AI Card Bao AICARDBAO.COM

Industry News

During their roadshow, Anthropic discussed their chips: 15 people worked together to develop custom chips using Claude.

When Anthropic recently met with potential IPO investors, it emphasized one particular aspect: its own chip-design capabilities.

Very little information has been made public—only a few details. A new hardware team consisting of 15 people has been formed to use Claude to accelerate the development of custom chips. According to the company, this approach takes less time than the established engineering processes used by NVIDIA and AMD. The valuation of this upcoming listing is reportedly close to $1.5 trillion.

Having 15 people work on chip development would be considered a joke in any semiconductor company. The development of a data center chip using advanced manufacturing processes usually requires the efforts of thousands of engineers over several years. Therefore, what investors really want to hear is not “we plan to produce chips”, but rather “our approach can lower these technical barriers”. This is essentially a proposition designed to influence valuation decisions.

The market's reaction has also been in this direction: The first sector to show signs of concern was the EDA industry. If models can truly take over tasks that require a significant amount of human effort, such as validation and layout, then the companies that sell these tools will face significant changes in their business models.

However, the information currently available does not include any results regarding film testing, manufacturing processes, or performance figures. The claim of “shortening development time” cannot be verified at this time.

Another issue arises regarding the list of shareholders. Amazon has its own developed Trainium technology, while Nvidia specializes in the production of GPUs – both companies are also significant investors in Anthropic. When customers who happen to be shareholders learn that you intend to develop your own chips, their reaction is likely to be less than enthusiastic.

In the prospectus updated this week, it is stated that the cost associated with computing power in the coming year will amount to 69.8 billion yuan. Even if achieving a savings of just a few percentage points in the development of self-developed chips proves feasible, that alone would be enough to tell a compelling story. As for whether such development is actually possible, that remains to be seen in the future.

Related