Jim Keller Says Tenstorrent Wants to Challenge Cerebras, Nvidia, and the AI Hardware Market

Tenstorrent CEO Jim Keller says the company is preparing to compete more aggressively in the AI hardware market, with plans that could involve a major licensing deal with Intel or Qualcomm. Keller also said Tenstorrent expects to outperform Cerebras on performance and cost as the company expands sales of its Blackhole based AI servers.

The comments come as competition in AI accelerators grows beyond Nvidia. Companies such as Cerebras, Groq, Tenstorrent, and several large chipmakers are trying to offer alternatives for AI training and inference, particularly for customers that want lower costs, better efficiency, or more flexible hardware designs.

Keller said Tenstorrent is not concerned about Cerebras becoming more prominent through its public market plans. Instead, he described the situation as an opportunity for Tenstorrent to show how its hardware compares in large scale deployments.

Tenstorrent Is Betting on Blackhole Servers for AI Inference

Tenstorrent’s current strategy is built around its Blackhole processors and Galaxy server systems. The company believes its architecture can provide competitive AI inference performance without the high cost often associated with large Nvidia deployments.

The company has previously promoted Blackhole Galaxy servers as a lower cost option for AI workloads. Keller argues that the advantage is not only raw performance, but also the total cost of deploying and operating a large AI system.

Tenstorrent says its approach to memory management is one of its advantages. AI models need large amounts of memory for data such as key value caches, which help systems process and generate responses efficiently.

Keller explained that Tenstorrent can keep this data in DRAM on the same chips handling decoding work. The company can also use local SRAM for some workloads. When more capacity is needed, data can move to DRAM, though that can reduce performance compared with keeping everything close to the processor.

Keller Mentions Intel and Qualcomm Meetings

Keller also confirmed that he has met with the CEOs of Intel and Qualcomm. He did not confirm any acquisition discussions, but said Tenstorrent hopes to secure a major business agreement with one of the companies.

The likely focus would be Tenstorrent’s RISC-V CPU intellectual property. RISC-V is an open instruction set architecture that is becoming more important in custom chips, embedded systems, AI hardware, and specialized processors.

A licensing agreement could give Intel, Qualcomm, or another company access to Tenstorrent’s CPU designs for future products. Keller also said that a hyperscaler is looking at Tenstorrent’s AI intellectual property for a smaller chip project.

Orders Are Growing as Customers Seek Alternatives

Tenstorrent says it has already received substantial interest in its AI hardware. One offshore customer reportedly ordered a 96 Galaxy pod configuration containing 3,072 Blackhole chips.

The company is also said to be working on 1,000 Galaxy servers, with roughly half already sold. If accurate, that would show growing demand from customers looking for alternatives to the largest AI hardware suppliers.

Keller suggested that some customers can reduce costs significantly by choosing Tenstorrent hardware instead of a large Nvidia deployment. However, performance, software compatibility, availability, and support will remain important factors for businesses deciding which AI platform to use.

Tenstorrent Is Also Preparing for an IPO

Tenstorrent is reportedly working toward an initial public offering as it expands its product range, manufacturing capacity, and global presence. The company will need to prove that it can deliver hardware at scale while maintaining strong software support for customers.

The AI chip market is becoming more competitive every year, but it is still dominated by a small number of suppliers. Tenstorrent’s challenge will be turning ambitious performance and cost claims into long term customer deployments.