Jim Keller Says That Tenstorrent Is Going After AI Markets That Aren’t “Served Well” By NVIDIA

Jim Keller, the iconic chip designer, and Tenstorrent's CEO believes that NVIDIA has yet to tap into certain AI market segments, creating a gap for other companies to fill.

Jim Keller Believes In AI Compute Scalability & Cost-Effectiveness As Tenstorrent To Target Markets That Are Ignored or Not Touched By NVIDIA

In an interview with Nikkei Asia, Jim Keller said that despite NVIDIA controlling a hefty share of the AI markets, there are still some gaps left that Team Green hasn't been able to fill in, and this has provided an opportunity to the likes of Tenstorrent and other AI companies involved in the race. Jim claims that Tenstorrent is all into building cost-effective solutions for the markets, given that NVIDIA has targeted the higher-end segment, which is unaffordable for several AI startups looking to get their hands on adequate computing power.

Related Story Samsung Reportedly Passes NVIDIA’s HBM3E Qualification Tests, Supply Expected To Initiate Next Quarter

Tenstorrent's CEO has emphasized the scalability of AI products, claiming that AI applications are expanding into smartphones, electric vehicles, and cloud services. Hence, the markets need a "scalable ecosystem," and that's where the firm's Galaxy AI computing system comes into play. Tenstorrent's Galaxy line of AI compute systems offer dense, scalable, and high-performance AI computations, as they are built upon an Ethernet-based mesh topology, which allows the addition of compute nodes much more easily.

We all know how "pro-Ethernet" Jim Keller is, and Tenstorrent has been one of the first firms to join the bandwagon, hence in terms of compute scalability, the firm definitely has an edge. Jim says that their Galaxy systems are three times more efficient than Nvidia's DGX while costing 33% less, simply due to not utilizing HBM and other costly resources that NVIDIA employs. Jim claims that the importance of HBM in the markets isn't sustainable, claiming that the memory solution is far more expensive and