Microsoft CEO Satya Nadella has recently lauded China based AI company- Deep Seek for its innovative AI computing architecture. He has set DeepSeek as a new benchmark for the Redmond giant’s artificial intelligence efforts, emphasizing the power of focused innovation. He was talking during a recent employee town hall meeting. The Verge reported that, Nadella specifically appreciated DeepSeek’s ability to enhance AI computing architecture and attain remarkable outcomes with a small team.
“What’s most impressive about DeepSeek is that it’s a great reminder of what 200 people can do when they come together with one thought and one play,” Nadella noted while addressing the Microsoft employees.
What He Meant
Satya Nadella praised DeepSeek’s R1 model for becoming a top-ranking app in the US Apple Store. He highlighted how a small team of 200 people worked with a clear vision to achieve this success. He also appreciated how DeepSeek transitioned from a research project to a popular consumer product, emphasizing its advanced computing architecture.
Jay Parikh, head of Microsoft’s CoreAI engineering group, agreed with Nadella’s views. He pointed out that DeepSeek’s success highlights the importance of teamwork and rapid innovation in the highly competitive AI space. This achievement not only earned praise from a top tech CEO but also set a new goal for Microsoft’s AI strategy. It serves as motivation for the company to continue investing heavily in AI technology.
Satya Nadella found DeepSeek’s system optimization impressive. He was particularly interested in how it works efficiently under Nvidia’s CUDA layer. He saw this as an example of new and advanced technology that could influence future computing developments.
Nvidia’s CUDA
Nvidia’s CUDA technology speeds up computing by using powerful graphics cards (GPUs) to process tasks faster. It breaks complex jobs into smaller pieces, running them simultaneously across many GPU cores. This makes processes like AI and deep learning much more efficient by relying on the GPU’s power, rather than just the regular CPU, allowing computers to handle large workloads quickly.
Chat GPT Versus DeepSeek
Launched on January 20, 2025, it was created by a 200-person team. The team claims that it was created for under $6 million, using 2,000 Nvidia H800 GPUs. This is much cheaper than GPT-4, which cost over $100 million. The company avoided US chip export restrictions by stockpiling Nvidia A100 chips and pairing them with less expensive ones. This approach, along with using less memory, makes DeepSeek more efficient and affordable compared to ChatGPT.
