What you need to know
- Microsoft is reportedly working on its own chips to train and power AI.
- The chips have reportedly been tested in secret since 2019 and could become available to Microsoft and OpenAI by next year.
- Creating its own chips would allow Microsoft to rely less on NVIDIA, which has a strong foothold in the AI server space at the moment.
Microsoft has plans to make its own AI chips, according to a recent report by The Information. In fact, the report states that the tech giant has been working on the chips since 2019 in secret. The efforts, assuming the result is chips Microsoft is happy using, will allow Microsoft to save money and shift away from relying on NVIDIA so heavily.
Select Microsoft and OpenAI employees already have the chips, allowing them to perform tests for large language models such as GPT-4. If Microsoft could use its own chips rather than NVIDIA GPUs to train LLMs and power AI, it could result in significant savings.
Microsoft has invested billions in OpenAI and uses GPT-4 to power Bing Chat. AI is being integrated into several Microsoft services as well. If Microsoft can successfully create chips that can train and power AI, the company could save money directly by using its own hardware and also benefit from improving the artificial intelligence its services rely on.
NVIDIA dominates the AI server space at the moment. It's estimated that ChatGPT may require 30,000 NVIDIA GPUs to meet demand. An NVIDIA A1000 GPU costs between $10,000 and $15,000, so NVIDIA could make as much as $300 million just off the GPUs that power ChatGPT. That was merely an estimate, but it gives us a gauge of the scope we're looking at.
Amazon, Google, and Meta already make their own AI chips. Microsoft following suit would give the company more control over hardware while also potentially saving millions, or even billions, of dollars.
