With AI rapidly developing, data centers must maximize the power of GPUs to meet huge processing needs. In that context, tensormesh has just received $4.5 million in seed investment capital, led by Laude Ventures and with the participation of Michael Franklin - a leading database expert.
This money was used to develop a commercial version of LMCache, an open source tool developed by co-founder Yihua Cheng. LMCache helps reduce AI operation costs by up to 10 times, and is now used by Google and Nvidia. tensormesh wants to turn this technology into a commercial product, making it easier for many other businesses to apply.
The core point of LMCache is key-value cushion memory (KV cache), allowing the system to save and reuse processed data, instead of deleting it after each request as in traditional methods. Tensormesh retains important data, helping AI models answer faster and save more resources. This method is especially effective for chatbot or virtual assistant applications, where the model must continuously refer to the history of the conversation or action.
According to CEO Junchen Jiang, although large AI companies can completely build similar systems themselves, this process requires many engineers and takes months to complete. With tensormesh, they can shorten the time and cost significantly, he said.
With a focus on memory optimization and GPU power, tensormesh is expected to become an important piece in the global AI infrastructure, as the demand for computing is increasing at an unprecedented rate.
