Amidst a competitive landscape in the graphics processing unit (GPU) sector, Nvidia is experiencing notable revenue growth while encountering challenges from emerging startups and established tech giants. The company has committed $20 billion to license technology and attract talent from Groq, a contender in the inference market founded by a former TPU engineer.
Google has developed Tensor Processing Units (TPUs) over the past decade, primarily for its own services, but has recently allowed Meta to utilize these resources, intensifying competition with Nvidia. Amazon is also expanding its chip offerings with Trainium and Inferentia, aimed at reducing Nvidia's market dominance. In parallel, Microsoft is progressing in chip development, having announced its AI inference chip, the Maia 200, while Meta plans to release four new silicon generations in the next two years.
The demand for AI inference solutions is spurring investment in startups like Cerebras, which has secured a $10 billion deal with OpenAI for wafer-scale chips. SambaNova raised $350 million after unsuccessful talks with Intel, focusing on AI solutions for businesses. Tenstorrent, valued at $2 billion, is also emerging as an alternative to conventional GPUs, reflecting a broader industry trend towards self-sufficiency in AI hardware.