research, and his sister, Daniela Amodei, is Anthropic's president and former vice president of security and policy at OpenAI.
Anthropic is one of many AI startups that explicitly claim to put ethics first. The company operates as a public benefit corporation, which requires it to achieve its business goals while weighing its social impact and demonstrating that it is committed to its purpose by regularly publishing social impact indicators. [Patagonia, ice cream brand Ben & Jerry's, and craft trading site Etsy have all adopted this corporate structure.] ]
Anthropic's founders were also actively involved in the effective altruism movement, a philosophical and social movement focused on using logical analysis to figure out the most effective ways to help those in need. Rooted in the effective altruistic movement, Anthropic raised $500 million from an investor named FTX founder Sam Bankman-Fried (whose moral credentials have been revoked), who was convicted of fraud in early November 2023. Anthropic's efforts to stand out in terms of ethics have attracted some interest (admiration). Jim Hare, a prominent VP analyst at Gartner, highlighted the startup's emphasis on "constitutional AI," which incorporates ethical principles into the training of AI models, in part through public consultation and in conjunction with ideas from sources such as the United Nations' Universal Declaration of Human Rights.
Others believe that Anthropic's ambitions are limited. Gary Marcus, professor emeritus of cognitive science at New York University and an expert on the constraints of "deep Xi" (the foundations of artificial intelligence), noted that Claude "still relies heavily on large language models." "We know that large language models are fundamentally opaque, often hallucinatory, and unreliable, which I think is an extremely bad underlying technology for AI to adhere to ethical values," he added. ”
It remains to be seen how Anthropic's new partnership will impact its efforts to position itself as an AI business that adheres to ethical values. (The company declined to comment for this article.) What happens if the business needs of AWS or Google and their customers are at odds with Anthropic's best practices, or raise bias or concerns about user privacy? Given the rapid pace of development in this space, we may not have to wait long to find an answer. (Fortune Chinese Network)
作者：葛继甫（Geoff Colvin）, Kylie Robison
译者：Zhy
At Fortune Plus, netizens have made many in-depth and thoughtful comments on this article. Let's take a look. You are also welcome to join us and talk about your ideas. Other hot topics today:
Check out the wonderful view of "The Middle-income Group in China Has Reached 400 Million".
Check out the wonderful views of "Fund Manager Salary Limit on Hot Search".