Microsoft announces powerful new chip for AI inference
TechCrunch • • business
Key Points:
- Microsoft has launched the Maia 200 chip, designed to enhance AI inference by running powerful AI models faster and more efficiently, boasting over 100 billion transistors and delivering over 10 petaflops in 4-bit precision.
- Maia 200 significantly improves upon its predecessor, Maia 100, providing approximately 5 petaflops of 8-bit performance and aiming to reduce AI inference costs and power consumption for businesses.
- The chip positions Microsoft to compete with other tech giants' custom AI chips like Google's TPU and Amazon's Trainium, offering 3x the FP4 performance of Amazon's Trainium3 and superior FP8 performance compared to Google's seventh-generation TPU.
- Maia 200 is already in use within Microsoft's AI initiatives,