The company also plans to expand its technology team in Bengaluru, intending to double its size in the next two years.
Cerebras, a competitor to Nvidia, is currently in discussions with the Indian government to enhance the country’s AI compute infrastructure, according to the company’s founder, Andrew Feldman. Feldman, who recently visited Delhi and met with government officials, highlighted India’s lack of data centers and supercomputers as a significant opportunity for Cerebras. He noted that while initial agreements and government declarations are a start, the real challenge lies in building the necessary infrastructure.
Cerebras is known for its unique approach to chipmaking, using entire silicon wafers to create large chips for AI supercomputers, contrary to the traditional method of cutting silicon wafers into smaller chips. The company also plans to expand its technology team in Bengaluru, intending to double its size in the next two years. Currently, with 38 employees in Bengaluru, Cerebras engages in AI developer work and does not segregate tasks based on regional locations.
Earlier in the year, Cerebras announced the construction of the first of nine AI supercomputers in collaboration with Abu Dhabi. This initiative aims to offer alternatives to systems that rely on Nvidia technology. The first supercomputer, Condor Galaxy 1 in Santa Clara, California, is operational and is expected to double in size soon. Additional systems are planned for Austin and Asheville, North Carolina, in the coming year.
Abu Dhabi-based tech conglomerate G42 plans to use Cerebras systems to provide AI computing services to sectors like healthcare and energy. Cerebras has also implemented OpenAI Chief Scientist Andrei Karpathy’s nanoGPT, which is considered the simplest and most compact code base for training and fine-tuning GPT models. Unlike nanoGPT, which trains models in the 100M parameter range, gigaGPT can train models well over 100B parameters.
