Nvidia is poorly poised to tackle the next phase of the
The maker of SN40L reconfigurable dataflow unit (RDU) chip has positioned itself as a power-efficient alternative meant for better inferencing speed and accuracy as compared to Nvidia GPUs. Backed by investors like SoftBank, Blackrock, Intel Capital, GV, Walden International, Temasek and GIC, the is valued at $5.1 billion and serves customers like Saudi Aramco, Accenture and Analog Devices.Liang said that billions of training dollars will be the prerogative of a select few companies like Meta, OpenAI and Google, mostly focussed in the US. Whereas bringing these models closer to the enterprise systems will be a priority.Most firms do not want to incur the cost of training a frontier model ground up, Liang said. Instead, an open-source pre-trained model, which is fine-tuned on a company’s private data, could achieve a higher level of accuracy at nearly 5% of the cost needed for training.Nvidia GPUs are not a sustainable solution to scale AI applications because every iteration of an AI model for a specific enterprise use-case will need new hardware, Liang said.
