Anthropic Explores Acquisition of AI Chip Startup MatX
Anthropic has engaged in discussions regarding the acquisition of artificial intelligence chip startup MatX, with the deal estimated at approximately $7 billion.
This move aims to bolster Anthropic’s endeavors in developing bespoke hardware for its rapidly expanding AI operations, according to two individuals familiar with the negotiations who spoke to Reuters.
Initially centered around a potential merger, the conversations have reportedly shifted toward a partnership, highlighting the AI laboratory’s ambition to gather the necessary resources and expertise to manufacture its own chips. This strategic initiative is intended to expedite Anthropic’s internal chip development efforts.
While Reuters is revealing these discussions for the first time, the reasons behind the current inactivity of the talks remain undisclosed.
MatX, established by ex-Google tensor processing unit (TPU) engineers, is now reportedly seeking to secure fresh capital, aiming for a valuation of around $4 billion, as revealed by a source who requested anonymity due to the sensitive nature of the matter.
Chip Development Goals Amid IPO Aspirations
As Anthropic continues to scale its Claude series of AI models, it seeks to create hardware that can satiate its considerable demand for data processing and lessen its dependency on chips manufactured by Nvidia.
Anticipating its initial public offering (IPO) within the year, Anthropic has been recruiting engineering and executive talent to expedite the chip design process, which typically spans several years.
Expectations for Anthropic’s IPO follow closely on the heels of SpaceX’s public listing, which reached a staggering $1 trillion valuation. Anthropic is striving for a valuation of $2 trillion, contingent upon achieving a revenue target of up to $200 billion by 2028, as earlier reported by Reuters.
Anthropic is expanding its in-house silicon team to craft custom chips that will enhance the operational efficiency and speed of Claude models. Furthermore, the company plans to maintain a multi-chip strategy by collaborating with various providers, extending from Nvidia to Google.
The design of chips is both costly and protracted. Developing a viable hardware solution can take over a year, with design expenses for each generation soaring into the hundreds of millions of dollars.
An acquisition of an AI chip startup like MatX would yield in-house design capabilities and potentially reduce long-term expenses. MatX is engaged in developing a chip that would be advantageous for training extensive AI models.
In its pursuit of computational power, Anthropic plans to allocate tens of billions of dollars for rental agreements with cloud service providers, while also pursuing direct chip purchases.
Specifically, the organization intends to procure $36 billion worth of AI chips from Google, alongside a $45 billion contract to utilize AI cloud computing resources from Nscale. A separate agreement with SpaceX entails a monthly payment of $1.25 billion through May 2029 for capacity across its data center networks.
Engagements with Other AI Chip Startups
The discussions with MatX imply Anthropic’s intent to develop a training chip, contrasting with other startups and competitors like OpenAI, which are focusing on processors optimized for chatbot response generation, a function known as inference. Sources suggest that Anthropic may also consider producing an inference chip.
Recently, Anthropic has convened meetings with a variety of AI chip startups, although no acquisition decisions have been made yet.
These meetings signify attempts by Anthropic’s engineers and executives to comprehend the diverse strategies in contemporary chip design.
This week, Anthropic welcomed Amir Salek, a veteran from Google in chip development, as part of its hardware push.
Additionally, in June, the company enlisted Clive Chan, a former chip engineer from OpenAI, who had previously contributed to the design of OpenAI’s recent chip unveiling.
Leading AI laboratories like Anthropic and OpenAI are increasingly concentrating on custom chip development, which can be tailored specifically to their models and workloads. This trend aims to provide significant performance enhancements and economic efficiencies.
At a recent conference, OpenAI claimed that its inaugural custom chip, dubbed Jalapeno, surpassed a comparable processor developed by Nvidia. OpenAI representatives highlighted the superior energy efficiency of their chip while performing inference calculations.
The endeavor to create custom chips may also provide Anthropic a safeguard against the constrained supply of Nvidia processors, which Nvidia announced would be limited through 2027.
In contrast, Google has pioneered TPUs, while Amazon has developed its own Trainium and Inferentia chips.
Anthropic has been ahead of the curve, utilizing models across hardware from various vendors, including Nvidia, Google, and Amazon, while seeking additional computational capacity to meet the rising demand for its models, including a strategic agreement with SpaceX to leverage its Colossus 1 facility, which hosts over 220,000 Nvidia chips.
Source link: Business-standard.com.
