SambaNova Challenges Nvidia With New Chip and Intel Backing
SambaNova unveiled its new SN50 chip, which it claims offers a 5x speedup and 3x lower inference cost than competing Nvidia products. The company also announced a new $350 million funding round led by Intel Capital, Vista, and T. Rowe Price. The funding supports a multi-year collaboration with Intel to develop AI inference solutions based on Intel's Xeon processors.
- Before this latest $350 million funding round, SambaNova Systems had already raised a total of $1.13 billion over four rounds since its founding in 2017. Its Series D in April 2021 brought in $676 million and valued the company at $5.1 billion. - The new SN50 chip's predecessor, the SN40L, is built on TSMC's 5nm process and features a three-tier memory architecture to handle large AI models. This design includes on-chip SRAM, co-packaged HBM, and off-package DDR DRAM, allowing it to process models with trillions of parameters. - Intel Capital was an early investor in SambaNova, participating in its Series B funding round in April 2019. The relationship between the two companies is also highlighted by the fact that Lip-Bu Tan, who became Intel's CEO, has been the chairman of SambaNova's board. - The collaboration with Intel is a multi-year agreement where SambaNova will use Intel's server chips and graphics cards to enhance the performance of its generative AI models. This partnership is aimed at strengthening their combined position against Nvidia, which holds an estimated 70-90% market share in AI accelerators. - The competitive landscape for AI chips is intense, with Nvidia currently dominating the market for AI training. However, the focus is increasingly shifting towards inference, the deployment of AI models, where companies like AMD, Groq, and SambaNova aim to offer more cost-effective solutions. - Enterprise hardware sales, particularly for high-value, complex systems like AI accelerators, often involve long sales cycles, typically ranging from 90 to over 180 days. Deals with an annual contract value of over $500,000 can take an average of 270 days to close. - For sales operations in this sector, accurate forecasting is critical but challenging due to long deal cycles and multiple stakeholders. Effective methodologies include opportunity stage forecasting, which weighs deals based on their pipeline progression, and increasingly, AI-powered predictive models that analyze historical data and deal engagement to improve accuracy.
