More than a half-dozen chipmakers are joining forces to launch the AI Platform Alliance, a consortium focused on supporting the development of artificial intelligence hardware.

The move was announced today by Ampere Computing LLC, which is leading the effort. Ampere is a well-funded developer of central processing units. The company’s customers include Google LLC and Oracle Corp., which offer its CPUs through their respective cloud platforms. 

The other chipmakers backing the AI Platform Alliance are Cerebras Systems Inc., FuriosaAI Inc., Graphcore Ltd., Kalray SADIR, Kinara Inc., Luminous Computing Inc., Neuchips Inc., Rebellions.ai and Sapeon. Ampere expects additional companies to sign up in the coming months.

Most of the consortium’s backers are venture-backed startups focused on developing AI chips. Along with Ampere, Cerebras and Graphcore are two of the biggest players in this ecosystem, having raised more than $1.4 billion from investors between them. Cerebras’ AI chip, the WSE-2, is the world’s largest processor with 2.6 trillion transistors. 

Not all of the AI Platform Alliance’s backers focus exclusively on the machine learning market. One of the participants is Kalray, a French company that develops so-called data processing units. Its chips are used to speed up data center storage infrastructure. 

Ampere said that one objective of the AI Platform Alliance will be to develop more cost-efficient machine learning hardware. Additionally, the group’s members will work together to make AI infrastructure more power-efficient and sustainable.

According to Ampere, another goal of the collaboration is to “validate joint AI solutions.” Such joint solutions might include servers that combine Ampere’s CPUs with machine learning chips made by the AI Platform Alliance’s other members. In the data center, machine learning chips such as graphics processing units are almost always deployed alongside CPUs. 

Ampere said that the consortium’s long-term objective is to provide a better alternative to GPUs. According to the company, the often steep cost of performing AI inference on graphics cards is one reason an alternative is necessary. 

The AI Platform Alliance is at least the second industry group that has launched in recent months to create more competition for Nvidia Corp., the world’s top supplier of data center GPUs.

In September, Intel Corp., Samsung Electronics Co. Ltd. and a number of other industry players formed the UXL Foundation. It’s a consortium focused on easing the task of writing AI applications that can run well on multiple machine learning processors. By making it easier to optimize AI software for chips other than Nvidia GPUs, the UXL Foundation could potentially make those chips more competitive.

Photo: Ampere