AI chip company Cerebras is now offering China’s market-shaking DeepSeek model to its US customers, claiming it outperforms traditional GPUs in inference tasks. The company’s large chips are well-suited for speedy AI model running rather than training.
DeepSeek has generated significant interest among enterprise customers due to its ability to match cutting-edge “reasoning” models while costing less to train and use. CEO Andrew Feldman stated that demand was overwhelming, with 10 days of sales surpassing expectations.
A demonstration showcasing DeepSeek’s speed saw it complete a chess program task in just 1.5 seconds, compared to 22 seconds on traditional GPUs. However, concerns over safety and biases have arisen, particularly when using the model to censor sensitive topics or generate bomb-making instructions.
Feldman emphasized that users should exercise caution when utilizing DeepSeek, acknowledging its Chinese worldview and potential limitations. He advised users to consider alternatives offered by US-based companies like Perplexity, which serve LLMs in the US.
Cerebras filed paperwork for an initial public offering in September, highlighting the growing interest in its AI services and specialized chips.
Source: https://fortune.com/2025/01/30/cerebras-china-deepseek-ai-fastest-crushed-demand-business-customers